Systems, devices, and methods for acoustic-signal based respiratory device use assessment

Abstract

Systems, devices, and methods for assessing respiratory device usage and breathing patterns using acoustic signal processing and machine learning may utilize signals, measurements, and / or recordings provided by a sensor system in communication (e.g., physical, sonic, etc.) with a subject and / or a respiratory device used by the subject. Audio signals captured by the sensor system may be analyzed and characterized based on audio signatures to identify event types such as shaking, priming, actuation, inhalation, exhalation, and breath holding. A respiratory device use assessment model and / or breathing assessment model evaluates the audio signals to determine whether the subject is using the respiratory device properly or executing breathing techniques correctly. Feedback is provided to the subject and / or healthcare providers regarding usage technique, breathing patterns, and recommendations for improvement. The system supports various respiratory devices including inhalers, PEP devices, peak flow meters, incentive spirometers, and respiratory muscle trainers.

Description

SYSTEMS, DEVICES, AND METHODS FOR ACOUSTIC-SIGNAL BASED RESPIRATORY DEVICE USE ASSESSMENTRELATED APPLICATIONS

[0001] This application claims priority to U. S. Provisional Patent Application Number 63 / 779,192, entitled “SYSTEMS, DEVICES, AND METHODS FOR EVALUATING AND PREDICTING USER RESPIRATORY HEALTH,” which was filed on 27 March 2025 and U. S. Provisional Patent Application Number 64 / 015,998, entitled “SYSTEMS, DEVICES, AND METHODS FOR ACOUSTIC-SIGNAL BASED RESPIRATORY DEVICE USE ASSESSMENT,” which was filed on 24 March 2026, both of which are incorporated by reference herein in their respective entireties.BACKGROUND

[0002] Effective medication delivery through metered dose inhalers (MDIs) presents several challenges for users with respiratory conditions. Users are often instructed to begin inhaling simultaneously with activating the MDI and / or triggering the MDI to dispense medication into their mouths so that the medication is carried by the inhaled breath through the airway and down into the distal parts of the lungs. When a user fails to do this, the medication sprayed into the mouth may be deposited on a surface of mouth and / or throat rather than reaching the lungs as intended, thereby significantly reducing medication’s effectiveness.

[0003] Another error that frequently occurs with MDI usage is improper alignment of the inhaler with user anatomy. For example, if the inhaler is not pointed directly towards the back of the user’s throat (e.g., toward the roof and / or side of the mouth), some, or all, of the medication may be deposited on the surface of the mouth and not be properly inhaled and delivered to the lungs effectively.

[0004] Other respiratory devices suffer from decreased efficacy when not properly aligned with the user’s face, mouth, nose, and / or throat.SUMMARY

[0005] Systems and devices for assessing respiratory device usage may include a microphone (e.g., electret condenser microphone and / or a MEMS microphone) configured to capture an audio signal when a subject may be using a respiratorydevice, a transceiver (e.g., a Bluetooth Low Energy (BLE) transceiver) configured to communicate the audio signal to a processor that may, or may not be resident in an external device (e.g., a computer, smartphone, cloud-computing device, and / or tablet). The processor may be configured to receive the audio signal, detect one or more acoustic events within the audio signal, identify an event type for each of the one or more acoustic events, and assess the subject's respiratory device usage technique based on the identified event types.

[0006] In some instances, the system may include an analog-to-digital converter configured to digitize the audio signal at a sampling rate of, for example, 44 kHz or 48 kHz with 16-bit resolution. Exemplary respiratory devices include, but are not limited to, a metered dose inhaler (MDI), a dry powder inhaler (DPI), a soft mist inhaler (SMI), a nebulizer, a positive expiratory pressure (PEP) device, an oscillating positive expiratory pressure (oPEP) device, a peak flow meter, an incentive spirometer, and a respiratory muscle trainer.

[0007] In some embodiments, the processor may be configured to preprocess the audio signal to generate a preprocessed audio signal prior to detecting the one or more events and the preprocessed audio signal may be analyzed to detect the one or more events. Preprocessing the audio signal may include bandpass filtering the audio signal to, for example, isolate frequency ranges associated with respiratory device use events and / or remove noise and / or applying adaptive noise reduction to the audio signal using, for example, spectral subtraction.

[0008] In some embodiments, detecting the one or more acoustic events may include applying amplitude thresholding, temporal smoothing, and slope analysis to the audio signal and / or preprocessed audio signal. Exemplary event types include, but are not limited to shaking the respiratory device, removing a cap from a mouthpiece, priming the respiratory device, actuating the respiratory device, subject inhalation, subject exhalation, breath holding, and replacing a cap on the mouthpiece.Additionally, or alternatively, detecting the one or more acoustic events may include segmenting the preprocessed audio signal into a plurality of event windows based on the detected acoustic events and extract features from each event window, wherein the features comprise one or more of amplitude, signal energy, frequency content, and event duration.

[0009] In some embodiments, assessing the subject's respiratory device usage technique comprises determining whether the subject executed events in a proper sequence and for a proper duration of time according to clinically validated, recommended, and / or clinically prescribed procedures. On some occasions, the processor may be further configured to provide feedback to the subject regarding the assessed respiratory device usage technique via a software application running on the external device. Exemplary feedback comprises a confirmation message when the subject's technique may be correct or targeted guidance when the subject's technique may be incorrect.

[0010] In some embodiments, the processor may be configured to input the audio signal and / or preprocessed audio signa into a respiratory device use assessment model trained on labeled audio recordings of respiratory device usage to classify the subject's usage technique as correct or incorrect. In these embodiments, the respiratory device use assessment model may be further configured to generate a confidence score indicating a probability that the subject's usage technique matches a correct or incorrect pattern and / or technique.Additionally, or alternatively, the respiratory device use assessment model may be configured to identify specific types of usage errors comprising one or more of insufficient inhalation force, improper timing of actuation, short or incomplete inhalation, lack of breath holding, lack of shaking or agitation of medication in a canister, and / or skipping a priming step.

[0011] In some embodiments, the system may further comprise an accelerometer configured to sense an orientation and / or movement of the respiratory device. Additionally, or alternatively, the system may further comprise a cloud computing platform configured to receive the audio signal from the external device and perform signal processing and machine learning classification thereon.

[0012] Methods disclosed herein may computer or processor implemented and / or executed and may comprise receiving an audio signal captured by a microphone when a subject using a respiratory device. The microphone may be attached to, resident within, and / or proximate to the respiratory device. In some embodiments, the audio signal may be preprocessed to remove noise therefrom. The preprocessing of the audio signal may comprise, for example, bandpass filtering the audio signal using digital Infinite Impulse Response (IIR) or Finite ImpulseResponse (FIR) filters, applying spectral subtraction to subtract an estimate of a noise spectrum from active segments of the audio signal, and / or segmenting and / or windowing all, or a portion, of the audio signal.

[0013] The audio signal and / or preprocessed audio signal may be analyzed, evaluated, or otherwise processed to detect one or more acoustic events using, for example, known audio signatures for events and / or a result intensity peak analysis and / or frequency analysis of the audio signal and / or preprocessed audio signal. In some embodiments, detecting the one or more acoustic events comprises analyzing the preprocessed audio signal to detect sound amplitude peaks and segmenting the preprocessed audio signal into a plurality of event windows based on the detected amplitude peaks. In these embodiments, detecting the one or more acoustic events may include extracting features from each event window, wherein the features comprise one or more of time-domain features, frequency-domain features, and cepstral features. Extracting frequency-domain features may involve applying a Fast Fourier Transform (FFT) to each event window to obtain a spectral representation thereof.

[0014] Once detected, one or more event types may be identified for each of the one or more acoustic events based on, for example, acoustic signatures associated with the event types. At times, identifying the event type may include comparing acoustic characteristics of each detected acoustic event to a library of audio signatures that correlates sound signatures with event types. Once the events are identified, the subject's respiratory device usage technique may be assessed and / or evaluated based on the identified event types.

[0015] At times, assessing the subject's respiratory device usage technique may include, for example, inputting the audio signal and / or preprocessed audio signal into a respiratory device use assessment model and receiving, from the respiratory device use assessment model, a classification of and / or indication for the subject's usage technique as correct or incorrect. In some embodiments, the method may further comprise receiving an identification of a specific type of usage error when the classification indicates incorrect usage and / or providing feedback (e.g., targeted guidance for correcting the subject's technique when the assessed technique may be incorrect) to the subject regarding the assessed respiratory device usage technique via a user interface.

[0016] In some embodiments, execution of the method may include triggering an alert to a healthcare provider when repeated incorrect usage may be detected over a configurable time window.

[0017] Methods disclosed herein may comprise receiving a labeled set of audio signals of respiratory device usage, wherein labels for the audio signals indicate, for example, identifications of respiratory events, start and / or stop times for respiratory events, a sequence of events, and / or indications whether respiratory device usage technique performed during capture of the audio signal was performed correctly or incorrectly. In some embodiments, data augmentation techniques (pitch shifting, noise injection, and time-stretching) may be applied to one or more of the labeled set of audio signals.

[0018] The labeled set of audio signals may be divided into a training set and a testing set and information regarding respiratory device usage technique may be received. This information may include, for example, indications of proper and improper usage, a correct sequence and / or timing of events for respiratory device usage, and / or an incorrect sequence and / or timing of events for respiratory device usage. In some embodiments, the labels for the audio signals may indicate one or more of event type, event start time, event stop time, event duration, and an indication of whether a subject performed a task associated with an event properly.

[0019] A first version of the respiratory device use assessment model may then be trained using the training set, wherein the respiratory device use assessment model may be configured to evaluate an audio signal to determine if a subject may be properly using a respiratory device and / or using a respiratory device and / or system in manner that is compliant with, for example, a recommendation, prescription, and / or know proper technique for respiratory device and / or system usage. The first version of the respiratory device use assessment model may be tested using the testing set and updated and / or iterated upon until a desired level of accuracy may be achieved, thereby generating a second and / or final version of the respiratory device use assessment model.

[0020] At times, the respiratory device use assessment model comprises and / or utilizes a deep neural network (DNN), a convolutional neural network (CNN), a recurrent neural network (RNN), and / or a long short-term memory network (LSTM).

[0021] In some embodiments, the respiratory device use assessment model may be configured to generate a binary classification output indicating correct or incorrect usage, a scaled output indicating a range of correct or incorrect usage, and / or perform multi-class classification to detect specific types of usage errors.

[0022] Methods, systems, and devices disclosed herein may further be deployed to train and / or generate a respiratory event characterization model configured to identify and characterize events within an audio signal and / or determine an audio signature for event types included in an audio signal based on acoustic characteristics of the event. In some embodiments, the respiratory event characterization model comprises a deep neural network (DNN), a convolutional neural network (CNN), a recurrent neural network (RNN), and / or a long short-term memory network (LSTM).

[0023] The method may include receiving a labeled data set of audio signals of respiratory device use and / or breathing exercises, wherein labels for the audio signals indicate event types and event timing and dividing the labeled data set of audio signals into a training set and a testing set. Exemplary events include one or more of shaking a respiratory device, removing a cap from a mouthpiece, priming the respiratory device, actuating the respiratory device, subject inhalation, subject exhalation, breath holding, and replacing a cap on the mouthpiece. Exemplary labels for the audio signals may indicate one or more of subject characteristics, respiratory device type, and environmental conditions proximate to a subject.

[0024] Then, a first version of the respiratory event characterization model may be trained using the training set. The first version of the respiratory event characterization model may then be tested and / or evaluated using the testing set and iterated upon using a result of the testing until a desired level of accuracy is achieved, thereby generating a second version of the respiratory event characterization model.

[0025] In some embodiments, the method may further comprise extracting features (one or more of event duration, amplitude, power, frequency, and frequency bands) from the audio signals of the training set generating a library of audio signatures that correlates sound signatures for identified events with annotations corresponding to the identified events.

[0026] The respiratory therapy systems disclosed herein may include and / or be in audio communication with a respiratory device (e.g., a metered dose inhaler (MDI), a dry powder inhaler (DPI), a soft mist inhaler (SMI), a nebulizer, a positive expiratory pressure (PEP) device, an oscillating positive expiratory pressure (oPEP) device, a peak flow meter, an incentive spirometer, and a respiratory muscle trainer) comprising a mouthpiece and a sensor system attached to or integrated with the respiratory device. The sensor system may include a microphone configured to capture an audio signal when a subject may be using the respiratory device and / or breathing, an optional an analog-to-digital converter configured to digitize the audio signal, and a transceiver configured to wirelessly communicate the digitized audio signal to a user device (e.g., a computer, smartphone, and / or tablet). On some occasions, the transceiver may be a Bluetooth Low Energy (BLE) transceiver configured to communicate with the user device using GATT protocols.

[0027] The user device may comprise a software application configured to receive the digitized audio signal from the transceiver, store the digitized audio signal, and communicate the digitized audio signal to a processor, which may be resident on the user device and / or external to the user device (e.g., a computer, cloud-computing device, and / or tablet). In some embodiments, the software application may be configured to associate labels and / or metadata (e.g., timestamps, device ID, session markers, and signal quality indicators) with the digitized audio signal. Additionally, or alternatively the software application may be configured to store the digitized audio signal in an uncompressed linear Pulse Code Modulation (PCM).wav format.

[0028] The processor may be configured to optionally preprocess the digitized audio signal, input the preprocessed audio signal into a respiratory device use assessment model, receive an output from the respiratory device use assessment model assessing the subject's respiratory device usage technique and communicate the output to a display device (e.g., a display of the user device) for display to the subject. In some embodiments, the respiratory device use assessment model may be configured to identify specific types of usage errors comprising one or more of insufficient inhalation force, improper timing of actuation, short or incomplete inhalation, and skipping a priming step.

[0029] In some embodiments, the processor may comprise and / or be in communication with an audio signal preprocessing module configured to perform noise suppression, peak detection, segmentation, and / or feature extraction. In some embodiments, the output may include a score indicating how closely the subject's actions align with clinically validated respiratory device usage procedures and / or a confidence score. On some occasions, the processor may be configured to trigger an alert to a healthcare provider when repeated improper usage and / or a change in usage (e.g., an increase of usage) is detected. In some instances, the output may include targeted guidance for correcting the subject's technique when the assessed technique is incorrect and / or can be improved. In some embodiments, the sensor system comprises an accelerometer configured to sense an orientation and / or movement of the respiratory device.

[0030] Additionally, or alternatively, a breathing assessment model may be trained by receiving information regarding breathing techniques such as a correct sequence and timing of events (e.g., inhalation, exhalation, breath holding, etc.) for performance of a breathing technique (e.g., box breathing, diaphragmatic breathing, 4-7-8 breathing, pursed-lip breathing, and Lamaze breathing). A labeled data set of audio signals of breathing exercise execution may be received. Labels for the audio signals may indicate, for example, whether breathing technique was performed correctly or incorrectly. An augmented labeled data set may then be generated using the information regarding breathing techniques. The augmented labeled data set may then be divided into a training set and a testing set and the training set may be used to train, or generate, a first version of the breathing assessment model. The breathing assessment model may be configured to, for example, evaluate an audio signal to assess a subject's breathing patterns and / or breathing technique execution. The first version of the breathing assessment model may be tested and / or evaluated using the testing set and results of the testing and / or evaluation may be used iterate upon the first version of the breathing assessment model until a desired level of accuracy may be achieved, thereby generating a second version of the breathing assessment model.

[0031] In some embodiments, the breathing assessment model may be configured to generate a binary and / or a scaled (e.g., 1-5 or 1-10) classification output indicating correct or incorrect breathing technique execution. Additionally, oralternatively, the breathing assessment model may be configured to perform multiclass classification to detect specific types of breathing patterns and / or technique execution errors.

[0032] In some embodiments, the breathing assessment model may be configured to identify when and for how long a subject inhales, exhales, and / or holds his or her breath. Additionally, or alternatively, the breathing assessment model may comprise and / or include a deep neural network (DNN), a convolutional neural network (CNN), a recurrent neural network (RNN), and / or a long short-term memory network (LSTM).

[0033] The breathing assessment model may be used to assessing a subject's breathing and / or execution of breathing technique by, for example, receiving an audio signal captured by a microphone when a subject may be breathing and / or performing a breathing exercise. The microphone may be a component of a sensor system like the sensor systems disclosed herein. Optionally, the audio signal may be preprocessed to generate a preprocessed audio signal. The audio signal and / or preprocessed audio signal may be input into a respiratory event characterization model and / or a breathing assessment model and an output from the respiratory event characterization model and / or the breathing assessment model may be received. The output may identify, evaluate, and / or characterize one or more breathing events within the audio signal. Then, the subject's breathing patterns and / or breathing technique execution may be assessed and / or evaluated based on, or responsively to, the output. In some embodiments, assessing the subject's breathing patterns and / or breathing technique execution comprises determining whether the subject executed events in a proper sequence and / or for a proper duration of time according to a prescribed breathing technique. In some instances, feedback may be provided to the subject regarding the assessed breathing patterns and / or breathing technique execution via, for example, a user interface. At times, the feedback may comprise targeted guidance for correcting the subject's breathing technique when the assessed technique may be incorrect.

[0034] Breathing events may include, but are not limited to subject inhalation, subject exhalation, and breath holding. These events may be identified by the respiratory event characterization model and / or otherwise determined using, for exemplary characteristics (e.g., frequency ranges, a sequence of sounds within anaudio signal (e.g., exhalation follows inhalation) and / or durations) known for and / or common to subject inhalation, subject exhalation, and breath holding. Additionally, or alternatively, events within an audio signal may be characterized using a procedure for a breathing technique. For example, if a breathing technique begins with an exhalation, then it may be assumed that the first event of an audio signal corresponds to exhalation and the next event either corresponds to breath holding or inhalation depending on, for example, the breathing technique procedure.

[0035] In some embodiments, an output of the respiratory event characterization model and / or the breathing assessment model may be validated to, for example, a confidence level for the output. In these embodiments, an error message may be communicated to a display device when the output may be not valid.BRIEF DESCRIPTION OF DRAWINGS

[0036] The accompanying drawings, which are incorporated in and constitute a part of this disclosure, illustrate exemplary embodiments and, together with the description, serve to explain the disclosed invention.

[0037] FIG. 1 is a block diagram of exemplary components included in a sensor system for use with one or more embodiments of a respiratory therapy device disclosed herein, in accordance with some embodiments of the present disclosure;

[0038] FIG. 2A1 is a schematic diagram of a side view of an exemplary medication inhaler system, in accordance with some embodiments of the present disclosure;

[0039] FIG. 2A2 is a schematic diagram of a front view of an exemplary medication inhaler system, in accordance with some embodiments of the present disclosure;

[0040] FIG. 2B provides a block diagram of a top view of an exemplary positive expiratory pressure (PEP) system, in accordance with some embodiments of the present disclosure;

[0041] FIG. 2C provides a block diagram of a top view of an exemplary oscillating positive expiratory pressure (oPEP) system, in accordance with some embodiments of the present disclosure;

[0042] FIG. 2D provides a block diagram of a top view of an exemplary peak flow meter system, in accordance with some embodiments of the present disclosure;

[0043] FIG. 2E provides a block diagram of a top view of an exemplary incentive spirometer system, in accordance with some embodiments of the present disclosure;

[0044] FIG. 2F provides a block diagram of a top view of an exemplary respiratory muscle trainer system, in accordance with some embodiments of the present disclosure;

[0045] FIG. 2G provides a block diagram of a top view of another exemplary respiratory device system with a face mask, in accordance with some embodiments of the present disclosure;

[0046] FIG. 2H provides a block diagram of a top view of an exemplary respiratory device system with a mouthpiece, in accordance with some embodiments of the present disclosure;

[0047] FIG. 211 provides a block diagram of an exemplary wearable respiratory monitoring system, in accordance with some embodiments of the present disclosure;

[0048] FIG. 2I2 provides a block diagram of user wearing the wearable respiratory monitoring system of FIG. 211, in accordance with some embodiments of the present disclosure;

[0049] FIG. 2J provides a block diagram of another exemplary wearable respiratory monitoring system, in accordance with some embodiments of the present disclosure;

[0050] FIG. 3 provides a block diagram of an exemplary system for performing one or more methods, or portions thereof, disclosed herein, in accordance with some embodiments of the present disclosure;

[0051] FIG. 4A provides a flowchart illustrating a method for identifying and / or characterizing events within an audio signal that captures sound made by a subject when breathing, interacting with, and / or using a respiratory device and / or a respiratory therapy, in accordance with some embodiments of the presentdisclosure;

[0052] FIG. 4B provides a flowchart illustrating an exemplary method for executing step 410 of the method of FIG. 4A, in accordance with some embodiments of the present disclosure;

[0053] FIG. 5A provides a graph of an eleven-second raw audio signal, in accordance with some embodiments of the present disclosure;

[0054] FIG. 5B provides a graph showing a preprocessed version of the raw audio signal of FIG. 5A, in accordance with some embodiments of the present disclosure;

[0055] FIG. 5C provides a graph showing a smoothed version of the audio signal of FIG. 5B, in accordance with some embodiments of the present disclosure;

[0056] FIG. 6A provides a graph of an amplitude of a denoised and smoothed audio signal of FIG. 5C, in accordance with some embodiments of the present disclosure;

[0057] FIG. 6B provides a graph of a STA / LTA ratio of the time series shown in FIG. 6A with a trigger threshold and a detrigger threshold superimposed thereon, in accordance with some embodiments of the present disclosure;

[0058] FIG. 6C provides a graph showing an event windowed version of the amplitude of the denoised and smoothed audio signal of FIG. 6A over time, in accordance with some embodiments of the present disclosure;

[0059] FIG. 7A provides a flowchart illustrating a method for training a respiratory event characterization model, in accordance with some embodiments of the present disclosure;

[0060] FIG. 7B provides a flowchart illustrating a method for training and / or generating a respiratory device use assessment model, in accordance with some embodiments of the present disclosure;

[0061] FIG. 7C provides a flowchart illustrating a method for training and / or generating a breathing assessment model, in accordance with some embodiments of the present disclosure;

[0062] FIG. 8 provides a flowchart illustrating another exemplary method for training a respiratory device use assessment model configured to assess subject’s technique when using a respiratory device and / or system, in accordance with some embodiments of the present disclosure;

[0063] FIG. 9 provides a flowchart illustrating an exemplary method for using a respiratory event characterization model to identify and / or characterize events within an audio signal that includes sound made by a subject when interacting with and / or using a respiratory device and / or a respiratory therapy, in accordance with some embodiments of the present disclosure;

[0064] FIG. 10 provides a flowchart illustrating an exemplary method for using a respiratory device use assessment model to assess a subject’s use of a respiratory device and / or respiratory therapy system, in accordance with some embodiments of the present disclosure;

[0065] FIG. 11 provides a flowchart illustrating an exemplary method for training a breathing assessment model, in accordance with some embodiments of the present disclosure;

[0066] FIG. 12 provides a flowchart illustrating an exemplary method for using a respiratory event characterization model to identify and / or characterize events within an audio signal that includes sound made by a subject when breathing and / or using a respiratory device, a respiratory therapy system, and / or respiratory monitoring system(s), in accordance with some embodiments of the present disclosure;

[0067] FIG. 13 provides a flowchart illustrating an exemplary method for using a breathing assessment model to assess a subject’s breathing patterns and / or breathing exercise execution technique, in accordance with some embodiments of the present disclosure.

[0068] Throughout the drawings, the same reference numerals and characters, unless otherwise stated, are used to denote like features, elements, components, or portions of the illustrated embodiments. Moreover, while the subject invention will now be described in detail with reference to the drawings, the description is done in connection with the illustrative embodiments. It is intended that changes and modifications can be made to the described embodiments without departing from the true scope and spirit of the subject invention as defined by the appended claims. In addition, it should be noted that the dimensions provided by some of the drawings are exemplary only.WRITTEN DESCRIPTION

[0069] Disclosed herein are systems, devices, and methods for assessing respiratory device and / or respiratory therapy system use by a subject using, for example, acoustic signal processing and machine learning. Exemplary respiratory device and / or respiratory therapy system include, but are not limited to, inhalers (e.g., metered dose inhalers (MDIs), dry powder inhaler (DPIs), soft mist inhalers (SMIs)), spirometers, incentive spirometers, oscillating positive expiratory pressure (oPEP) devices, positive expiratory pressure (PEP) devices, peak flow meters, respiratory muscle trainers, nebulizer, small volume nebulizers (SVNs) and the like. The systems, devices, and methods disclosed herein may be used to, for example, monitor respiratory device / respiratory therapy system usage, provide feedback to subjects regarding their respiratory device / respiratory therapy system usage, alert medical professionals to conditions that may warrant medical interventions, and / or encourage proper respiratory device / respiratory therapy system usage by a diverse population of subjects.

[0070] The systems and devices included herein often employ a respiratory device / respiratory therapy system equipped with a microphone (e.g., a high fidelity microphone) that captures audio signals during use so that sounds associated with respiratory device / respiratory therapy system usage (e.g., priming, actuation, exhalation, breath holding (e.g., absence of sound), and / or inhalation) may be detected, converted into a digital signal, and / or recorded. These audio signals may be transmitted to a companion software application running on an external device (e.g., a smart phone or user device), where it may be processed to, for example, assess the user’s respiratory device / system usage technique, breathing patterns, and / or therapy compliance. Additionally, or alternatively, the audio signals may be temporarily stored a memory of the external device in, for example, a raw.wav format for future processing and analysis. Once network connectivity is available for the external device, the recorded data (e.g., raw.wav file) may be securely uploaded to a cloud computing platform, where, for example, signal processing algorithms and / or models may be used to analyze the recorded data. Additionally, or alternatively, recorded data may be analyzed by an onboard processor and / or external computing device (e.g., a computer, cloud computing platform, smartphone, etc.). This analysis may include, for example, inputting the recorded data intopreprocessing pipeline that includes noise filtering, peak detection, windowing, and / or feature extraction so that, for example, meaningful acoustic events and / or signatures present in the recorded data may be identified and converted them into a structured set of features representative of the subject’s technique. These features may then be input into a pre-trained machine learning classification model (e.g., a respiratory device use assessment model) that has been trained on a comprehensive dataset of labeled respiratory device / respiratory therapy system usage patterns. The model evaluates the input and generates a confidence score indicating how closely the subject’s actions align with clinically validated respiratory device / respiratory therapy system usage procedures and / or parameters and feedback is provided to the subject and / or a caregiver for the subject.

[0071] The systems, devices, and methods disclosed herein provide, for example, a seamless integration of hardware embedded in and / or in communication with a respiratory device and / or respiratory therapy system, wireless communication between the hardware and an external device, software running on the external device (e.g., one or more mobile applications), cloud computing, and artificial intelligence. The systems, devices, and methods disclosed herein are configured to, for example, enable automated feedback to subjects regarding the respiratory device / respiratory therapy system usage, support adherence monitoring for subjects, and / or empower healthcare providers with actionable insights on the respiratory device / respiratory therapy system usage of various subjects under their care. In some instances, the systems, devices, and methods disclosed herein may be configured to enhance medication effectiveness, improve subject outcomes, and advance digital health technology.

[0072] A goal of the systems, devices, and methods disclosed herein is to automatically evaluate whether the subject has executed the correct sequence and timing of actions (e.g., shaking the inhaler, priming the dose, coordinating actuation with inhalation, and sustaining adequate inhalation duration) of respiratory device / respiratory therapy system usage so that, for example, subjects may achieve optimal treatment outcomes (e.g., deposition of the medication in the lungs) and, consequently, achieve one or more intended therapeutic effect(s).

[0073] The system architecture disclosed herein comprises several integrated components that span hardware, software, and cloud-based analytics. Theseinclude: (1) a microphone embedded in and / or proximate to a respiratory device / respiratory therapy system (e.g., an embedded miniature electret or MEMS microphone, optimized for capturing fine-grained acoustic variations from within the respiratory device / respiratory therapy system body) configured to record audio signals (e.g., raw audio data files) when a subject is using a respiratory device / respiratory therapy system; (2) a transceiver (e.g., a Bluetooth Low Energy (BLE) transceiver) configured to communicate the recorded audio signals to an external device running (3) a custom software application that may be configured to, for example, receive, buffer, and / or store the audio signals in, for example, a high-resolution linear.wav format, along with metadata such as timestamps and device IDs and communicate same to a cloud computing platform; and (4) a secure cloud computing platform that receives and archives the data from the software application and is configured to perform downstream analytics thereon. In some embodiments, the cloud computing platform 310 may include an audio signal preprocessing module configured to perform, for example, noise suppression, peak and event detection, segmentation, and feature extraction using both time- and frequency-domain techniques; and a machine learning classification subsystem (e.g., a deep neural network or ensemble model) trained on a labeled dataset of respiratory device / respiratory therapy system usage sessions to classify each instance as, for example, correct or incorrect and optionally provide a probabilistic confidence score. These components may work together to form a robust pipeline for the continuous assessment of subject technique and support closed-loop feedback to both subjects and clinicians. In some embodiments, the system architecture, model, and / or a portion of a model disclosed herein may include, be resident on, and / or operate within user device 360 and / or software application 365.

[0074] Communication between the components, systems, and software disclosed herein may be compliant with one or more security and / or privacy protocols to, for example, anonymize or de-identify data and / or signals as well as analysis thereof and determinations made therefrom. To that end, subject-identifying information and other information (e.g., health-related information) may be replaced with other identifiers (e.g., a binary string and / or an alpha-numeric code) when, for example, data is in transit between two or more components. In some embodiments, the data disclosed herein may be encrypted via, for example, one or more forms ofencryption for secure data transmission such as single key, or symmetric, encryption via, for example, application of an encryption algorithm (e.g., advance encryption standard (AES), triple data encryption standard (3DES), Twofish, etc.) to the data and / or asymmetric encryption via application of an asymmetric encryption (e.g., Rivest-Shamir-Adleman (RSA) and / or elliptic curve cryptography (ECC) to the data. Additionally, or alternatively, one or more network encryption protocols (e.g., SSL / TLS and / or MACSec) may be used when communicating data as disclosed herein.

[0075] Turning now to the drawings, FIG. 1 is a block diagram of exemplary components included in a sensor system 100 for use with one or more embodiments of a respiratory therapy device disclosed herein. Sensor system 100 may include a housing 105, an imaging / light sensing device 110, one or more port(s) 115, a microphone 120 (which may be embodied as, for example, an acoustic and / or vibration sensor), a power source 130, user interface 140, a transceiver 150, an accelerometer 160, a sound- and / or vibration-producing device 170, a processing device 180, and / or a memory 190. In some cases, sensor system 100 may not include all the components shown in FIG. 1. For example, in some embodiments, sensor system 100 may not include imaging / light sensing device 110, user interface 140, accelerometer 160, sound-and / or vibration-producing device 170, processing device 180, and / or memory 190.

[0076] Housing 105 may be configured to be moisture and / or impact resistant and may be permanently and / or removably attached to a respiratory therapy device. When permanently attached, housing 105 may be integrated into the hardware of a respiratory therapy device and / or affixed via, for example, chemical, heat, and / or vibrational bonding. When removably attached, housing 105 may be attached to a respiratory therapy device via, for example, a magnet, VELCRO®, glue, and / or a mechanical attachment device such as a strap, clasp, and / or snap. Housing 105 may include one or more communication and / or power port(s) 105 that are configured to enable the charging of power source 130 (when embodied as a rechargeable battery) and / or communication with an external device such as a mobile phone, computer, and / or wearable device (e.g., a smartwatch).

[0077] Imaging / light sensing device 110 may be configured to image a user’s face and / or detect light one and / or around the user’s face and / or facial features,which may then be stored on memory 190 and / or communicated to transceiver 150 for wired and / or wireless communication to an external processing device such as a smart phone, tablet computer, desktop computer, and / or cloud-based computing system for further processing and / or analysis. In some cases, transceiver 150 may be configured for low-power, short-range, secure communication to a user’s mobile phone. For example, transceiver 150 may be configured as a Bluetooth® and / or Bluetooth® Low Energy (BLE) device. Power source 130 may be any power source and / or battery configured to provide power to the components of sensor system 100.

[0078] Microphone 120 may be configured to measure or detect vibrations and / or sound made by the subject when, for example, interacting with and / or proximate to a respiratory therapy device and / or a component thereof. Exemplary microphone 120 include, but are not limited to, microphones and / or a device configured to measure a change in the amplitude, frequency, and / or intensity of vibrations or movement of a respiratory therapy device and / or a component thereof. In some embodiments, acoustic and / or vibration measurements sensed by microphone 120 may be stored in memory 190 and / or communicated to transceiver 150 for communication to an external processing device. In some embodiments, microphone 120 may be a low-power miniature electret condenser and / or a low-power omnidirectional MEMS (Micro-Electro-Mechanical Systems) microphone, optimized for capturing fine-grained acoustic variations proximate to and / or from within the inhaler body. Additionally, or alternatively, microphone 120 may be configured to capture a broad dynamic range of acoustic signals and have a broad frequency response (e.g., a frequency response spanning approximately 20 Hz to 20 kHz) so that nuanced acoustic signatures generated during all phases of inhaler use may be accurately captured. An analog signal from the microphone may be digitized using, for example, an onboard low-power analog-to-digital converter (ADC) 125 that may be configured to sample the analog signal at, for example, a 16-bit resolution and 44.1 or 48 kHz to retain fine temporal and spectral detail. The digitized signal (sometimes referred to herein as the “raw acoustic signal” or “raw acoustic data”) may then be packetized and wirelessly transmitted in, for example, real-time to, for example, a user device (e.g., user device 360 and / or a software application running thereon (e.g., software application 365)) via transceiver 150.

[0079] User interface 140 may be any user interface configured to, for example, receive input from and / or provide output to a user. User interface 140 may be embodied as, for example, a button, a keypad, a speaker, a microphone, a touch screen, an indicator light, and a dial. Exemplary subject and / or user input received via user interface 140 may include, for example, an on / off instruction, a selection of a program or routine the subject is (or will be) using, and an instruction to communicate information stored in memory 190 to an external device. Exemplary output provided to a subject and / or user via user interface 140 includes, but is not limited to, a message and / or graphic displayed on a display screen, a tone, a progress bar, and the lighting of an indicator light. In some embodiments, user interface 140 may be configured to provide feedback to a subject regarding use of a respiratory therapy device and / or system 100 and / or images captured and / or determined by imaging / light sensing device 110. Additionally, or alternatively, user interface 140 may be configured to provide feedback to a subject and / or user regarding one or more operations of, for example, microphone 120, transceiver 150, and / or power source 130. Additionally, or alternatively, user interface 140 may be configured to provide feedback regarding when and / or for how long microphone 120 is sensing sound made by a subject while, for example, breathing and / or interacting with a respiratory therapy device or a component thereof (e.g., one-way valve 830 and / or sound- and / or vibration-producing device 170). Additionally, or alternatively, user interface 140 may be configured to provide feedback to a subject indicating that he or she has inhaled and / or exhaled for a predetermined length of time and / or has shaken the assembly for a predetermined length of time. User interface 140 may be embodied as, for example, one or more lights, speakers, and / or display windows and / or touch screens configured to display, for example, text and / or icons to the subject and / or receive input from a subject.

[0080] Sound-and / or vibration-producing device 170 may be any device configured to make a sound and / or vibration in response to, for example, movement of a respiratory device system and / or a flow of gas and / or medication through a component (e.g., mouthpiece and / or valve) of a respiratory therapy device as may occur when, for example, the subject inhales medication and / or exhales. Exemplary movement includes, but is not limited to, shaking a respiratory therapy device, placing a mouthpiece of a respiratory therapy device into a subject’s mouth, and / or aflow of gas and / or medication through a component (e.g., mouthpiece and / or valve) of a respiratory therapy device. Exemplary sound-and / or vibration-producing devices 170 may be mechanical and / or electronic and, in some instances, may include, but are not limited to, reeds, gaskets, ball-bearings, and electronic vibration devices (e.g., buzzers that may be activated responsively to, for example, a subject command and / or movement of the respiratory therapy device). In some cases, air flow across and / or through sound-and / or vibration-producing device 170 may cause vibration by triggering a mechanism or causing turbulent or otherwise disturbed airflow.

[0081] Memory 190 may be any memory configured to store data and / or sets of instructions to be executed by processing device 180. Data stored on memory 190 may include, but is not limited to, measurements detected by imaging / light sensing device 110, microphone 120, accelerometer 160, sound- and / or vibration-producing device 170, inputs received from user interface 140 and / or communications received from and / or transmitted by transceiver 150. The sets of instructions stored in memory 190 may be configured to instruct processing device 180 to execute one or more methods disclosed herein and / or a step thereof. Additionally, or alternatively, the sets of instructions may be configured to instruct the processing device 180 to perform a particular function (e.g., on / off, record measurement, etc.) and / or set a time period for performance of the function. Processing device 180 may be any device configured to execute one or more instructions stored in memory 190 and / or provided via port 115 and / or user interface 140. Exemplary processing devices include, but are not limited to, field programmable gate arrays (FPGAs), applicationspecific integrated circuits (ASICs), and central processing units (CPUs).

[0082] Accelerometer 160 may be configured to sense, for example, an orientation in the X-, Y-, and / or Z-planes, proper acceleration, coordinate acceleration, and / or motion of a respiratory therapy device like the respiratory device systems disclosed herein and / or a component thereof. In some embodiments, accelerometer 160 may be / include a six axis accelerometer and / or gyroscope configured to, for example, sense motion and / or rotational velocity, acceleration, and / or changes in displacement.

[0083] FIG. 2A1 is a schematic diagram of a side view and FIG. 2A2 is a schematic diagram of a front view of an exemplary inhaler 201 that includes anoptional canister housing 210, a canister 215, a medication conduit 235, and a mouthpiece 220 with an opening 230. Optionally, inhaler 201 may include sensor system 100. Canister 215 may be a canister, or container, of medication configured for inhalation into a subject’s lungs to, for example, relieve respiratory distress and / or improve breathing. Optional canister housing 210 may be configured to house canister 215 and / or couple canister 215 to mouthpiece 220. Mouthpiece 220 may be in communication with canister 215 via medication conduit 235 and configured to fit into a subject’s mouth so that, upon activation, medication from canister 215 may be communicated to medication conduit 235 and directed into the subject’s mouth via opening 230. Prior to and / or when the medication is released from canister 215 into the opening, the subject may inhale the medication into his or her lungs as directed and / or prescribed by, for example, a physician or health care provider. In some embodiments, inhaler 201 may be an MDI or DPI, and, in these embodiments, a method for using inhaler 201 may be similar to the use of these devices.

[0084] FIG. 2B provides a block diagram of a top view of an exemplary positive expiratory pressure (PEP) device 202 that includes a body 212, a mouthpiece 222, and sensor system 100 affixed to and / or positioned proximate to mouthpiece 222 so that sounds made by a subject when exhaling into, or otherwise using, PEP device 202 may be detected, recorded, communicated, and / or analyzed as, for example, disclosed herein.

[0085] FIG. 2C provides a block diagram of a top view of an exemplary oscillating positive expiratory pressure (oPEP) device 203 that includes a body 213, a mouthpiece 223, and sensor system 100 affixed to and / or positioned proximate to mouthpiece 223 so that sounds made by a subject when exhaling into, or otherwise using, oPEP device 203 may be detected, recorded, communicated, and / or analyzed as, for example, disclosed herein.

[0086] FIG. 2D provides a block diagram of a top view of an exemplary peak flow meter device 204 that includes a body 214, a mouthpiece 224, and sensor system 100 affixed to and / or positioned proximate to mouthpiece 224 so that sounds made by a subject when exhaling into, or otherwise using, peak flow meter 204 may be detected, recorded, communicated, and / or analyzed as, for example, disclosed herein.

[0087] FIG. 2E provides a block diagram of a top view of an exemplary incentive spirometer 205 that includes a body 25, a mouthpiece 225, and sensor system 100 affixed to and / or positioned proximate to mouthpiece 225 so that sounds made by a subject when exhaling into, or otherwise using, incentive spirometer 205 may be detected, recorded, communicated, and / or analyzed as, for example, disclosed herein.

[0088] FIG. 2F provides a block diagram of a top view of an exemplary respiratory muscle trainer 206 that includes a body 216, a mouthpiece 226, and sensor system 100 affixed to and / or positioned proximate to mouthpiece 226 so that sounds made by a subject when exhaling into, inhaling from, or otherwise using, respiratory muscle trainer 206 may be detected, recorded, communicated, and / or analyzed as, for example, disclosed herein.

[0089] FIG. 2G provides a block diagram of a top view of an exemplary respiratory device 207 with a face mask 227, a body 217, and a tube 237 connecting body 217 to face mask 227. Respiratory device 207 also includes sensor system 100 affixed to and / or positioned proximate to facemask 227 so that sounds made by a subject when exhaling into, inhaling from, or otherwise using, respiratory device 207 may be detected, recorded, communicated, and / or analyzed as, for example, disclosed herein. Respiratory device 207 may be embodied as, for example, a positive expiratory pressure (PEP) device, a continuous positive airway pressure (CPAP) device, a bilevel positive airway pressure (BPAP) device, and / or a nebulizer.

[0090] FIG. 2H provides a block diagram of a top view of an exemplary respiratory device 208 with a mouthpiece 228, a body 218, and a tube 238 connecting body 218 to mouthpiece 228. Respiratory device 208 also includes sensor system 100 affixed to and / or positioned proximate to mouthpiece 228 so that sounds made by a subject when exhaling into, inhaling from, or otherwise using, respiratory device 208 may be detected, recorded, communicated, and / or analyzed as, for example, disclosed herein. Respiratory device 208 may be embodied as, for example, a nebulizer.

[0091] In some embodiments, sensor system 100 and / or one or more components thereof may be integrated into a housing and / or mouthpiece of inhaler 201, PEP device 202, oPEP device 203, peak flow meter device 204, incentive spirometer 205, respiratory muscle trainer 206, respiratory device 207, and / orrespiratory device 208 and / or may be permanently or removably affixed thereto by way of, for example, a mechanical attachment mechanism (e.g., a clip, a snap, or tongue and groove), a chemical attachment mechanism (e.g., glue or epoxy), and / or thermal and / or vibrational bonding. In many cases, sensor system 100 may be arranged on / within inhaler 201, PEP device 202, oPEP device 203, peak flow meter device 204, incentive spirometer 205, respiratory muscle trainer 206, respiratory device 207, and / or respiratory device 208 so as to capture sound made by the subject when using (e.g., breathing into and / or out of) the respective device.

[0092] FIG. 211 provides a block diagram of an exemplary wearable respiratory monitoring system 209 that includes a housing 254 and system 100. Housing 254 may be configured to enable wearing of system 209 so that sounds made by the user while breathing may be detected by microphone 120 and saved in memory 190 and / or communicated to an external device (e.g., a smart phone and / or computer) via transceiver 150. For example, housing 254 may be configured as a strip with attachment mechanism that may be configured to attach system 209 to a wearer’s face (e.g., over the user’s nose proximate to the user’s nostrils and / or nose bridge and / or under the user’s nose). Exemplary attachment mechanisms include, but are not limited to, adhesives, glue, magnets, and straps. A shape and / or size of housing 254 may be configured to fit different parts of a user’s face or body (e.g., a relatively long and thin strip (e.g., 4-7 cm long and 0.5-3 cm wide) for housing 254 configured to fit over a wearer’s nose and a relatively thin oval shape when positioned under the nose). Additionally, or alternatively, housing may be sized (e.g., small, medium, and large) to appropriately fit users of different sizes.

[0093] An example of how system 209 may be worn by a user 21 is provided by FIG. 2I2, which is a diagram illustrating user 21 wearing a first respiratory monitoring system 209a over her nose proximate to her nostrils and a second respiratory monitoring system 209b between her nose and upper lip. Although user 21 is shown wearing first and second systems 209a and 209b, this need not always be the case. For example, in some embodiments, user may wear only first system 209a or second system 209b. In some embodiments, housing 254 may be configured for attachment to a device (e.g., a facemask, breathing cannula, and / or respirator)

[0094] FIG. 2J provides a block diagram of another exemplary wearable respiratory monitoring system 211 configured as a housing 256 with a wearing mechanism 234 configured to allow the user to, for example, wear housing 256 around his or her neck (as, for example, a necklace) and / or on her wrist (as, for example, a bracelet). Wearable respiratory monitoring system(s) 209 and / or 211 may be worn when, for example, a user is sleeping, exercising, practicing breathing exercises, and / or performing activities of daily life.

[0095] FIG. 3 provides a block diagram of an exemplary system 300 for performing one or more methods, or portions thereof, disclosed herein. System 300 includes a cloud computing platform 310, a communication network 320, a computer or terminal 330, a display device 340, a database 350, a user device 360 with a software application 365 installed thereon, respiratory therapy system 201, 202, 203, 204, 205, 206, and / or 207, a subject using or who may use the respiratory device 370, and one or more optional third party users A-N 375A-375N.

[0096] The components of system 300 may be coupled together via wired and / or wireless communication links. In some instances, wireless communication of one or more components of system 300 may be enabled using short-range wireless communication protocols designed to communicate over relatively short distances (e.g., BLUETOOTH®, near field communication (NFC), radio-frequency identification (RFID), and Wi-Fi) with, for example, a computer or personal electronic device (e.g., tablet computer or smart phone) as described below. In some instances, communication network 320 is the Internet, a cellular network, and / or a near-field communication network like Bluetooth®. In many embodiments, respiratory therapy system 201, 202, 203, 204, 205, 206, and / or 207 and / or a transceiver thereof may be in communication with user device 360 and / or software application 365 via direct wired and / or a wireless communication protocol. Sound recordings from microphone 120 may be received by one or more components of system 300 via, for example, software application 365, user device 360, communication network 320, database 350, and or computer 330.

[0097] Cloud computing platform 310 may be any cloud computing platform 310 configured to run a machine learning and / or artificial intelligence software program and / or support a machine learning and / or artificial intelligence architecture configured to build, tune, update, and / or validate machine learning models such asTensorFlow. Exemplary cloud computing platforms 310 include, but are not limited to, Amazon Web Service (AWS), Rackspace, and Microsoft Azure. Exemplary machine learning and / or artificial intelligence architectures include neural networks, deep neural networks, artificial neural networks, Bayesian networks, and / or software or hardware that utilizes artificial intelligence. The machine learning models disclosed herein may incorporate a variety of models such as decision trees, linear regression models, logistic regression models, neural networks, classifiers, support vector machines (SVMs), inductive logic programming, ensembles of models (e.g., using techniques such as bagging, boosting, random forests, etc.), genetic algorithms, Bayesian networks. Examples of neural networks can include convolutional neural networks (CNNs) such as U-Net architectures and Residual Networks (Res-Net). The machine learning models may be configured to perform a variety of tasks including, for example, a regression, a classification, a clustering, or a segmentation. The machine learning models can be trained using a variety of approaches, such as deep learning, association rules, inductive logic, clustering, maximum entropy classification, learning classification, etc. In some cases, the machine learning models may use supervised learning. In other cases, the machine learning models use unsupervised learning.

[0098] Computer 330 may be configured to act as a communication terminal to cloud computing platform 310 via, for example, communication network 320 and may facilitate provision of the results machine learning and / or artificial intelligence calculations (e.g., training and / or testing of a model and / or tuning of a model) performed on cloud computing platform 310 and / or may enable communication between a user or programmer and cloud computing platform 310 by providing access to cloud computing platform 310. Exemplary computers 330 include desktop and laptop computers, servers, tablet computers, personal electronic devices, mobile devices (e.g., smart phones), and the like. Exemplary display devices 340 are computer monitors, tablet computer devices, and displays provided by one or more of the components of system 300. In some instances, display device 340 may be resident in computer 330. Computer 330 may be communicatively coupled to database 350, which may be configured to store information generated by one or more systems, devices, and / or processes disclosed herein such as sound recordings, identifying information for respiratory devices, being used, identifyinginformation for subject(s) using a respiratory device, demographic information for a subject of a respiratory device, medical information for a subject of a respiratory device. In some embodiments, database 350 may be an electronic medical record database. Additionally, or alternatively, database 350 may be configured to store inputs, used for machine learning and / or sets of instructions for computer 330 and / or cloud computing platform 310.

[0099] User device 360 may be any device operated by user 370 and / or a third party user 375A-375N to, for example, communicate with respiratory therapy system 201, 202, 203, 204, 205, 206, and / or 207 and / or a transceiver thereof.Exemplary user devices 360 include, but are not limited to smartphones, smart watches and tablet computers that have an operating system (e.g., iOS® and Android®) and software application 365 installed thereon.[000100] Software application 365 may be configured to, for example, establish and / or maintain communication with respiratory therapy system 201, 202, 203, 204, 205, 206, and / or 207 and / or a transceiver thereof via, for example, a persistent and / or intermittent BLE connection using GATT (Generic Attribute Profile) protocols. Software application 365 may be designed to, for example, manage connection states, handle packet reassembly, and buffer incoming audio data in real time, even under fluctuating network or power conditions. Software application 365 may further be configured to locally store received audio data on a memory of user device 360 in, for example, a format that enables lossless storage and facilitates detailed downstream analysis, including high-resolution signal processing, machine learning inference, and cloud-based archival (e.g., an uncompressed linear Pulse Code Modulation (PCM).wav format, RAW format and / or a Moving Picture Experts Group (MPEG)-4 Part 14 (MP4) format) to preserve the complete spectral and temporal integrity of the received audio data and / or signal carrying same. In some cases, software application 365 may be configured to associate metadata such as timestamps, device ID, session markers, and signal quality indicators with received audio data and / or signals, which may be logged to, for example, support traceability and diagnostics in later processing stages.[000101] Software application 365 may be configured to initiate a secure upload protocol to transfer the received audio data and / or signals and / or preprocessed versions of the received audio data and / or signals to a cloud-based analyticsplatform running on cloud computing platform 310 for processing when user device 360 establishes an active connection (e.g., Wi-Fi and / or a cellular network) with communication network 320. At times, this transfer may be performed over encrypted HTTPS channels using, for example, industry-standard TLS protocols to ensure data confidentiality, integrity, and compliance with healthcare privacy regulations such as HIPAA. In some embodiments, software application 365 may be configured to, for example, perform integrity checks and / or attach metadata to the received audio data and / or signals and / or preprocessed versions of the received audio data and / or signals to, for example, to enrich the data package. Exemplary metadata includes, but is not limited to, session identifiers, subject authentication tokens, signal timestamps, and contextual usage markers (e.g., device battery level, connection status, subject location if permitted).[000102] FIG. 4A provides a flowchart illustrating a method 400 for identifying and / or characterizing events within an audio signal that captures sound made by a subject while breathing and / or interacting with (e.g., using) a respiratory device and / or a respiratory therapy system like the respiratory devices and / or a respiratory therapy systems disclosed herein. Method 400 may be executed via, for example, any of the systems (e.g., system 300) and / or system components (e.g., cloud computing platform 310 and / or respiratory therapy system 201, 202, 203, 204, 205, 206, and / or 207) described herein.[000103] In step 405, a set of audio signals and labels and / or annotations describing the audio signals may be received by, for example, a user device (e.g., user device 360), a software application (e.g., software application 365), and / or a computer (e.g., computer 330 and / or cloud computing platform 310). One or more audio signals of the set may correspond to analog audio data detected by a microphone (e.g., microphone 120) and / or a corresponding digital signal (e.g., a digitized version of the analog signal prepared by, for example, ADC 125) when a user is interacting (e.g., passively and / or actively) with a respiratory therapy system (e.g., respiratory therapy system 201, 202, 203, 204, 205, 206, 207, 208) and / or breathing (e.g., breathing naturally or performing one or more breathing exercises (e.g., box breathing, skip breathing, diaphragmatic breathing, 4-7-8 breathing, pursed-lip breathing, Lamaze breathing, etc.) and / or interacting with a respiratory monitoring system like respiratory monitoring system 209 and / or 211.[000104] In some instances, the audio signal received in step 405 may be a clean (i.e., does not include any ambient or extraneous noise) audio signal that includes only sounds generated by the subject when using the respiratory device, respiratory therapy system, and / or respiratory monitoring system. In other embodiments, the audio signals received in step 405 may be a composite signal audio signal that includes sounds generated by the subject when using the respiratory device, respiratory therapy system, and / or respiratory monitoring system as well as ambient or extraneous noise from the subject’s environment that occurs during use such as noise and / or sounds generated by devices (e.g., household appliances, speech, traffic, HVAC systems, lawn equipment), objects, and / or living things (other people, pets, etc.) proximate to the subject.[000105] Labels, annotations and / or descriptions for the audio signals of the set may include notes regarding one or more events that occur during a length of the audio signal taken by, for example, a clinician or individual trained to recognize events within the audio signals. These notes may be, for example, event type, event start and / or stop time, event duration, an indication of the subject performed a task associated with the event properly and / or in line with clinical guidelines for respiratory device and / or respiratory therapy system use, information about the respiratory device and / or respiratory therapy system used to generate the audio signal, and / or information about the subject of the respiratory therapy system generating the audio signal (e.g., age, gender, diagnosis, respiratory health indicators, medications the subject is taking, etc.). Exemplary events that may occur during an audio signal include, but are not limited to, inhaling, exhaling, holding the breath, removing a cap from a respiratory therapy system, shaking a respiratory therapy system, the subject placing a mouthpiece of a respiratory therapy system into his / her / their mouth, exhalation, inhalation, activation of a metered dose inhaler, replacement of the cap, and priming a respiratory therapy system. In some embodiments, the annotations may include notes, or metadata, relating to, for example, date, time, geolocation, respiratory device characteristics, subject ID, and / or weather conditions, humidity levels, and / or air quality metrics for an environment proximate to the subject when the audio signal is captured.[000106] In some embodiments, the labeled audio signals received in step 405 may correspond to audio information captured when a subject is executing proper(e.g., clinically validated and / or prescribed) breathing technique and / or respiratory device / respiratory therapy system usage technique and / or variations of improper breathing technique and / or respiratory device / respiratory therapy system usage technique. Breathing technique may be improper when, for example, the user does not inhale or exhale according to timing established by the breathing technique. For example, if a breathing technique requires a subject to take two seconds to inhale, hold the breath for two seconds, and then exhale for two seconds and the subject inhales for two seconds and then immediately exhales (i.e., does not hold the breath for two seconds), then their execution of the technique would be improper or incorrect. Respiratory device and / or respiratory therapy system usage technique may be improper when, for example, steps of a usage technique are performed in the wrong order, are improperly performed (e.g., duration of time it takes to execute a step of respiratory device / respiratory therapy system usage technique is too long or short), and / or are not performed at all. In some embodiments, the labels for audio signals that capture improper respiratory device / respiratory therapy system usage technique may indicate how the technique is improper and / or how a subject may fix, or correct, the respiratory device / respiratory therapy system usage technique.Additionally, or alternatively, the labels for one or more audio signals of the data set may indicate one or more parameters for and / or characteristics of the respiratory device and / or respiratory therapy system such as respiratory device / respiratory therapy system type, manufacturer, usage history (e.g., age of the respiratory device / respiratory therapy system, how often the respiratory device / respiratory therapy system is used, when the respiratory device / respiratory therapy system was last used, etc.), and an amount of medication present within a portion of the respiratory device / respiratory therapy system.[000107] In some embodiments, the labels for one or more audio signals of the data set may indicate one or more parameters for and / or characteristics of the usage of the respiratory device and / or respiratory therapy system. For example, a label for an audio file may provide accelerometry information (via e.g., an accelerometer like accelerometer 160) for a system like system 100 and / or a respiratory monitoring and / or therapy system incorporating system 100 that provides information about an orientation of the respiratory therapy system and / or a subject and / or a position of the accelerometer. Additionally, or alternatively, the labels for one or more audio signalsof the data set may indicate and / or correspond to one or more features of noise and / or background sounds included within the audio signal that do not correspond to use of the respiratory device / respiratory therapy system. For example, a label for an audio signal may indicate that a police siren is heard during seconds 5-8 of a 10 second audio signal / recording. In another example, a label for an audio signal may indicate that ambient noise in the form of a ventilation system is present for the entirety of the audio signal.[000108] In some cases, the audio signals and / or recordings may have been collected from, for example, controlled clinical trials and / or real-world usage scenarios and the audio signals may have been annotated by human experts or medical professionals (who may observe the subjects while they are generating the audio signals) to indicate whether the breathing technique and / or respiratory device usage technique was performed correctly or incorrectly, based on, for example, objective criteria and / or physician-guided protocol compliance. The audio signals and / or recordings may include events that record a wide variety of subject behaviors, respiratory device types, ambient noise environments, and error modes, ensuring that the model analyze and evaluate audio signals and / or recordings made under diverse conditions with diverse subject populations. In some embodiments, one or more data augmentation techniques such as pitch shifting, noise injection, bandpass filtering, and / or time-stretching may be applied to the audio signals and / or recordings to, for example, improve model robustness and / or accuracy.[000109] In step 410, one or more acoustic events within the audio signals may be detected and, in step 415 the event may be identified and / or characterized.Exemplary event types include, but are not limited to, shaking a respiratory therapy system, removing a cap from the mouthpiece of a respiratory therapy system, priming a respiratory therapy system, activating a respiratory therapy system, subject inhalation, subject exhalation, subject coughing, breath holding, and replacing a cap on the mouthpiece. In some cases, events may be identified and / or characterized using, for example, features of the acoustic event and / or labels and / or annotations associated with the respective audio signal and / or portion of the audio signal including the detected acoustic event. For example, if a 10s audio signal has annotations indicating that between 1-2.5 seconds the subject was shaking an MDI, between 2.5-3 seconds the subject removed the cap from the MDI, between 3-6seconds the subject exhaled, and between 6-10 seconds the subject put the MDI mouthpiece in his mouth and inhaled medication then, acoustic events may be detected and / or identified using these annotations. For example, different event types (e.g., inhalation, exhalation, canister shaking, dose priming (if applicable), actuation of a medication spray, the subject’s exhalation, the subject's inhalation, mouthpiece cap removal, and / or movement of components (e.g., valves, balls, reeds, etc.) may produce a unique acoustic signature, that may be marked by and / or differentiated from one another using changes (e.g., sudden or gradual) in signal amplitude, intensity, and / or frequency content and / or the absence of sound, which may be a signature for breath holding. For example, an acoustic signature for an actuation phase of an MDI may be characterized by a high-amplitude, short-duration burst with broadband frequency content. In another example, an acoustic signature for an onset of a subject’s inhalation when using a respiratory device embodied as an inhaler may be characterized by as a sustained rise in lower-frequency components of an audio signal due to turbulent airflow entering a mouthpiece of the inhaler. Further details regarding how steps 410 and 415 may be performed are shown in FIG. 4B and discussed below.[000110] In step 420, an audio signature, or a range of audio signatures, for an event type may be determined. In some embodiments, execution of step 420 may include analysis of a plurality (100-10,000,000) of identified events of the same type to, for example, identify and / or characterize one or more features (e.g., event duration, amplitude, power, frequency, frequency bands, and / or changes thereto) common a particular event type that are present within an audio signal corresponding to the event type.[000111] In some instances, the audio signature(s) may be associated with information (received in, for example, step 405) in addition to, or rather than, event type. For example, audio signature(s) may be associated with subject characteristics (e.g., age, height, weight, diagnosis, etc.), respiratory device / respiratory therapy system type, breathing technique the subject is trying to execute, and / or environmental conditions proximate to a subject. For example, an audio signature for a child’s (e.g., 2.5 feet tall) inhalation from an MDI may be different from an audio signature for an adult male (e.g., 6 feet tall) performing the same task and execution of step 420 may include labeling and / or associating an audio signature for an eventtype (in this case, inhalation from an MDI) with a subject type or characteristic (in this case, height and / or age).[000112] Additionally, or alternatively, execution of step 420 may include associating an audio signature for an event type with a diagnosis and / or general medical condition of a subject. For example, an audio signature for inhalation from an MDI by a subject may be associated with a diagnosis of the subject (e.g., asthma, chronic obstructive pulmonary disease (COPD), pneumonia, or lung cancer) so that differences in the audio signature for an event type across various subjects with various respiratory health indications may be classified and / or identified.[000113] Additionally, or alternatively, execution of step 420 may include associating an audio signature for a respiratory event with respiratory device and / or respiratory therapy system information including, but not limited to, respiratory device and / or respiratory therapy system type (e.g., MDI, nebulizer, peak flow meter, etc.), manufacturer, and / or form factor.[000114] In some embodiments, the annotations of step 405 may provide information about ambient sounds and / or noise that may be included within the audio signal so that, for example, the ambient sound and / or noise may also be identified in step 415 and / or an audio signature for the ambient sound and / or noise may also be determined in step 420.[000115] In step 425, a library of audio signatures that correlates a sound signature for one or more identified event(s) within an audio signal with annotations corresponding to the identified event(s) of the respective audio signal may be generated and saved in a database (e.g., database 350) (step 430).[000116] FIG. 4B provides a flowchart illustrating an exemplary method of executing step 410. In step 435, an audio signal (e.g., an audio signal received in step 405) may be filtered or otherwise processed to remove noise and / or extraneous sound. For example, in some embodiments, execution of step 435 may include bandpass filtering the audio signal to isolate one or more frequency ranges, or frequency bands, of interest (e.g., frequency bands associated with respiratory device use events). This bandpass filtering may be implemented using, for example, digital Infinite Impulse Response (IIR) or Finite Impulse Response (FIR) filters.[000117] Additionally, or alternatively, execution of step 435 may include application of one or more adaptive noise reduction techniques to the audio signaland / or bandpass filtered audio signal. Exemplary noise reduction techniques include, but are not limited to, spectral subtraction, where an estimate of the noise spectrum calculated during silent or low-energy portions of the signal is subtracted from active segments of the audio signal. Additionally, or alternatively, execution of step 435 may include applying one or more filters to the raw, or preprocessed, acoustic signal that leverage a minimum mean square error (MMSE) estimation strategy to suppress noise while preserving signal fidelity, voice activity detection (VAD) algorithms and / or non-negative matrix factorization (NMF) to separate structured background noise or overlapping sounds from sounds made by the subject when using the respiratory device.[000118] FIGs. 5A-5C provide an example of how step 435 may be executed, wherein FIG. 5A provides a graph 501 of an eleven-second raw audio signal showing sound amplitude as a function of time that may be received in step 435. FIG. 5B provides a graph 502 showing a preprocessed version of the raw audio signal of FIG. 5A following removal of low-frequency noise during execution of step 435. FIG. 5C provides a graph showing a smoothed version of the audio signal of FIG. 5B following removal of low-frequency noise during execution and additional smoothing as may be performed during execution of step 435.[000119] In step 440, the preprocessed signal may be analyzed to detect sound amplitude peaks and / or identify events and these peaks and / or events may be used to segment the preprocessed audio signal into one or more window(s) (step 445) so that features (e.g., amplitude overtime, changes in amplitude, signal energy, changes in signal energy, frequency, frequency bands, amplitude peak, segment duration, etc.) may be extracted from one or more of the window(s) (step 450). The extracted features and / or events may then be identified and / or characterized (step 455). For example, identified amplitude peaks may be used during execution of step 450 and / or 455 to, for example, detect, identify, and / or localize prominent acoustic events that may correspond to one or more operational phases of respiratory device / respiratory therapy system use. At times, execution of 450 and / or 455 may include extraction and / or determination of feature vectors, temporally aggregated feature vector sequences, and / or frame-level feature vectors of the preprocessed audio signal and / or an event window included therein.[000120] In some embodiments, detection of peaks and / or events present within an audio signal (step 440 and / or 410) may be executed by, for example, employing amplitude thresholding, temporal smoothing, and / or slope analysis, that, on some occasions, may be optimized for real-time processing and / or robustness to noise artifacts. Slope analysis may be used to, for example, detect rapid changes in slope, which may be indicative of impulsive events such as medication canister (e.g., canister 215) pressurization or actuation. By correlating these slope changes with the timing of amplitude crossings, differentiation between relevant acoustic events and spurious spikes caused by, for example, background transients, noise, and / or subject movement may be achieved so that, for example, only relevant acoustic events are further analyzed and / or incorporated into an audio signature for an acoustic event type.[000121] In some embodiments, execution of step 440 may include application of peak validation heuristics to the preprocessed audio signal. Exemplary peak validation heuristics include, but are not limited to, constraints on minimum peak duration, inter-peak timing (e.g., ensuring inhalation follows actuation within a physiologically plausible event window), and spectral shape analysis to match the expected frequency profile of each inhaler phase. In some cases, a machine-learned peak classifier may be trained to distinguish respiratory-device-use-related events from non-respiratory-device-use-related sounds using, for example, short, windowed feature vectors around each candidate peak.[000122] In some embodiments, a result of the execution of step 445 may be a set of time-stamped event markers for the preprocessed audio signal, wherein each time-stamped event may correspond to a phase of the subject’s interaction with the respiratory device and / or respiratory therapy system. These markers may, for example, enable segmentation of the signal (step 445) for targeted feature extraction and / or serve as high-level indicators for compliance monitoring of a technique a subject employs while using a respiratory device as, for example, described herein.[000123] FIG. 6A provides a graph 601 of an amplitude of the denoised and smoothed audio signal of FIG. 5C (e.g., the audio signal received in step 405) over time (shown in blue) with detected events (in this case, twenty shown in pink) of various lengths of time superimposed thereon that may be generated via, for example, execution of step 445. FIG. 6B provides a graph 602 of a STA / LTA ratio ofthe time series shown in FIG. 6A with a trigger threshold (shown in red) and a detrigger threshold (shown in blue) superimposed thereon.[000124] In step 445, detected events within the preprocessed audio signal may be used to segment the preprocessed audio signal into, for example, shorter event windows using, for example, windowing techniques that, on some occasions, may be chosen for their ability to reduce spectral leakage while preserving transient information. On some occasions, the event windows may overlap in time.[000125] At times, a size and / or duration of an event window may be responsive to size and / or duration to a duration of an event, or characteristics of sound (e.g., a sustained amplitude of sound and / or a pattern of sound amplitude) within a particular event window. In some embodiments, event window duration and / or range may overlap to, for example, balance time and frequency resolution to capture both rapid and sustained acoustic phenomena. Use of overlapping event windows may ensure that events are accurately captured and / or analyzed and / or are not lost between event windows and / or help maintain continuity in temporal dynamics. An example of how step 445 may be executed is provided by FIG. 6C, which shows the preprocessed audio signal of FIG. 6A with various event windows superimposed thereon, wherein each event window corresponds to an amplitude peak (step 440) and / or event (step 445).[000126] In step 450, the preprocessed audio signal and / or one or more event windows therein may be analyzed to, for example, extract features thereof and / or identify characteristics and / or event types therein. Exemplary features include, but are not limited to, sound frequency(ies), amplitudes, power, and / or intensity present within a segment; variations of sound frequency(ies), amplitudes, power, and / or intensity present within a segment; segment duration; and features of other segments within the preprocessed audio signal (e.g., segments that are adjacent in time to the segment under study). In some embodiments, the feature extraction and / or identification of step 450 may include identifying an audio signature for a type of event within one or more segments. In some cases, step 450 may be executed for each individual segment as, for example, an independent analysis unit that may maintain temporal locality so that events that are identified within an event window may be associated with their respective time of occurrence. Additionally, or alternatively, execution of step 450 may include generation of a multi-dimensionalfeature vector that captures a broad range of acoustic characteristics within each segment.[000127] In some embodiments, execution of step 450 may include analysis of one or more segments in the time domain so that, for example, acoustic amplitude and / or waveform structure, which may be used to deduce acoustic signal power may be extracted from the acoustic signal of a particular event window and / or associated with a particular event. This signal power may then be used to, for example, identify and / or classify events (step 415) as, for example, high-energy events (e.g., removal and / or replacement of a cap from / on a respiratory device, placing the respiratory device in the subject’s mouth, etc.), short-time energy events, which may be characterized by variability in amplitude, which may be useful for detecting signal complexity, and / or determine characteristics of signal noise or periodicity which may be used to, for example, detect turbulence in airflow.[000128] Additionally, or alternatively, execution of step 415 may include analysis of one or more event windows in the frequency domain via application of a Fast Fourier Transform (FFT) to one or more event windows to obtain its spectral representation. This spectral representation of the acoustic signal of an event window may be used to, for example, differentiate between tonal and broadband events, distinguish sharp bursts from steady-state inhalation, and / or capture the frequency below which a specified percentage of total spectral energy resides. In some cases, spectral features of an event window may be useful for identifying whether, for example, a signal corresponds to a smooth inhalation or a sudden mechanical click.[000129] Additionally, or alternatively, execution of step 415 may include analysis of peaks within a spectral envelope of an event window, and / or group of event windows, to determine one or more formants (e.g., resonant frequencies of the vocal tract or device cavity) thereof. These formants may be used to, for example, identify one or more resonant frequencies and / or signatures within an audio signal and / or event window thereof that may correspond to resonant frequencies and / or signature(s) of, for example, one or more events (e.g., actuation of a canister (e.g., canister 215) of medication within an inhaler (e.g., respiratory device 201), airflow within an inhaler chamber (e.g., medication conduit 235), airflow through a mouthpiece of a respiratory device (e.g., mouthpiece 220, 222, 223, 224, 225, 226,227, or 228), and / or airflow through a tube of a respiratory device (e.g., medication conduit 235, 237, or 238)), which may provide, for example, device-specific information regarding how the subject is using the respiratory device.[000130] Additionally, or alternatively, execution of step 415 may include computation of one or more derivative features of an audio signal within an event window or plurality of event windows. These derivative features may be used to, for example, capture and / or analyze a dynamic evolution of features within the analyzed audio signal over time, which may provide information about how the acoustic profile changes across event windows, or sets of event windows, within an audio signal. These temporal dynamics may be used to, for example, understand coordination between various events (e.g., inhalation, actuation, exhalation, etc.) of respiratory device / respiratory therapy system usage to, for example, assess subject technique and / or determine compliance with instructions for use.[000131] In some embodiments, one or more results of the analysis of step 450 and / or 415 may be analyzed in isolation to, for example, individually classify and / or characterize of events within each event window. Additionally, or alternatively, one or more results of the analysis of step 450 and / or 415 may be aggregated into a temporal sequence (e.g., a plurality of event windows or the whole audio signal) to, for example, classify and / or characterize of events within the temporal sequence. This may be accomplished using, for example, temporal pooling, statistical summarization, sequence modeling, and / or recurrent neural networks (RNNs) to evaluate subject respiratory device use technique.[000132] FIG. 6C provides a graph 603 showing an event windowed version of the amplitude (in decibels (dB)) of the denoised and smoothed audio signal of FIG.6A over time with a plurality of event windows 610 (in this case, eleven) superimposed thereon. Each event window is associated with an identified event (step 415), wherein a first event window 610A includes a portion of the audio signal with a rapid, sharp, and solitary spike in amplitude, which has the audio signature of a subject removing a cap from a respiratory therapy device’s (in the form of a MDI) mouthpiece; a second event window 610B includes a portion of the audio signal with a period of an oscillating amplitude, which has an audio signature that corresponds to the subject shaking the MDI; a third event window 610C includes rapid, sharp, and solitary spike in amplitude, which has the audio signature of a subject placing themouthpiece of the MDI in his or her mouth; a fourth event window 61 OD includes a portion of the audio signal corresponding to noise; fifth event window 61 OE includes a portion of the audio signal with a sustained period sound with a relatively consistent amplitude, which has an audio signature that corresponds to the subject exhaling; sixth event window 61 OF includes a portion of the audio signal with a period of a moderate (e.g., 1800 to -1800) amplitude that lasts for approximately 0.4s, which has an audio signature that corresponds to the subject actuating release of the medication from the MDI; seventh event window 610G includes a portion of the audio signal with a sustained period (approximately 0.7s) of sound within an amplitude range of 5000 to -5000, which has an audio signature that corresponds to the subject inhaling the medication from the MDI; eighth event window 610H includes a portion of the audio signal with a very low amplitude, which has an audio signature that corresponds to the subject holding his or her breath following inhalation of the medication from the MDI; ninth event window 610J includes a portion of the audio signal with a rapid, sharp, and solitary spike in amplitude, which has the audio signature of a subject replacing the cap on the MDI mouthpiece; and tenth event window 61 OK includes a portion of the audio signal that corresponds to a noise caused by the user when interacting with the respiratory therapy system such as putting the device down and / or replacing the cap to the mouthpiece. In some embodiments, where an event window occurs in time along the sequence of event windows may be used to identify and / or characterize an event within an event window. For example, the audio signal of FIG. 6C has two event windows (first and ninth event window 610A and 610J) of the audio signal that include a rapid, sharp, and solitary spike in amplitude at the beginning and ending of the audio signal. From its position within the time sequence, it may be deduced that that the audio signal included in first event window 610A corresponds to removing the cap from the MDI’s mouthpiece at the beginning of an MDI usage routine and that the audio signal included in ninth event window 610J corresponds to replacing the cap on the MDI’s mouthpiece once the MDI has been used.[000133] FIG. 7A provides a flowchart illustrating a method 700 for training a respiratory event characterization model configured to identify and / or characterize events within an audio signal that includes sound made by a subject when breathing and / or interacting with (e.g., using or preparing for use) a respiratory device and / or arespiratory therapy system like the respiratory devices and / or a respiratory therapy systems disclosed herein. Method 700 may be executed via, for example, any of the systems (e.g., system 300) and / or system components (e.g., cloud computing platform 310, respiratory therapy system 201, 202, 203, 204, 205, 206, and / or 207), and / or respiratory monitoring system (e.g., respiratory monitoring system 209 and / or 211) described herein.[000134] In step 705, a labeled set of audio signals, recordings, and / or portions thereof (e.g., 1,000-10,000,000 audio signals, recordings, and / or portions thereof) of breathing exercises, respiratory device use, and / or respiratory therapy system use may be received. On some occasions, the audio signals and / or recordings received in step 705 may include noise and / or ambient sound and, on other occasions, the audio signals and / or recordings may be pristine (i.e., no noise and / or ambient sound). In some embodiments, the labeled set of audio signals, recordings, and / or portions thereof may be similar to those received in step 405 and / or may be received from the library generated in step 425 of method 400. In some embodiments, execution of step 705 may include receiving and / or accessing the library of step 425.[000135] In some embodiments, the audio signals and / or recordings of step 705 may be analyzed and / or processed to generate one or more vectors that represent the audio signal / recording and / or a portion thereof (e.g., an event). Additionally, or alternatively, the audio signals and / or recordings of step 705 may be and / or include feature vectors, temporally aggregated feature vector sequences, and / or frame-level feature vectors. Labels used to label the labeled set of audio signals / recordings include, but are not limited to, start and stop times for activities performed during the audio signal / recording, activity types performed at certain times within the audio signal / recording, equipment used when making the audio signal / recording, and / or demographic information about a subject generating the audio signal.[000136] In step 710, the labeled data set of audio signals, recordings, and / or portions thereof, may be divided into a training set and testing set (e.g., 60% for training and 40% for testing; 70% for training and 30% for testing; 80% for training and 20% for testing, etc.). In step 715, one or more machine learning and / or Al software and / or architecture inputs may be received. Exemplary inputs received in step 715 include, but are not limited to, how the model is to be generated, trained, tuned, and / or iterated upon via execution of method 700 and / or any other methoddisclosed herein. In some embodiments, the inputs received in step 715 may select a machine learning, or model, architecture that may be, for example, a deep neural network (DNN), a convolutional neural network (CNN) tailored for temporal signal features, and / or a robust ensemble classifier such as a random forest and / or a gradient boosting machine (e.g., XGBoost). In some implementations, recurrent neural networks (RNNs) and / or long short-term memory networks (LSTMs) may be used so that the model may be configured to model temporal dependencies across sequential portions of an audio signal (e.g., frames), so that the model and / or a system using the model may capture the evolution of events like inhalation buildup, actuation burst, and exhalation taper.[000137] In step 720, the training data set of step 710 may be input into a modelgeneration architecture configured with the inputs of step 715 so that a first version of a respiratory event characterization model may be trained and / or generated. The respiratory event characterization model may be configured to identify (e.g., identify event type) and / or characterize (average event amplitude, peak event amplitude, event start time, event stop time, and / or event duration) events within an audio signal that includes sounds generated by a subject when breathing and / or using a respiratory device and / or respiratory therapy system.[000138] In some embodiments, execution of step 720 may include extraction of an acoustic pattern or audio signature of one or more events present in an audio signal / recording and / or feature vector(s) representing an event and evaluation of how closely the extracted acoustic pattern matches known extracted acoustic patterns for the event. The first version of respiratory event characterization model may then be tested using the testing data set (step 725) and iterated upon and / or updated responsively to the testing until a desired level of accuracy is achieved, thereby generating a second version of the respiratory event characterization model (step 730). In step 735, the first and / or second versions of the respiratory event characterization model may be saved and process 700 may end.[000139] FIG. 7B provides a flowchart illustrating a method 701 for training and / or generating a respiratory device use assessment model configured to assess respiratory device and / or respiratory therapy system usage technique. Method 701 may be executed via, for example, any of the systems (e.g., system 300) and / or system components (e.g., cloud computing platform 310, respiratory therapy system201, 202, 203, 204, 205, 206, and / or 207), and / or respiratory monitoring system (e.g., respiratory monitoring system 209 and / or 211) described herein.[000140] Process 701 may be begin with execution of step(s) 705, 710, 715, 720, 725, 730, and / or 735 and / or be a continuation of process 700. In step 740, information regarding proper and improper use of a respiratory device and / or respiratory therapy system may be received. This information may include, but is not limited to, clinically validated procedures, prescribed methods, or techniques, for respiratory device and / or respiratory therapy system use, audio signatures for various events associated with indications of proper and improper use, procedures for the use of respiratory devices and / or respiratory therapy systems, proper and improper event sequences, proper and improper event durations and / or ranges of durations, age and / or morbidity specific event characteristics, and / or respiratory device and / or respiratory therapy system specific event characteristics. For example, instructions for use of a respiratory therapy system embodied as an MDI received in step 735 may include 1 ) prime the inhaler when necessary (prior to their first use and / or every 1-3 weeks), 2) shake the inhaler, 3) exhale fully to empty lungs, 4) remove the cap from the inhaler’s mouthpiece and place it into the subject’s mouth, 5) begin to inhale, 6) dispense medication by, for example, activating the medication inhaler apparatus (e.g., press down on canister of medication), 7) complete a deep inhalation, 8) hold your breath (e.g., 2-10 seconds) following complete dispensation of a dose of medication to allow the medication to settle in the lungs and / or contact lung tissue, and 9) exhale slowly and then breathe normally. Among other things, proper execution of this technique helps maximize deposition of the medication within the lungs, maximizes the effectiveness of the medication, and ensures better management of respiratory conditions. In addition, some medication inhalers need to be primed prior to their first use, on some occasions, over time (e.g., every 1-3 weeks) and this step would be performed prior to step 1 as needed.[000141] In step 742 the information received in step 740 is used to generate an augmented labeled data set. In some embodiments, the information about proper and / or improper respiratory device and / or respiratory therapy system usage may be generalized (i.e., not specific to the data or subjects of the labeled dataset of step 705) to, for example, proper instructions for use for various respiratory device and / or respiratory therapy system and / or common mistakes subjects make when usingrespiratory devices and / or respiratory therapy systems. In these embodiments, execution of step 742 may include analyzing and / or processing the labeled set of audio signals, recordings, and / or portions thereof (e.g., 1,000-10,000,000 audio signals, recordings, and / or portions thereof) of breathing exercises, respiratory device use, and / or respiratory therapy system received in step 905 to determine whether or not they exhibit proper or improper usage technique and then adding an additional label, or annotation, with a result of this determination to the respective audio signals, recordings, and / or portions. In some instances, results of the proper or improper determination may be binary (e.g., 1 for proper, 0 for improper) and / or on a scale (e.g., 1-5 or 1-10) with, for example, a relatively low number indicating improper technique and a relatively high number indicating proper technique. In some cases, the determination of whether or not the audio signals, recordings, and / or portions thereof exhibit proper or improper usage technique may be performed using the second version of the respiratory event characterization model and / or an output thereof, wherein the audio signals, recordings, and / or portions thereof are analyzed using the respiratory event characterization model to characterize (e.g., identify, determine a feature (e.g., length of time, position in a sequence of events, etc.) one or more events within the respective audio signals, recordings, and / or portions thereof. Output of this analysis may be associated with the respective audio signals, recordings, and / or portions thereof as, for example, part of the execution of step 742.[000142] Additionally, or alternatively, the labels associated with the labeled data sets of step 705 may include indications of whether and / or how one or more of the audio signals, recordings, and / or portions thereof are associated with sounds made when the subject of the audio signal, recording, and / or portion thereof is executing proper and / or improper respiratory device and / or respiratory therapy system usage technique and / or proper / improper indications for the audio signals, recordings, and / or portions thereof of the labeled data set may be received in step 740 and associated with the respective audio signals, recordings, and / or portions thereof in step 742 to generated the augmented labeled data set.[000143] In step 744, the augmented labeled data set may be divided into a training set and testing set (e.g., 60% for training and 40% for testing; 70% for training and 30% for testing; 80% for training and 20% for testing, etc.). In step 744,one or more machine learning and / or Al software and / or architecture inputs may be received. In some embodiments, execution of step 744 may be similar to execution of step 715 and the machine learning and / or Al software and / or architecture inputs may, or may not, be specific to training a respiratory device use assessment model.[000144] In step 748, the training data set of step 744 may be input into a modelgeneration architecture optionally configured with the inputs of step 744 so that a first version of a respiratory device use assessment model may be trained and / or generated. The respiratory device use assessment model may be configured to evaluate and / or assess a subject’s technique when using a respiratory device and / or respiratory therapy system like the ones disclosed herein to, for example, determine whether the subject is using the respiratory device and / or respiratory therapy system in a manner consistent with its prescribed use (e.g., proper) and, if not, how that use is inconsistent with its prescribed use (e.g., improper).[000145] The first version of respiratory device use assessment model may then be tested using the testing data set (step 750) and iterated upon and / or updated responsively to the testing until a desired level of accuracy is achieved, thereby generating a second version of the respiratory device use assessment model (step 752). In step 735, the first and / or second versions of the respiratory device use assessment model may be saved and process 701 may end.[000146] FIG. 7C provides a flowchart illustrating a method 702 for training and / or generating a breathing assessment model configured to assess respiratory device and / or respiratory therapy system usage technique. Method 702 may be executed via, for example, any of the systems (e.g., system 300) and / or system components (e.g., cloud computing platform 310, respiratory therapy system 201, 202, 203, 204, 205, 206, and / or 207), and / or respiratory monitoring system (e.g., respiratory monitoring system 209 and / or 211) described herein.[000147] Process 702 may be begin with execution of step(s) 705, 710, 715, 720, 725, 730, and / or 735 and / or be a continuation of process 700. In step 760, information regarding breathing techniques executed during the generation of the labeled data set of audio signals and / or recordings, breathing exercise execution, and / or noise may be received. In some cases, the information received in step 760 may include indications of proper and improper (e.g., compliant and / or non-compliant) execution of a breathing technique(s). Information received in step 760may include, but is not limited to, clinically validated procedures, prescribed methods, and / or defined techniques for performing breathing exercises to, for example, practice mindfulness (e.g., focus and / or relaxation), train for childbirth, practice using a respirator, and / or practice breathing in low-oxygen situations as may occur when using a tank of air while firefighting or scuba diving. In some cases, the information may include routines, or procedures, for when and / or how often to inhale or exhale over a time period, whether to inhale and / or exhale through the nose and / or mouth, how long inhalations and / or exhalations are recommended to take, and / or when to hold the breath.[000148] In step 762 the information received in step 760 is used to generate an augmented labeled data set. In some embodiments, the information about proper and / or improper execution of breathing techniques may be generalized (i.e., not specific to the data or subjects of the labeled dataset of step 705) to, for example, instructions for proper, or preferred, execution of breathing exercises performed while, for example, a subject is using a respiratory monitoring systems like the respiratory monitoring systems disclosed herein. In these embodiments, execution of step 762 may include analyzing and / or processing the labeled set of audio signals, recordings, and / or portions thereof (e.g., 1,600-10,000,000 audio signals, recordings, and / or portions thereof) of breathing exercises received in step 905 to determine whether or not they exhibit proper, preferred and / or compliant breathing technique execution and then adding an additional label, or annotation, with a result of this determination to the respective audio signals, recordings, and / or portions. In some instances, results of the proper or improper determination may be binary (e.g., 1 for proper, 0 for improper) and / or on a scale (e.g., 1-5 or 1-10) with, for example, a relatively low number indicating improper technique and a relatively high number indicating proper technique. In some cases, the determination of whether or not the audio signals, recordings, and / or portions thereof exhibit proper or improper execution of breathing technique may be performed using the second version of the respiratory event characterization model and / or an output thereof, wherein the audio signals, recordings, and / or portions thereof are analyzed using the respiratory event characterization model to characterize (e.g., identify, determine a feature (e.g., length of time, position in a sequence of events, etc.)) one or more events within the respective audio signals, recordings, and / or portions thereof. Output of this analysismay be associated with the respective audio signals, recordings, and / or portions thereof as, for example, part of the execution of step 762.[000149] Additionally, or alternatively, the labels associated with the labeled data sets of step 705 may include indications of whether and / or how one or more of the audio signals, recordings, and / or portions thereof are associated with sounds made when the subject of the audio signal, recording, and / or portion thereof is executing proper and / or improper breathing technique and / or proper / improper indications for the audio signals, recordings, and / or portions thereof of the labeled data set may be received in step 760 and associated with the respective audio signals, recordings, and / or portions thereof in step 762 to generated the augmented labeled data set.[000150] In step 764, the augmented labeled data set may be divided into a training set and testing set (e.g., 60% for training and 40% for testing; 70% for training and 30% for testing; 80% for training and 20% for testing, etc.). In step 764, one or more machine learning and / or Al software and / or architecture inputs may be received. In some embodiments, execution of step 764 may be similar to execution of step 715 and the machine learning and / or Al software and / or architecture inputs may, or may not, be specific to training a breathing assessment model.[000151] In step 768, the training data set of step 764 may be input into a modelgeneration architecture optionally configured with the inputs of step 764 so that a first version of a breathing assessment model may be trained and / or generated. The breathing assessment model may be configured to evaluate and / or assess a subject’s technique when breathing and / or performing a breathing exercise to, for example, determine whether the subject is breathing correctly and / or executing a breathing technique properly.[000152] The first version of breathing assessment model may then be tested using the testing data set (step 770) and iterated upon and / or updated responsively to the testing until a desired level of accuracy is achieved, thereby generating a second version of the breathing assessment model (step 772). In step 735, the first and / or second versions of the breathing assessment model may be saved and process 702 may end.[000153] FIG. 8 provides a flowchart illustrating another exemplary method 800 for training a respiratory device use assessment model configured to assess subject’s technique when using a respiratory device and / or system like therespiratory devices and / or a respiratory therapy systems disclosed herein. Method 800 may be executed via, for example, any of the systems (e.g., system 300) and / or system components (e.g., cloud computing platform 310 and / or respiratory therapy system 201, 202, 203, 204, 205, 206, and / or 207) described herein.[000154] In step 805, information regarding respiratory device and / or respiratory therapy system usage technique may be received. Exemplary information received in step 805 includes, but is not limited to, a correct sequence and / or timing of events for respiratory device and / or system usage. For example, when the respiratory device is an MDI, the information received in step 805 may include a correct, or clinically validated, sequence and timing of MDI shaking, priming, actuation, and a full inhalation with sufficient flow rate and duration. Additionally, or alternatively, information received in step 805 may include audio signatures, feature vectors, temporally aggregated feature vector sequences, and / or frame-level feature vectors for one or more events that occur (or should occur) when a subject is using a respiratory device and / or respiratory therapy system. Additionally, or alternatively, information received in step 805 may include indications for proper and / or improper respiratory device respiratory therapy system usage technique(s) and / or a level of importance (e.g., how much proper or improper usage technique may impact a subject’s health or the efficacy of a treatment delivered and / or provided via the respiratory device and / or respiratory therapy system) for the indications. On some occasions, execution of step 805 may be similar to execution of step 735, described above.[000155] In step 810, a labeled set of audio signals, recordings, and / or portions thereof (e.g., 1,000-10,000,000 audio signals, recordings, and / or portions thereof) of a subject using a respiratory device and / or respiratory therapy system may be received. On some occasions, execution of step 810 may be similar to execution of step 405 and / or 705, described above and / or may include receiving and / or accessing the library of step 425.[000156] In step 815, the set of labeled audio signals, recordings, and / or portions thereof, may be divided into a training set and testing set. In step 820, one or more machine learning and / or Al software and / or architecture inputs may be received. On some occasions, a manner in which step 815 is executed and / or 820 may be similar to execution of step 710 and / or 715, respectively.[000157] In step 825, the training data set of step 815 may be input into a modelgeneration architecture configured with the inputs of step 820 so that a first version of a respiratory device use assessment model may be trained and / or generated. The respiratory device use assessment model may be configured to evaluate an audio signal and / or recording of a subject using a respiratory device and / or respiratory therapy system to determine one or more aspects, or characteristics, of how the subject is using the respiratory device and / or respiratory therapy system and / or determine if the subject is properly using the respiratory device and / or respiratory therapy system and, optionally, when the user is improperly using the respiratory device and / or respiratory therapy system, determine why and / or how to fix the subject’s technique. Additionally, or alternatively, the respiratory device use assessment model may be configured to provide one or more recommendations for a change (e.g., increased or decrease dosage or use, changing a time of day of use, changing a type of respiratory device and / or respiratory therapy system being used, etc.) in the routine and / or technique of respiratory device and / or respiratory therapy system usage.[000158] In some embodiments, execution of step 825 may include extraction of an acoustic pattern of one or more events present in an audio signal / recording and / or feature vector(s) representing an event and evaluation of how closely the extracted acoustic pattern matches known extracted acoustic patterns and / or audio signatures for the event. Additionally, or alternatively, execution of step 825 may include analysis of a training data set to identify (e.g., identify event type) and / or characterize (average event amplitude, peak event amplitude, event start time, event stop time, and / or event duration) events within an audio signal that includes sounds generated by a subject using a respiratory device and / or respiratory therapy system as, for example, described above with regard to step(s) 410, 415, 435, 440, 445, 450, and / or 455.[000159] In some embodiments, the respiratory device use assessment model may be configured to generate a binary classification output (e.g., correct vs. incorrect). Additionally, or alternatively, a respiratory device use assessment model may be configured to perform multi-class or multi-label respiratory device use classification, which may enable the detection of specific types of usage errors. For example, a respiratory device use assessment model may be configured to identify:(1) insufficient inhalation force, detected by low sustained energy and absence of turbulence patterns; (2) improper timing of actuation, such as when the spray occurs before or after inhalation instead of during it; (3) short or incomplete inhalation, which presents as a truncated airflow segment; or (4) skipping the priming step, which is inferred from the absence of a low-amplitude, low-frequency pre-actuation event. Optionally, one or more error classifications generated by the respiratory device use assessment model may be mapped to, for example, clinical recommendations and / or subject feedback messages.[000160] The first version respiratory device use assessment model may then be tested using the testing data set (step 830) and iterated upon and / or updated responsively to the testing until a desired level of accuracy is achieved, thereby generating a second version of the respiratory device use assessment model (step 835). At step 840, the first and / or second version of the respiratory device use assessment model may be saved.[000161] FIG. 9 provides a flowchart illustrating an exemplary method 900 for using a respiratory event characterization model to identify and / or characterize events within an audio signal that includes sound made by a subject when interacting with and / or using a respiratory device and / or a respiratory therapy system like the respiratory devices and / or a respiratory therapy systems disclosed herein. Additionally, or alternatively, method 900 may be executed to assess a subject’s respiratory device and / or respiratory therapy system usage technique using a respiratory device use assessment model. Method 900 may be executed via, for example, any of the systems (e.g., system 300) and / or system components (e.g., cloud computing platform 310 and / or respiratory therapy system 201, 202, 203, 204, 205, 206, and / or 207) described herein.[000162] In step 905, an audio signal that captures sounds made by a subject using a respiratory device and / or respiratory therapy system may be received. In some embodiments, the audio signal of step 905 may be a raw analog and / or digital audio signal captured by a microphone like microphone 120 and / or a digital signal corresponding to an analog audio signal that has been converted into the digital signal by, for example, an analog-to-digital converter like ADC 125.[000163] Optionally, in step 910, the received audio signal may be preprocessed and / or analyzed to generate a preprocessed audio signal. Preprocessing the audiosignal may, for example, remove noise and / or unwanted acoustic interference from the audio signal that could degrade the performance of subsequent feature extraction and classification of sounds within the preprocessed signal. Exemplary unwanted acoustic interference may originate from, for example, household appliances, speech, traffic, HVAC systems, lawn equipment, the subject’s movement, and noise generated by others in proximity to the subject. In some embodiments, execution of step 910 may be similar to execution of step 410, 435, 440, 445, 450, and / or 455.[000164] In step 915, the audio signal received in step 905, the preprocessed audio signal of step 910, and / or the portion thereof may be input into a respiratory event characterization model (e.g., the respiratory event characterization model of step 730) and / or respiratory device use assessment model (e.g., the respiratory device use assessment model of step 772 and / or 1135) to, for example, identify and / or characterize events for which sounds have been captured in the audio signal received, the preprocessed audio signal, and / or the portion thereof and / or assess respiratory device usage technique. Additionally, or alternatively, execution of step 915 may include inputting a result of the preprocessing of step 910 into the respiratory event characterization model. The respiratory event characterization model may analyze the input of step 915 and generate an output (step 920) to, for example, identify and / or characterize one or more events included in the audio signal by, for example, determining when, in time, the event occurs, an event type, an event duration, an event amplitude, an order of two or more events within an audio signal, a type of respiratory therapy system used, respiratory therapy system type, a technique or procedure the subject used when using the respiratory therapy system, and / or whether sounds made by the respiratory therapy system are within expected and / or clinically validated sounds made by respiratory therapy systems that share characteristics with the respiratory therapy system used when making the audio signal. In some embodiments, this characterization and / or identification may be similar to execution of step(s) 410, 415, 435, 440, 445, 450, and / or 455 and / or may use one or more audio signatures, such as the audio signatures of step 425 and / or information associated therewith.[000165] In step 925, the output of the step 920 may be validated and / or tested to determine, for example, a confidence level and / or verify that it corresponds to oneor more expected outputs. When the output is not valid, an error message may be communicated to, for example, display device and / or terminal (e.g., a display device of a user device like user device 360, a computer like computer 330, a display device like display device 340, a cloud computing platform like cloud computing platform 310, and / or a third party user like third party user 375) (step 930).[000166] When the output of step 920 is valid, it may optionally be used to assess respiratory device usage technique to, for example, evaluate how the subject is using the respiratory therapy system and / or determine whether the subject is using the respiratory therapy system in a recommended, clinically validated, and / or prescribed manner (e.g., whether the subject executed different events in the proper sequence and / or for a proper length of time, whether the subject completed a treatment routine, whether the subject executed all events within a prescribed routine, etc.). Additionally, or alternatively, in some embodiments, a subject characteristic may be determined via execution of step 935. In these embodiments, the determined subject characteristic may be, for example, an indication of lung health (e.g., peak expiratory flow, whether there is fluid in the lungs, whether the lungs are obstructed as may occur with COPD, etc.).[000167] In some embodiments, execution of step 935 may include a comparison of two separate results of process 900 for a particular subject that were performed at different times (e.g., separated by a day, a plurality of days, a week, a month, etc.) to, for example, assess the subject’s disease progression and / or respiratory health over time to, for example, determine if use of the respiratory therapy system has slowed disease progression and / or improved the respiratory health of the subject over time and / or determine if a change in respiratory therapy system use and / or routine is recommended and / or needed. Additionally, or alternatively, execution of step 935 may include evaluating the output of step 920 (e.g., a characterization of an event within an audio signal) using a library (e.g., the library of step 425) of audio signatures that correlates one or more identified event(s) with an annotation corresponding to the identified event(s) to determine a feature thereof. For example, the assessment of step 930 may include receiving an output in step 920 that indicates that a portion of an audio signal that corresponds to an inhalation of medication event and that characterization may be used to query a library of audio signatures (e.g., the library of step 425) for inhalation of medicationevents that match, or are similar to, the characterization of step 920. Annotations, or evaluations, that correspond to matching inhalation of medication events found in the library may then be used to assess respiratory usage technique in step 935. For example, if an inhalation of medication event characterized in step 920 corresponds to an inhalation of medication event audio signature in the library associated with an annotation that indicates it corresponds to proper inhalation of medication, that annotation may be used to assess respiratory device usage technique in step 935.[000168] In step 940, an indication of one or more results of execution of steps 920 and / or 935 may be provided to a display device and / or user (e.g., a user device like user device 360, a computer like computer 330, a display device like display device 340, a cloud computing platform like cloud computing platform 310, and / or a third party user like third party user 375). In some embodiments, the indication may be an event identification and / or classification. Additionally, or alternatively, the indication may be an evaluation of the subject’s respiratory therapy system usage technique, an indication of the subject’s health, and / or a recommendation.[000169] FIG. 10 provides a flowchart illustrating an exemplary method 1000 for using a respiratory device use assessment model to assess a subject’s use of a respiratory device and / or respiratory therapy system. Method 1000 may be executed via, for example, any of the systems (e.g., system 300) and / or system components (e.g., cloud computing platform 310 and / or respiratory therapy system 201, 202, 203, 204, 205, 206, and / or 207) described herein.[000170] Initially, step(s) 905 and, optionally, step 910 may be performed and an audio signal and / or recording and / or preprocessed audio signal and / or recording may be input into a respiratory device use assessment model like the respiratory device use assessment model generated via execution of method 701 and / or 800 (step 1005). The respiratory device use assessment model may, for example, characterize the audio signal and / or recording and / or preprocessed audio signal and / or recording as, for example, described herein to, for example, identify events and / or characteristics of identified events in order to evaluate whether the subject has correctly used the respiratory device and / or respiratory therapy system as, for example, described herein. Additionally, or alternatively, step 1005 may be executed by inputting an output from a respiratory event characterization model into the respiratory device use assessment model to evaluate whether the subject hascorrectly used the respiratory device and / or respiratory therapy system as, for example, described herein.[000171] Optionally, in some embodiments, the output of step 1010 may be evaluated to, for example, determine a subject’s compliance with a recommended or prescribed technique for using a respiratory device and / or respiratory therapy system as, for example, described herein (step 1015). At times, execution of step 1015 may include using a library (e.g., the library of method 400) of event characterizations and / or known / pre-determined audio signatures that correlate one or more events and / or event characterizations of the audio signal, audio signal recording, and / or preprocessed audio signal and / or audio signal recording of step 905 and / or 910 with an annotation or evaluation (e.g., proper technique, improper technique, etc.) of same (step 1015).[000172] Output from the respiratory device use assessment model (step 1010) and / or a result of evaluation of step 1015 may then be optionally verified and / or validated (step 1020) and / or a confidence level for the output and / or evaluation may be determined. When the output and / or evaluation is not valid, an error message may be sent to the subject (step 1025). In some embodiments, execution of step 1020 may include determining how the output and / or evaluation is invalid and / or how a subject may improve the validity and, in these embodiments, the error message of step 1025 may include these determinations and / or recommendations for improving respiratory device and / or respiratory therapy system usage technique. For example, one or more time stamped events present in a received audio signal may be analyzed to determine whether the subject inhaled immediately after actuation of a medication canister of an inhaler respiratory device and / or inhaled for a sufficient duration of time following actuation. Additionally, or alternatively, execution of step 1020 may include determination of a confidence metric, which may quantify the respiratory device use assessment model’s certainty in its classification and / or evaluation decision. On some occasions (e.g., when respiratory device use assessment model is a neural network model), this confidence metric may indicate the probability that the respiratory device usage matches a correct or incorrect pattern so that, for example, the system may be able to distinguish between high-confidence misuses that warrant medical and / or clinical intervention and low-confidence cases that may require reanalysis or additional data.[000173] In step 1030, an indication of the assessment the subject’s respiratory device utilization (e.g., output of step 1010 and / or evaluation of step 1015) may be provided to the subject and / or a display device. In some cases, execution of step 1030 includes providing the indication to a mobile application interface (e.g., software application 365) running on a user device (e.g., user device 360) and optionally uploaded to a third party device (e.g., third party user A-N 375A-375N) such as a clinician-facing dashboard. Execution of step 1030 enables subjects to receive immediate feedback and coaching on proper respiratory device usage technique, while allowing healthcare providers to monitor adherence trends, detect usage errors, and intervene when necessary to improve therapeutic outcomes. For example, when the subject’s technique is deemed correct with high confidence, the indication may be a confirmation message, such as “Good job! Proper usage detected.” If the subject’s technique is deemed incorrect, the indication may provide targeted guidance, such as “Try to inhale more deeply during actuation” or “Remember to shake the inhaler before use.” Indications that include messages of this type may be also include other media (e.g., icons, images, animations, etc.) and / or links (e.g., hyperlinks) by which to get more information. Thus, execution of method 1000 transforms passive respiratory device usage into an active, intelligent monitoring system that supports precision respiratory care.[000174] In some cases, execution of step 1030 may include packaging output of steps 1010 and / or 1015 together with a rich set of metadata from, for example, the user device that may, for example, enhance traceability, personalization, and clinical relevance for the output of steps 1010 and / or 1015. This metadata may include, but is not limited to, one or more timestamps, subject information, device ID, firmware version, geolocation, session duration, signal quality metrics (e.g., signal-to-noise ratio), and / or battery status of the inhaler or mobile device. On some occasions, a plurality of indications and / or metadata may be aggregated over time to, for example, generate compliance reports, trend visualizations, and predictive analytics, allowing providers to identify subjects at risk of poor disease control due to improper inhaler technique. Ultimately, this feedback loop between subjects, devices, and clinicians supports a comprehensive, real-time adherence ecosystem that enhances therapeutic outcomes and promotes long-term behavioral change.[000175] Additionally, or alternatively, execution of step 1030 may include deployment of an escalation mechanism for clinical oversight. For example, when repeated misuse is detected over a configurable time window (e.g., more than three incorrect uses within 48 hours), the indication may be used to automatically trigger alerts or recommendations to the healthcare provider via, for example, a secure clinician dashboard, SMS / email notification, or integration with an electronic health record (EHR) system via FHIR or HL7 APIs. These alerts may include a summary of the detected error types, usage trends, and suggested clinical actions, such as scheduling a telehealth consultation or adjusting medication instructions.[000176] The systems, devices, and method disclosed herein may be used to, for example, provide a respiratory therapy device and / or drug delivery device with an embedded and / or attached microphone configured to capture acoustic signals associated with use and / or medication delivery events, including shaking, priming, actuation, and inhalation. On some occasions, the microphone may be placed and / or configured to optimize signal fidelity based on, for example, internal airflow and mechanical resonance characteristics of the respiratory therapy device and / or drug delivery device and / or a housing thereof. The acoustic signals may be analyzed using one or more methods and / or models disclosed herein to, for example, determine one or more characteristics of the user’s technique while using the respiratory therapy device and / or drug delivery device and / or drug deployment events.[000177] On some occasions, acoustic signals detected by the microphones disclosed herein (or digitized versions thereof) may be communicated to a processor (e.g., a mobile device, a computer, and / or a cloud computing environment) using one or more wireless communication protocols (e.g., Bluetooth Low Energy (BLE)) with, for example, real-time buffering and / or lossless data transfer. A receiving processor may be running a software program configured to, for example, receive the digitized acoustic signals and, for example, maintain low-latency BLE communication while storing raw PCM audio along with, for example, device telemetry and / or session metadata.[000178] On some occasions, digitized acoustic signals may be preprocessed using, for example, bandpass filtering, adaptive noise suppression using, for example, spectral subtraction and / or Wiener filtering, event-based segmentation,and / or a dynamic noise profiling system configured to, for example, learn and subtract background noise profiles in real-time from digitized acoustic signals and / or recordings to enhance acoustic event clarity.[000179] In some embodiments, digitized acoustic signals and / or preprocessed digitized acoustic signals may be processed to detect peaks therein using, for example, a peak detection algorithm that employs, for example, amplitude thresholding, temporal smoothing, and / or slope analysis to identify discrete inhaler-related events in a denoised acoustic signal. On some occasions, detected peaks may be validated using, for example, one or more multi-dimensional feature vector(s) and / or physiological timing constraint(s) between expected acoustic events.[000180] In some embodiments, the digitized acoustic signals and / or preprocessed digitized acoustic signals may be divided into one or more portions, frames, or windows for further processing. This division (also referred to herein as “windowing”) may be performed using one or more peaks detected within the digitized acoustic signals and / or preprocessed digitized acoustic signals so that, for example, each window has a peak therein. One or more of the windows may then be further analyzed to extract, for example, time-domain, frequency domain, and / or cepstral features (e.g., delta coefficients and / or delta-delta coefficients) thereof that may include, for example, MFCCs and / or spectral entropy. On some occasions, these features may be used to capture dynamic acoustic signatures of one or more events (e.g., actuation, inhalation, exhalation, shaking, etc.) present within the digitized acoustic signals and / or preprocessed digitized acoustic signals.[000181] Also disclosed herein are one or more artificial intelligence and / or machine learning model(s) trained using labeled audio recordings of respiratory therapy device and / or drug delivery device and configured to classify usage technique as, for example, correct or incorrect with a probabilistic confidence score. Additionally, alternatively, the artificial intelligence and / or machine learning model(s) may be configured to identify misuse of a respiratory therapy device and / or drug delivery device and / or distinguish between different types of misuse including, for example, early actuation, insufficient inhalation force, skipped exhalation, and skipped priming.[000182] On some occasions, the systems, devices, and methods disclosed herein may be configured to provide feedback to a user and / or a caregiver regarding his / her / their technique. At times, this feedback may be accompanied by one or moreoptional calibrated confidence metrics (e.g., SoftMax probability with Platt scaling) to, for example, support real-time usage feedback and alert generation. In some embodiments, the systems, devices, and / or method disclosed herein may be configured to generate and / or provide user feedback in a mobile application running on the user’s device based on, for example, classification results, including text, visual cues, and personalized recommendations. Additionally, or alternatively, the systems, devices, and methods disclosed herein may be configured to provide a thresholdbased clinical escalation protocol that transmits alerts to a clinician when repeated improper respiratory therapy device and / or drug delivery device use is detected.[000183] Additionally, or alternatively, the systems, devices, and methods disclosed herein may be configured to provide metadata-enhanced session management of respiratory therapy device and / or drug delivery device usage via, for example, association of metadata (e.g., timestamp, user ID, device ID, and environmental context) with the acoustic signals and / or recordings configured to enhance usage traceability and / or compliance tracking. Additionally, or alternatively, the metadata and acoustic signals from a plurality of users may be aggregated to enable, for example, population health analytics to, for example, assess regional and / or environmental links to respiratory health.[000184] FIG. 11 provides a flowchart illustrating another exemplary method 1100 for training a breathing assessment model configured to assess a subject’s breathing patterns and / or performance of a breathing exercise or routine while, for example, wearing and / or using one or more of the respiratory monitoring systems disclosed herein. Method 1100 may be executed via, for example, any of the systems (e.g., system 300) and / or system components (e.g., cloud computing platform 310, respiratory therapy system 201, 202, 203, 204, 205, 206, and / or 207), and / or respiratory monitoring system(s) (e.g., respiratory monitoring system(s) 209 and / or 211) described herein.[000185] In step 1105, information regarding breathing techniques and / or how to properly and / or improperly execute them may be received. Exemplary information received in step 1105 includes, but is not limited to, a name for the breathing technique, a correct sequence and / or timing of events (e.g., inhaling, exhaling, breath holding, etc.) for performance of a breathing technique and / or what indicatesproper / improper breathing technique. On some occasions, execution of step 1105 may be similar to execution of step 760, described above.[000186] In step 1110, a labeled set of audio signals, recordings, and / or portions thereof (e.g., 1,000-10,000,000 audio signals, recordings, and / or portions thereof) of a subject executing a breathing technique may be received. On some occasions, execution of step 1110 may be similar to execution of step 405 and / or 705, described above and / or may include receiving and / or accessing the library of step 425.[000187] In step 1115, the set of labeled audio signals, recordings, and / or portions thereof, may be divided into a training set and testing set. In step 1120, one or more machine learning and / or Al software and / or architecture inputs may be received. On some occasions, a manner in which step 1115 is executed and / or 1120 may be similar to execution of step 710 and / or 715, respectively.[000188] In step 1125, the training data set of step 1115 may be input into a model-generation architecture configured with the inputs of step 1120 so that a first version of a breathing assessment model may be trained and / or generated. The breathing assessment model may be configured to evaluate an audio signal and / or recording generated by a subject while breathing and / or executing a breathing technique. The subject may be breathing during sleep, waking hours, while exercising, and / or performing breathing technique. The audio signal, audio signal recording, and / or a portion thereof may be received from and / or via a component of a sensor system like sensor system 100 and / or a component thereof (e.g., microphone 120).[000189] In some embodiments, execution of step 1125 may include extraction of an acoustic pattern of one or more events present in an audio signal / recording and / or feature vector(s) representing an event and evaluation of how closely the extracted acoustic pattern matches known extracted acoustic patterns and / or audio signatures for the event. Additionally, or alternatively, execution of step 1125 may include analysis of a training data set to identify (e.g., identify event type) and / or characterize (average event amplitude, peak event amplitude, event start time, event stop time, and / or event duration) events within an audio signal that includes sounds generated by a subject using a respiratory device and / or respiratory therapy system as, for example, described above with regard to step(s) 410, 415, 435, 440, 445, 450, and / or 455.[000190] In some embodiments, the breathing assessment model may be configured to generate a binary classification output (e.g., correct vs. incorrect). Additionally, or alternatively, a breathing assessment model may be configured to perform multi-class or multi-label breathing classification, which may enable the detection of specific types of breathing patterns (or the lack thereof) and / or technique execution errors. For example, a breathing assessment model may be configured to identify when and for how long a user inhales, exhales, and / or holds his or her breath to, for example, determine whether, for example, the subject is breathing normally, has sleep apnea, is breathing efficiently, and / or is breathing in a manner that is consistent with a breathing technique. Optionally, one or more determinations and / or error classifications generated by the breathing assessment model may be mapped to, for example, clinical recommendations and / or subject feedback messages.[000191] The first version breathing assessment model may then be tested using the testing data set (step 1130) and iterated upon and / or updated responsively to the testing until a desired level of accuracy is achieved, thereby generating a second version of the breathing assessment model (step 1135). At step 1140, the first and / or second version of the breathing assessment model may be saved.[000192] FIG. 12 provides a flowchart illustrating an exemplary method 1200 for using a respiratory event characterization model to identify and / or characterize events within an audio signal that includes sound made by a subject when breathing and / or using a respiratory device, a respiratory therapy system, and / or respiratory monitoring system(s) like the respiratory devices, a respiratory therapy systems, and / or respiratory monitoring system(s) disclosed herein. Method 1200 may be executed via, for example, any of the systems (e.g., system 300) and / or system components (e.g., cloud computing platform 310 respiratory therapy system 201, 202, 203, 204, 205, 206, and / or 207, and / or respiratory monitoring system(s) 209 and / or 211) described herein.[000193] In step 1205, an audio signal that captures sounds made by a subject while breathing and / or using a respiratory device, respiratory therapy system, and / or respiratory monitoring system(s) may be received. In some embodiments, the audio signal of step 1205 may be a raw analog and / or digital audio signal captured by a microphone like microphone 120 and / or a digital signal corresponding to an analogaudio signal that has been converted into the digital signal by, for example, an analog-to-digital converter like ADC 125.[000194] Optionally, in step 1210, the received audio signal may be preprocessed and / or analyzed to generate a preprocessed audio signal. In some embodiments, execution of step 1210 may be similar to execution of step 410, 435, 440, 445, 450, 455 and / or 910.[000195] In step 1215, the audio signal received in step 1205, the preprocessed audio signal of step 1210, and / or the portion thereof may be input into a respiratory event characterization model (e.g., the respiratory event characterization model of step 730) and / or breathing assessment model (e.g., the breathing assessment model of step 772 and / or 1135) to, for example, identify and / or characterize events corresponding to sounds included in the audio signal and / or recording of step 1205, the preprocessed audio signal and / or recording of step 1210 assess the subject’s breathing patterns and / or breathing exercise execution and / or technique.[000196] The respiratory event characterization model may analyze the input of step 1215 and generate an output (step 1220) to, for example, identify, evaluate, and / or characterize one or more events included in the audio signal by, for example, determining when, in time, the event occurs, an event type, an event duration, an event amplitude, an order of two or more events within an audio signal, a type of device and / or system used to obtain the audio signal of step 1205, a technique or procedure the subject used when breathing or performing a breathing exercise, and / or whether sounds made by the subject, device, and / or system are within expected and / or clinically validated sounds. In some embodiments, this characterization, evaluation, and / or identification may be similar to execution of step(s) 410, 415, 435, 440, 445, 450, 455, and / or 920 and / or may use one or more audio signatures, such as the audio signatures of step 425 and / or information associated therewith.[000197] In step 1225, the output of the step 1220 may be validated and / or tested to determine, for example, a confidence level and / or verify that it corresponds to one or more expected outputs. When the output is not valid, an error message may be communicated to, for example, display device and / or terminal (e.g., a display device of a user device like user device 360, a computer like computer 330, a display device like display device 340, a cloud computing platform like cloud computing platform 310, and / or a third party user like third party user 375) (step 1230).[000198] When the output of step 1220 is valid, it may optionally assess and / or be used to assess the subject’s breathing patterns and / or breathing exercise technique to, for example, evaluate how the subject is using the respiratory therapy system, determine whether the subject has a breathing problem (e.g., sleep apnea, rapid breathing, depressed breathing, shallow breathing, etc.), determine whether the subject is executing the breathing technique properly, and / or determine whether the subject is using the respiratory therapy system, respiratory device, and / or breathing monitoring system in a recommended, clinically validated, and / or prescribed manner (e.g., whether the subject executed different events in the proper sequence and / or for a proper length of time, whether the subject completed a breathing routine, whether the subject executed all events within a prescribed breathing routine, etc.). Additionally, or alternatively, in some embodiments, a subject characteristic may be determined via execution of step 1235. In these embodiments, the determined subject characteristic may be, for example, an indication of the subjects overall and / or lung health (e.g., whether the subject has troubled breathing, has a respiratory condition, etc.).[000199] In some embodiments, execution of step 1235 may include a comparison of two separate results of process 1200 for a particular subject that were performed at different times (e.g., separated by a day, a plurality of days, a week, a month, etc.) to, for example, assess the subject’s progression and / or respiratory health over time. Additionally, or alternatively, the breathing assessment model may be configured to provide one or more recommendations for a change (e.g., increased or decrease dosage or use, changing a time of day of use, changing a type of respiratory device and / or respiratory therapy system being used, etc.) in the routine and / or breathing technique used.[000200] In step 1240, an indication of one or more results of execution of steps 1220 and / or 1235 may be provided to a display device and / or user (e.g., a user device like user device 360, a computer like computer 330, a display device like display device 340, a cloud computing platform like cloud computing platform 310, and / or a third party user like third party user 375). In some embodiments, the indication may be an event identification and / or classification. Additionally, or alternatively, the indication may be an evaluation of the subject’s breathing patterns, an evaluation ofthe subject’s breathing technique execution, an indication of the subject’s health, and / or a recommendation.[000201] FIG. 13 provides a flowchart illustrating an exemplary method 1300 for using a breathing assessment model to assess a subject’s breathing patterns and / or breathing exercise execution technique. Method 1300 may be executed via, for example, any of the systems (e.g., system 300) and / or system components (e.g., cloud computing platform 310 and / or respiratory therapy system 201, 202, 203, 204, 205, 206, and / or 207) described herein.[000202] Initially, step(s) 1205 and, optionally, step 1210 may be performed and an audio signal and / or recording and / or preprocessed audio signal and / or recording may be input into a breathing assessment model like the breathing assessment model generated via execution of method 702 and / or 1100 (step 1305). The breathing assessment model may, for example, characterize the audio signal and / or recording of step 1205 and / or the preprocessed audio signal and / or recording of step 1210 as, for example, described herein to, for example, identify events and / or characteristics of identified events in order to characterize, assess, and / or evaluate the subject’s breathing patterns and / or breathing technique execution. Additionally, or alternatively, step 1305 may be executed by inputting an output from a respiratory event characterization model into the breathing assessment model to, for example, identify events and / or characteristics of identified events in order to characterize, assess, and / or evaluate the subject’s breathing patterns and / or breathing technique execution.[000203] Optionally, in some embodiments, the output of step 1310 may be evaluated to, for example, determine a breathing pattern for the subject and / or determine whether the subject has properly executed a breathing technique (step 1315). At times, execution of step 1315 may include using a library (e.g., the library of method 400) of event characterizations and / or known / pre-determined audio signatures that correlate one or more events and / or event characterizations of the audio signal, audio signal recording, and / or preprocessed audio signal and / or audio signal recording of step 1205 and / or 1210 with an annotation or evaluation (e.g., proper technique, improper technique, etc.) of same (step 1315).[000204] Optionally, output from the breathing assessment model (step 1310) and / or a result of evaluation of step 1315 may be verified and / or validated (step1320) and / or a confidence level for the output and / or evaluation may be determined. When the output and / or evaluation is not valid, an error message may be sent to the subject (step 1325). In some embodiments, execution of step 1320 may include determining how the output and / or evaluation is invalid and / or how a subject may improve the validity and, in these embodiments, the error message of step 1325 may include these determinations and / or recommendations for improving breathing patterns and / or execution of breathing techniques. Additionally, or alternatively, execution of step 1320 may may include determination of a confidence metric, which may quantify the breathing assessment model’s certainty in its classification and / or evaluation decision. On some occasions (e.g., when breathing assessment model is a neural network model), this confidence metric may indicate the probability that the respiratory device usage matches a correct or incorrect pattern so that, for example, the system may be able to distinguish between high-confidence misuses that warrant medical and / or clinical intervention and low-confidence cases that may require reanalysis or additional data.[000205] In step 1330, an indication of the output of step 1310 and / or an evaluation of the output assessment the subject’s respiratory device utilization (e.g., evaluation of step 1315) may be provided to the subject and / or a display device. In some cases, execution of step 1330 includes providing the indication to a mobile application interface (e.g., software application 365) running on a user device (e.g., user device 360) and optionally uploaded to a third party device (e.g., third party user A-N 375A-375N) such as a clinician-facing dashboard. Execution of step 1330 enables subjects to, for example, receive feedback and coaching on, for example, proper execution of breathing technique, while allowing healthcare providers to monitor adherence trends, detect usage errors, and intervene when necessary to improve therapeutic outcomes. For example, when the subject’s technique is deemed correct with high confidence, the indication may be a confirmation message, such as “Good job! Proper technique executed.” If the subject’s technique is deemed incorrect, the indication may provide targeted guidance, such as “Try to inhale more deeply” or “Remember to hold your breath for two seconds before exhaling.” Indications that include messages of this type may be also include other media (e.g., icons, images, animations, etc.) and / or links (e.g., hyperlinks) by which to get moreinformation. In some cases, execution of step 1330 and / or method 1300 may resemble execution of step 1030 and / or method 1000, respectively.

Claims

CLAIMSWe claim:

1. A system for assessing respiratory device usage, the system comprising:a microphone configured to capture an audio signal when a subject is using a respiratory device;a transceiver configured to communicate the audio signal to a processor; and the processor configured to:receive the audio signal;detect one or more acoustic events within the audio signal;identify an event type for each of the one or more acoustic events; and assess the subject's respiratory device usage technique based on the identified event types.

2. The system of claim 1, wherein the microphone is an electret condenser microphone or a MEMS microphone.

3. The system of any one of claims 1-2, wherein the transceiver is a Bluetooth Low Energy (BLE) transceiver.

4. The system of any one of claims 1-3, further comprising an analog-to-digital converter configured to digitize the audio signal at a sampling rate of 44.1 kHz or 48 kHz with 16-bit resolution.

5. The system of any one of claims 1-4, wherein the respiratory device is selected from the group consisting of a metered dose inhaler (MDI), a dry powder inhaler (DPI), a soft mist inhaler (SMI), a nebulizer, a positive expiratory pressure (PEP) device, an oscillating positive expiratory pressure (oPEP) device, a peak flow meter, an incentive spirometer, and a respiratory muscle trainer.

6. The system of any one of claims 1-5, further comprising:preprocessing the audio signal by applying bandpass filtering the audio signal to isolate frequency ranges associated with respiratory device use events.

7. The system of any one of claims 1-6, further comprising:preprocessing the audio signal by applying adaptive noise reduction to the audio signal using spectral subtraction.

8. The system of any one of claims 1-7, wherein detecting the one or more acoustic events comprises applying amplitude thresholding, temporal smoothing, and slope analysis to the preprocessed audio signal.

9. The system of any one of claims 1-8, wherein the event type is selected from the group consisting of shaking the respiratory device, removing a cap from a mouthpiece, priming the respiratory device, actuating the respiratory device, subject inhalation, subject exhalation, breath holding, and replacing a cap on the mouthpiece.

10. The system of any one of claims 1-9, wherein the processor is further configured to segment the audio signal into a plurality of event windows based on the detected acoustic events and extract features from each event window, wherein the features comprise one or more of amplitude, signal energy, frequency content, and event duration.

11. The system of any one of claims 1-10, wherein assessing the subject's respiratory device usage technique comprises:determining whether the subject executed events in a proper sequence and for a proper duration of time according to clinically validated procedures.

12. The system of any one of claims 1-11, wherein the processor is further configured to provide feedback to the subject regarding the assessed respiratory device usage technique via a software application running on the external device.

13. The system of claim 12, wherein the feedback comprises a confirmation message when the subject's technique is correct or targeted guidance when the subject's technique is incorrect.

14. The system of any one of claims 1-13, wherein the processor is configured to input the audio signal into a respiratory device use assessment model trained on labeled audio recordings of respiratory device usage to classify the subject's usage technique as correct or incorrect.

15. The system of claim 14, wherein the respiratory device use assessment model is further configured to generate a confidence score indicating a probability that the subject's usage technique matches a correct or incorrect pattern.

16. The system of any one of claims 14-15, wherein the respiratory device use assessment model is configured to identify specific types of usage errors comprising one or more of insufficient inhalation force, improper timing of actuation, short or incomplete inhalation, and skipping a priming step.

17. The system of any one of claims 1-16, further comprising a cloud computing platform configured to receive the audio signal from the external device and perform signal processing and machine learning classification thereon.

18. The system of any one of claims 1-17, further comprising an accelerometer configured to sense an orientation of the respiratory device.

19. A method for assessing respiratory device usage, the method comprising:receiving, by a processor, an audio signal captured by a microphone when a subject is using a respiratory device;detecting, by the processor, one or more acoustic events within the audio signal;identifying, by the processor, an event type for each of the one or more acoustic events based on acoustic signatures associated with the event types; and assessing, by the processor, the subject's respiratory device usage technique based on the identified event types.

20. The method of claim 19, further comprising preprocessing the audio signal by bandpass filtering the audio signal using digital Infinite Impulse Response (IIR) orFinite Impulse Response (FIR) filters and applying spectral subtraction to subtract an estimate of a noise spectrum from active segments of the audio signal.

21. The method of any one of claims 19-20, wherein detecting the one or more acoustic events comprises:analyzing the audio signal to detect sound amplitude peaks; and segmenting the audio signal into a plurality of event windows based on the detected amplitude peaks.

22. The method of claim 21, further comprising extracting features from each event window, wherein the features comprise one or more of time-domain features, frequency-domain features, and cepstral features.

23. The method of claim 22, wherein extracting frequency-domain features comprises applying a Fast Fourier Transform (FFT) to each event window to obtain a spectral representation thereof.

24. The method of any one of claims 19-23, wherein identifying the event type comprises comparing acoustic characteristics of each detected acoustic event to a library of audio signatures that correlates sound signatures with event types.

25. The method of any one of claims 19-24, wherein assessing the subject's respiratory device usage technique comprises:inputting the audio signal into a respiratory device use assessment model; andreceiving, from the respiratory device use assessment model, a classification of the subject's usage technique as correct or incorrect.

26. The method of claim 25, further comprising receiving, from the respiratory device use assessment model, an identification of a specific type of usage error when the classification indicates incorrect usage.

27. The method of any one of claims 19-26, further comprising providing feedback to the subject regarding the assessed respiratory device usage technique via a user interface.

28. The method of claim 27, wherein the feedback comprises targeted guidance for correcting the subject's technique when the assessed technique is incorrect.

29. The method of any one of claims 19-28, further comprising triggering an alert to a healthcare provider when repeated incorrect usage is detected over a configurable time window.

30. A method for training a respiratory device use assessment model, the method comprising:receiving a labeled set of audio signals of respiratory device usage, wherein labels for the audio signals indicate whether respiratory device usage technique was performed correctly or incorrectly;dividing the labeled set of audio signals into a training set and a testing set; receiving information regarding proper and improper respiratory device usage technique, wherein the information comprises a correct sequence and timing of events for respiratory device usage;training a first version of the respiratory device use assessment model using the training set, wherein the respiratory device use assessment model is configured to evaluate an audio signal to determine if a subject is properly using a respiratory device;testing the first version of the respiratory device use assessment model using the testing set; anditerating upon the first version of the respiratory device use assessment model until a desired level of accuracy is achieved, thereby generating a second version of the respiratory device use assessment model.

31. The method of claim 30, wherein the labels for the audio signals further indicate one or more of event type, event start time, event stop time, event duration, and an indication of whether a subject performed a task associated with an event properly.

32. The method of any one of claims 30-31, wherein the respiratory device use assessment model comprises a deep neural network (DNN), a convolutional neural network (CNN), a recurrent neural network (RNN), or a long short-term memory network (LSTM).

33. The method of any one of claims 30-32, wherein the respiratory device use assessment model is configured to generate a binary classification output indicating correct or incorrect usage.

34. The method of any one of claims 30-33, wherein the respiratory device use assessment model is configured to perform multi-class classification to detect specific types of usage errors.

35. The method of any one of claims 30-34, further comprising applying data augmentation techniques to the labeled set of audio signals, wherein the data augmentation techniques comprise pitch shifting, noise injection, and time-stretching.

36. A method for training a respiratory event characterization model, the method comprising:receiving a labeled data set of audio signals of respiratory device use and / or breathing exercises, wherein labels for the audio signals indicate event types and event timing;dividing the labeled data set of audio signals into a training set and a testing set;training a first version of the respiratory event characterization model using the training set, wherein the respiratory event characterization model is configured to identify and characterize events within an audio signal;testing the first version of the respiratory event characterization model using the testing set; anditerating upon the first version of the respiratory event characterization model until a desired level of accuracy is achieved, thereby generating a second version of the respiratory event characterization model.

37. The method of claim 36, wherein the events comprise one or more of shaking a respiratory device, removing a cap from a mouthpiece, priming the respiratory device, actuating the respiratory device, subject inhalation, subject exhalation, breath holding, and replacing a cap on the mouthpiece.

38. The method of any one of claims 36-37, wherein the respiratory event characterization model is configured to determine an audio signature for each event type based on acoustic characteristics of the event.

39. The method of any one of claims 36-38, further comprising:extracting features from the audio signals of the training set, wherein the features comprise one or more of event duration, amplitude, power, frequency, and frequency bands; andgenerating a library of audio signatures that correlates sound signatures for identified events with annotations corresponding to the identified events.

40. The method of any one of claims 36-39, wherein the labels for the audio signals further indicate one or more of subject characteristics, respiratory device type, and environmental conditions proximate to a subject.

41. The method of any one of claims 36-40, wherein the respiratory event characterization model comprises a deep neural network (DNN), a convolutional neural network (CNN), a recurrent neural network (RNN), or a long short-term memory network (LSTM).

42. A respiratory therapy system comprising:a respiratory device comprising a mouthpiece;a sensor system attached to or integrated with the respiratory device, the sensor system comprising:a microphone configured to capture an audio signal when a subject is using the respiratory device;an analog-to-digital converter configured to digitize the audio signal; anda transceiver configured to wirelessly communicate the digitized audio signal to a user device;a user device comprising a software application configured to: receive the digitized audio signal from the transceiver;store the digitized audio signal; andcommunicate the digitized audio signal to a processor; and the processor configured to:input the audio signal into a respiratory device use assessment model; receive an output from the respiratory device use assessment model assessing the subject's respiratory device usage technique; and communicate the output to the user device for display to the subject.

43. The respiratory therapy system of claim 42, wherein the software application is configured to associate metadata with the digitized audio signal, wherein the metadata comprises one or more of timestamps, device ID, session markers, and signal quality indicators.

44. The respiratory therapy system of any one of claims 42-43, wherein the processor comprises or communicatively coupled to an audio signal preprocessing module configured to perform noise suppression, peak detection, segmentation, and feature extraction.

45. The respiratory therapy system of any one of claims 42-44, wherein the output comprises a score indicating how closely the subject's actions align with clinically validated respiratory device usage procedures.

46. The respiratory therapy system of any one of claims 42-45, wherein the processor is further configured to trigger an alert to a healthcare provider when repeated improper usage is detected.

47. The respiratory therapy system of any one of claims 42-46, wherein the respiratory device is selected from the group consisting of a metered dose inhaler (MDI), a dry powder inhaler (DPI), a soft mist inhaler (SMI), a nebulizer, a positive expiratory pressure (PEP) device, an oscillating positive expiratory pressure (oPEP) device, a peak flow meter, an incentive spirometer, and a respiratory muscle trainer.

48. The respiratory therapy system of any one of claims 42-47, wherein the sensor system further comprises an accelerometer configured to sense an orientation of the respiratory device.

49. The respiratory therapy system of any one of claims 42-48, wherein the transceiver is a Bluetooth Low Energy (BLE) transceiver configured to communicate with the user device using GATT protocols.

50. The respiratory therapy system of any one of claims 42-49, wherein the software application is configured to store the digitized audio signal in an uncompressed linear Pulse Code Modulation (PCM).wav format.

51. The respiratory therapy system of any one of claims 42-50, wherein the respiratory device use assessment model is configured to identify specific types of usage errors comprising one or more of insufficient inhalation force, improper timing of actuation, short or incomplete inhalation, and skipping a priming step.

52. The respiratory therapy system of any one of claims 42-51, wherein the output comprises targeted guidance for correcting the subject's technique when the assessed technique is incorrect.

53. A method for training a breathing assessment model, the method comprising: receiving information regarding breathing techniques, wherein the information comprises a correct sequence and timing of events for performance of a breathing technique;receiving a labeled data set of audio signals of breathing exercise execution, wherein labels for the audio signals indicate whether breathing technique was performed correctly or incorrectly;generating an augmented labeled data set using the information regarding breathing techniques;dividing the augmented labeled data set into a training set and a testing set; training a first version of the breathing assessment model using the training set, wherein the breathing assessment model is configured to evaluate an audio signal to assess a subject's breathing patterns and / or breathing technique execution;testing the first version of the breathing assessment model using the testing set; anditerating upon the first version of the breathing assessment model until a desired level of accuracy is achieved, thereby generating a second version of the breathing assessment model.

54. The method of claim 53, wherein the breathing techniques comprise one or more of box breathing, diaphragmatic breathing, 4-7-8 breathing, pursed-lip breathing, and Lamaze breathing.

55. The method of any one of claims 53-54, wherein the breathing assessment model is configured to generate a binary classification output indicating correct or incorrect breathing technique execution.

56. The method of any one of claims 53-55, wherein the breathing assessment model is configured to perform multi-class classification to detect specific types of breathing patterns and / or technique execution errors.

57. The method of any one of claims 53-56, wherein the breathing assessment model is configured to identify when and for how long a subject inhales, exhales, and / or holds his or her breath.

58. The method of any one of claims 53-57, wherein the breathing assessment model comprises a deep neural network (DNN), a convolutional neural network(CNN), a recurrent neural network (RNN), or a long short-term memory network (LSTM).

59. A method for assessing a subject's breathing, the method comprising:receiving, by a processor, an audio signal captured by a microphone when a subject is breathing and / or performing a breathing exercise;optionally preprocessing the audio signal to generate a preprocessed audio signal;inputting the audio signal or the preprocessed audio signal into a respiratory event characterization model and / or a breathing assessment model;receiving an output from the respiratory event characterization model and / or the breathing assessment model, wherein the output identifies and / or characterizes one or more breathing events within the audio signal; andassessing the subject's breathing patterns and / or breathing technique execution based on the output.

60. The method of claim 59, wherein the one or more breathing events comprise one or more of subject inhalation, subject exhalation; and breath holding.

61. The method of any one of claims 59-60, further comprising validating the output to determine a confidence level for the output.

62. The method of claim 61, further comprising communicating an error message to a display device when the output is not valid.

63. The method of any one of claims 59-62, wherein assessing the subject's breathing patterns and / or breathing technique execution comprises determining whether the subject executed events in a proper sequence and for a proper duration of time according to a prescribed breathing technique.

64. The method of any one of claims 59-63, further comprising providing feedback to the subject regarding the assessed breathing patterns and / or breathing technique execution via a user interface.

65. The method of claim 64, wherein the feedback comprises targeted guidance for correcting the subject's breathing technique when the assessed technique is incorrect.

66. The method of any one of claims 59-65, further comprising comparing results of the method performed at different times to assess the subject's respiratory health over time.

Images