Pose detection system for personal head wearable device

Abstract

This document discloses methods and systems related to the field of posture detection. A system for a personal head-worn wearable device includes a first electrode and a second electrode. The first and second electrodes measure bioelectrical signals. The system also includes one or more non-transitory computer-readable media storing instructions that, when executed by the system, cause the system to analyze the bioelectrical signals using a stored feature model of the posture signals to identify the posture signals within the bioelectrical signals, and to generate interface signals in the process of identifying the posture signals. The posture signals are one of a double-jaw clenching signal, a triple-jaw clenching signal, and a long-jaw clenching signal.

Description

[0001] This application is a divisional application of Chinese National Application No. 2021800323913, filed on May 19, 2021, entitled "Posture Detection System for Personal Head Wearable Device". Cross-reference to related applications

[0002] This application claims the benefit of French patent application No. 2005020, filed on May 19, 2020, which is incorporated herein by reference in its entirety. Background Technology

[0003] Wearable devices are electronic devices configured to be worn by a user and perform various functions. Wearable devices can be configured to use electrodes to detect electrical activity in the user's brain. These electrodes can capture different types of electrical signals, such as electroencephalography (EEG or brain signals), electrocardiography (ECG), electromyography (EMG or muscle signals), and electrooculography (EOG or eye movement signals).

[0004] While electrodes can provide reliable measurements of bioelectrical signal patterns associated with EEG, ECG, EOG, and EMG readings, these signals are difficult to use as control signals. For example, wearable neural interfaces configured to capture muscle movement may confuse spontaneous muscle movements (control signals) during exercise with unintentional or involuntary muscle movements (erroneous signals or noise). This limits the practical application of neural interfaces when the user is at rest, resulting in limited practical utility. Summary of the Invention

[0005] This paper discloses methods and systems related to the field of posture detection. These systems and methods can be used in personal wearable devices, such as personal head-worn devices. Examples of head-worn devices include smart glasses, headphones (generally walkie-talkie headsets, headsets or earbuds, wireless earbuds, VR / AR headsets, earphones, earplugs), etc. Wearable devices can be configured to detect electrical activity in a user's brain using electrodes connected to the wearable device, and configured to make contact with a portion of the user's head when the wearable device is worn. For example, electrodes placed in earbuds, earpieces, or near the ear (especially the ear canal or concha) (also known as intraauricular electrodes) can provide exceptional contact. These electrodes can be installed in audio earbud devices or headphones to capture different types of electrical signals, such as the EEG, ECG, EMG, and EOG signals mentioned above.

[0006] The systems disclosed herein can relate to using a neural interface of a head-mounted wearable device to perform various hands-free control functions. For example, the system can be trained to capture clear control signals via the neural interface. The neural interface can be configured to control the operation of a personal wearable device and / or control another device associated with the personal wearable device via those control signals.

[0007] To provide clear control signals independent of user movement, bioelectrical signals associated with facial muscle movements, such as jaw movements, can be used as control signals. However, users may unconsciously trigger control signals while speaking, eating, drinking, or chewing. Specific embodiments of the invention disclosed herein aim to provide a neural interface for wearable devices that can provide clear signals to control the wearable device or other devices in all possible situations. In this regard, the neural interface can be configured to activate in response to the detection of jaw postures involving voluntary muscle contractions (e.g., jaw-controlled postures), including a wearer's double jaw clenching, triple jaw clenching, or long jaw clenching, because these jaw postures have very distinctive characteristics compared to single jaw clenching and other typical jaw movements listed above.

[0008] In a specific embodiment of the present invention, a posture detection system for a personal head-worn wearable device is provided. The system includes a first electrode and a second electrode, wherein the first electrode and the second electrode measure bioelectrical signals. The system also includes one or more computer-readable media storing instructions that, when executed by the system, cause the system to analyze the bioelectrical signals to identify the posture signal in the bioelectrical signals using a stored signature model of the posture signals, and to generate an interface signal when identifying the posture signal in the bioelectrical signals. The posture signal is one of a double-jaw clenching signal, a triple-jaw clenching signal, and a long-jaw clenching signal.

[0009] In a specific embodiment of the present invention, a wearable posture recognition system is provided. The system includes a first earpiece and a second earpiece. The system also includes a first electrode and a second electrode, wherein the first electrode and the second electrode are located on the outer surface of the first earpiece, and wherein the first electrode and the second electrode measure bioelectrical signals. The system further includes a third electrode and a fourth electrode, wherein the third electrode and the fourth electrode are located on the outer surface of the second earpiece, and wherein the third electrode and the fourth electrode measure the bioelectrical signals. The system also includes one or more computer-readable media storing instructions, which, when executed by the system, cause the system to analyze the bioelectrical signals to identify posture signals within the bioelectrical signals, and to generate interface signals after identifying the posture signals within the bioelectrical signals. The identification of the posture signals within the bioelectrical signals uses a combination of data measured by the first electrode and the second electrode and data measured by the third electrode and the fourth electrode.

[0010] In a specific embodiment of the present invention, a posture recognition system is provided. The system includes a first electrode and a second electrode, wherein the first electrode and the second electrode measure bioelectrical signals. The system also includes a user interface output and a computer-readable medium storing one or more instructions. When the instructions are executed by the system, the system generates a prompt to perform a posture associated with the posture signal, analyzes the bioelectrical signal to locate the posture signal in the bioelectrical signal using stored features of the posture signal, generates an interface signal when identifying the posture signal in the bioelectrical signal, and updates the stored features using the posture signal to generate modified stored features. The interface signal is output to the user interface output. The stored features are default features associated with the posture.

[0011] In a specific embodiment of the present invention, a posture detection method for a personal head-worn device is provided. The method includes measuring bioelectrical signals using a first electrode and a second electrode, wherein the first electrode and the second electrode are located on the personal head-worn device. The method further includes analyzing the bioelectrical signals to identify the posture signal within the bioelectrical signals using a stored feature model of the posture signal. The method further includes generating an interface signal when identifying the posture signal within the bioelectrical signals. The posture signal is one of a double-jaw clenching signal, a triple-jaw clenching signal, and a long-jaw clenching signal. Attached Figure Description

[0012] Figure 1This includes examples of pose detection systems and flowcharts of a set of methods according to specific embodiments of the invention disclosed herein.

[0013] Figure 2 A diagram including the human masseter muscle and auricle.

[0014] Figure 3 Examples of screens provided on a display during the training process of a system, including specific embodiments of the invention disclosed herein.

[0015] Figure 4 Examples of cross-ear and mono-ear configurations are included, based on specific embodiments of the invention disclosed herein.

[0016] Figure 5 Examples of bimanual bite signals measured by electrodes and a sliding window, including specific embodiments of the invention disclosed herein.

[0017] Figure 6 Examples of signals measured by electrodes at different times and durations, including specific embodiments of the invention disclosed herein.

[0018] Figure 7 Examples of earpieces containing electrodes at different locations, including specific embodiments of the invention disclosed herein.

[0019] Figure 8 Examples of connector configurations, including specific embodiments of the invention disclosed herein. Detailed Implementation

[0020] This document discloses in detail methods and systems related to the field of posture detection based on the above-described invention. The methods and systems disclosed in this section are non-limiting embodiments of the invention, are for illustrative purposes only, and are not intended to limit the full scope of the invention.

[0021] Specific embodiments of the present invention relate to a posture detection system and method. The system may include a neural interface (also known as a brain-computer interface or BCI) of a wearable device (such as a head-mounted wearable device) to perform various hands-free control functions. In particular, the technology relates to capturing clear control signals via the neural interface. The neural interface may be configured to control the operation of the wearable device and / or control other devices, such as controlling a smartphone with earpieces.

[0022] A system according to a specific embodiment of the present invention may include one or more computer-readable media storing instructions that, when executed by the system, cause the system to perform certain actions or method steps. The computer-readable medium may be non-transitory. The computer-readable medium may be internal to the system or external to the system. In this disclosure, actions and / or method steps are described as actions / steps that the system is "configured" to perform; in a sense, the system's structure is configured to perform these actions / steps (e.g., a processing block of the system may execute instructions to cause the system to operate in a certain way, perform certain actions, and / or provide certain outputs).

[0023] Figure 1 This includes examples of a posture detection system 100 and a process flow 150 of a set of methods, according to specific embodiments of the invention disclosed herein. The posture detection system can be used in wearable devices, such as personal wearable devices. Personal wearable devices used in this disclosure are electronic devices that perform multimedia activities. Specific embodiments of the invention disclosed herein relate to wearable devices in the form of headphones, such as earbuds. However, the system of the invention can be used in any type of device, including any type of wearable device such as headbands, wristbands, smartwatches, smart rings, belts, etc.

[0024] A system according to a specific embodiment of the invention may include electrodes, such as electrodes 102 and 103. The electrodes may be placed in contact with the user's skin, for example, when the user wears a personal head-worn device containing the electrodes. The electrodes may measure bioelectrical signals, as shown in step 151 of process 150. As used herein, a bioelectrical signal is a physically detectable signal generated by the nervous system or internal brain operations of a user (e.g., a user wearing a wearable device with the electrodes described herein). The term is also used to refer to the same signal that is electrically sampled, measured, and analyzed by a posture detection system. The bioelectrical signal may be transmitted along a signal processing path, from an analog electrical signal emitted by the wearer's brain to digitization and storage in a computer-readable medium, and undergo additional processing, such as the application of digital filtering and other digital post-processing. The term may refer to a continuous phenomenon or its discrete sampling, the relevant definition of which is self-evident from the surrounding environment.

[0025] Muscle contraction can be triggered by the brain, which transmits electrical signals along motor neurons to the muscles. When muscle fibers receive the trigger signal transmitted along the motor neurons and contract, they also generate electrical activity. The electrical activity from the contracting muscles and motor neurons triggers the displacement of surrounding charged particles (ions), which in turn causes the displacement of surrounding ions, creating a cascading effect. This mechanism allows the electrical activity generated by muscle contraction and motor neurons to travel through the fluids filling the body, all the way to the skin.

[0026] Electrodes made of conductive materials (such as metals) can come into contact with the skin and capture changes in the skin's surface potential. These electrodes can be called surface electrodes. This technique is called EMG. This recording technique can use at least two different electrodes (at least one "measurement" electrode and at least one "reference" electrode). Multiple measurement electrodes can be placed at different locations on the skin.

[0027] Figure 2 A schematic diagram 200 includes the human masseter muscle 201. Masseter muscle contraction (or “jaw clenching,” as used herein) is a voluntary action of the subject. For the purposes of this invention, it can be considered that the subject’s mouth is closed, the upper and lower teeth are in contact with each other, and then the subject contracts the masseter muscle. Since the subject’s jaw is already “closed” and the upper and lower teeth are in contact, there is no obvious movement of the jaw itself except for changes in the shape, size, and hardness of the masseter muscle during contraction.

[0028] In a specific embodiment of the invention, the muscle activity of the masseter muscle can be measured and used via a neural interface in a personal head-worn device to perform certain functions. As previously described, this muscle activity can be recorded using an EMG system. In a specific embodiment of the invention, at least one electrode, such as electrodes 102 and 103, can be placed inside or near the subject's ear, and thus near the target electrical signal source area, which maximizes signal strength, as signal strength can decrease with distance from the signal source.

[0029] Figure 2 It also includes a schematic diagram of human auricular anatomy 250. To measure bioelectrical signals, electrodes can be placed anywhere inside or near the ear and in contact with the subject's skin. Examples include the ear canal 251, inferior concha 252, superior concha 253, earlobe 254, tragus 255, etc., or around the ear 260. Electrodes can also be located on the surface of wearable devices (such as earplugs) so that they are in contact with the skin when the subject wears the device.

[0030] Reference Figure 1System 100 may include processing blocks, such as processing block 104. The processing block may access memory 105 located internally or externally to processing block 104. The memory may store instructions for execution by the processing unit, such as instructions for performing method steps, like step 152 for analyzing signals and step 153 for generating interface signals, as will be described in more detail herein. The processing block may include any type of processing unit, such as a microcontroller, microprocessor, ASIC, FPGA, etc. The processing block may be implemented as a CPU, GPU, TPU, FPU, etc., and may use any memory component, such as flash memory, RAM, SRAM, ROM, etc. The processing block may be implemented by one or more processors operating in parallel.

[0031] Electrodes can be connected to other hardware units (not shown in the diagram) using wires or any type of electrical connection system, such as a hardware analog front-end (AFE). The AFE can be embedded within the device where the electrode operates or externally (e.g., on an external circuit board). The AFE may contain analog differential amplifiers, analog filters, and analog amplifiers to add gain to the system and / or other hardware modules for further processing of the electrode measurement signal. With only one measuring electrode and one reference electrode, the AFE's input is the differential signal between the measuring and reference electrodes. With multiple measuring electrodes, the AFE can have multiple "channels," each consisting of a differential signal between one measuring electrode and one reference electrode.

[0032] In addition to the aforementioned "measurement" and "reference" electrodes, the system may include other electrodes. For example, the system may include a "bias" or "drive right leg" electrode for canceling common-mode measurements across different channels. The system may also include a "ground" electrode for establishing a common ground between the electronic device and the main body.

[0033] The system may include an analog-to-digital converter (ADC). The AFE can amplify the electrical signal recorded for each channel to maximize the voltage range of the electrical signal, for example, within the limit of the maximum voltage allowed at the ADC input. For example, if the ADC allows input signals between 0V and 2V, and the signal recorded at the AFE input is between -0.1mV and 0.1mV, then an ideal AFE would apply a gain of 10,000 and a DC component of 1V.

[0034] An ADC can discretize a signal at a given sampling frequency. For example, at a sampling frequency of 250Hz, the discrete values ​​of the signal are measured every 4ms. An ADC can also digitize a signal based on its resolution. For example, if the ADC input range is 2V and the resolution is 10 bits, each voltage measurement can be rounded to 1.953 mV (2V / 2...). 10The closest multiple of (1.953mV).

[0035] Then, the discretized and digitized signals can be transmitted to the system's processing blocks, for example... Figure 1 The processing block 104. The processing block can perform certain operations, such as signal processing operations, prediction operations, data transfer operations, etc., as will be described in more detail in this disclosure. The processing block can perform some of the operations described herein, for example, by executing instructions stored in memory.

[0036] The processing block can perform signal processing operations, such as filtering (high-pass, low-pass, band-pass, notch, or any combination thereof), slicing, padding, artifact removal using advanced techniques (such as ICA, Riemann geometry), channel reconstruction, channel rereference, etc.

[0037] The processing block can perform predictive operations, such as classifying labels to identify certain muscle contractions, recognizing abnormal signals, etc. It can be based on rule-based models or pre-trained supervised machine learning models (such as neural networks, support vector classifiers, logistic regression, decision trees, any ensemble method, or linear classifiers), unsupervised methods such as clustering (such as mixture models, k-means and variable methods), regression (such as linear regression, Bayesian models), reinforcement learning (such as Q-learning, Monte Carlo methods), etc.

[0038] The processing block can also perform data transfer operations, including, for example, transferring recorded data to an electronic device and / or outputting any signal processing and / or prediction. The electronic device can be a personal user device, such as… Figure 1 The system includes a personal user device 120, such as a smartphone or personal computer. The electronic device can also be a remote server with which the system communicates to perform certain operations, such as further processing. The system can communicate with the electronic device wirelessly (e.g., via Bluetooth, WiFi, etc.) or via a wired connection. Additional processing and display can then be performed on the electronic device associated with the system.

[0039] In a specific embodiment of the invention, as shown in step 152, the system is configured to analyze the bioelectrical signals measured by the electrodes. The bioelectrical signals can be analyzed to identify postural signals within them. Postural signals can be signal representations of postures performed by the wearer, included in the bioelectrical signals measured by the electrodes. Postural signals can be signals associated with postures involving masseter muscle contraction, such as the aforementioned posture involving jaw clenching.

[0040] In specific embodiments of the present invention, the system can identify specific posture signals of a specific posture from bioelectrical signals measured by electrodes. One posture can be a "double jaw clench," where the subject clenches their jaw twice. There may be a delay between each clench, which can be a predetermined delay or a threshold-limited delay. For example, the delay can be set to less than 1 second and can be optimized as needed, for example, to 0.8 seconds. Another posture can be a "triple jaw clench," where the subject clenches their jaw three times. There may be a delay between each clench, which can be a predetermined delay or a threshold-limited delay. For example, the delay can be set to less than 1 second and can be optimized as needed, for example, to 0.8 seconds. Yet another posture can be a "long jaw clench," where the subject clenches their jaw and maintains masseter muscle contraction for a predetermined period of time, for example, at least 1 second. Therefore, in specific embodiments of the present invention, the posture signal to be identified from the bioelectrical signals can be one of a double jaw clench signal, a triple jaw clench signal, or a long jaw clench signal.

[0041] The reason for choosing the above postures instead of "single jaw clenching" (e.g., the subject clenching his jaw once) is that these postures can generate a posture signal whose features are more easily identifiable from common postures performed by the subject (e.g., chewing, swallowing, speaking, etc.).

[0042] In a specific embodiment of the present invention, the system is configured to generate an interface signal when recognizing a posture signal in a bioelectrical signal, as shown in step 153 of process 150. The interface signal may be a feedback signal to notify the user that the posture has been recognized, a control signal from a wearable device or related device, or a training signal to train the posture recognition model of the posture recognition system.

[0043] Interface signals can be feedback signals that help the user recognize gesture signals. Feedback can be auditory, visual, or tactile, such as beeps or audible information, information on a display, vibration, etc. The user interface output can be located within the personal head-mounted wearable device itself. For example, the user interface output can be speaker 106 of system 100. Alternatively or in combination, the user interface output can be a controllable vibration source or a display on the wearable device. Alternatively or in combination, the user interface output can be located on a device that operates with the personal head-mounted wearable device, such as personal user device 120. For example, the interface signal can be a representation of a gesture signal displayed on display 128 or an auditory message delivered via speaker 126.

[0044] Interface signals can be not only explicit feedback that a gesture has been recognized, but also control signals for the system and / or other devices associated with the system. For example, based on the gesture performed by the user identified from the gesture signal, the control signal can be used to perform certain actions or trigger certain events on devices associated with the system. Thus, interface signals can be used to control play / pause functions, start / end call functions, etc. Therefore, the system according to a specific embodiment of the present invention can be used to control devices, including hands-free control.

[0045] Electronic devices can then be controlled by performing gestures and recognizing gesture signals in bioelectrical signals, for example... Figure 1 The personal user device 120 described herein can perform a variety of control functions, such as playing / pausing music, making calls using a True Wireless headset or a walkie-talkie with an on / off button, grabbing objects using VR headsets, triggering noise cancellation on sleep buds, triggering an ASSR test on a hearing aid, etc.

[0046] Postural signals within bioelectrical signals can be identified using a stored feature model of the posture signals. In specific embodiments of the invention, the stored feature model can be a default feature model associated with posture. The default feature model can be configured based on unique bioelectrical signal patterns in the time and / or frequency domains specific to posture. Advantageously, the default feature model is based on bioelectrical signal patterns similar to those between individuals. The default feature model can be pre-stored within the system or in an external device (such as a server) that processes system data. The default feature model can be a neural network trained using data from multiple posture signals to predict the postures performed by the user. The default feature model can be a classifier where features of the posture signals are stored, for example, as embeddings of variables in the weights or functions of the neural network.

[0047] The default feature model can be calibrated through the training process, adaptive learning process, domain transfer, or other parametric or non-parametric methods. During training, users can interact with the system and provide data so that the system can "learn" from specific users and improve the default model accordingly. In this way, the default feature model can be adjusted for each user, as users can update their stored feature models, which will be discussed in detail below.

[0048] In a specific embodiment of the invention, the system is configured to generate prompts to perform a pose associated with a pose signal, and to update the stored feature model using the pose signal to generate a revised stored feature model. The revised stored feature model is a stored feature model that has been calibrated using data from a training process performed by the user.

[0049] The training process can be performed when the system is first used so that it can correctly "learn" the poses, such as when the wearable device is used for the first time. If the user wants to adjust the model, an additional training process can be performed (e.g., if the user realizes that performing a pose in a certain way does not trigger the expected results). In this way, the system can be calibrated to improve the performance of the pose detection algorithm and enhance the user experience when performing poses.

[0050] The training module can provide users with a program to learn to perform clear jaw muscle movements, thereby activating the neural interface. The training module can be configured to prompt users to perform specific jaw postures. Users can attempt to perform these jaw postures, and when the neural interface detects a specific jaw posture, it can generate a feedback signal. This feedback signal can be displayed, for example, on a graphical user interface that displays the measurement signal, played back as an auditory feedback signal (e.g., a beeping sound played by a speaker on a wearable device), and / or provided in the form of a tactile signal (e.g., earplug vibration). Therefore, users can optimize their jaw posture to improve the performance of the posture recognition system.

[0051] The training process can be aided by a companion application running on a personal user device (e.g., personal user device 120). The application can be associated with the system so that the system can still input and output data through the application, even if the application runs on a device not part of the system. This allows the user to perform signal recognition while wearing the head-mounted wearable device and view the process on an external device. Figure 3 This includes, for example, screen examples provided by a companion application running on a personal user device 120. The user can install and open the companion application on the personal user device. The personal user device can provide a notification to open the application when a connection to the wearable device is detected. As a first step in the training process, the system (e.g., via the application) can provide information about how the technology behind the system works. This step might include videos and animations showing masseter muscle contractions, images and messages explaining how electrodes are used to record facial movements, and how technologies such as artificial intelligence translate signals into system actions. In subsequent steps, the system can check whether the wearable device is correctly connected to the user's skin and whether the SNR is good enough. For example, the system can provide notifications via the application that the electrodes are correctly positioned and prompt the user to move or clean the device if it is not working.

[0052] The application can also provide explanations about the posture to be performed. Screen 300 includes a screen example from personal user device 120 illustrating the procedure the user might follow. Messages, such as message 301, may be displayed on the screen of the personal user device or provided as audio messages via the personal user device's speaker or the wearable device itself. The application can also provide explanations about feedback to help the user understand what the correct posture is. Screen 310 includes a screen example from personal user device 120 with explanations of the feedback the user might receive.

[0053] The two options can be performed independently or sequentially, and can be repeated multiple times as needed by the user. First, the user begins the test and can be asked to perform a posture every 3-10 seconds, for example, for 1-2 minutes, and receive appropriate feedback. Second, over a period of time, the user can be asked to perform the posture whenever he / she wants, for example, a 1-2 minute attempt, and receive appropriate feedback. Screen 320 includes an example screen from the personal user device 120, which displays a graph of signals 302 measured by electrodes and feedback signals 303 provided by the system when a posture is detected.

[0054] After the system is trained, the model is updated in other ways and / or the features associated with the posture signal are correctly stored. The system can use this stored feature model to search for posture signals in the bioelectrical signals measured by the electrodes by using the stored feature patterns. The system may include specific hardware modules or software routines that perform such recognition. The use of the aforementioned signal processing modules allows the system to process data from the signals measured by the electrodes and provide predictions based on this data, which will be explained in more detail below. Recognition can be aided by artificial intelligence.

[0055] The system may include various devices for providing feedback to a user. For example, the system may include one or more speakers. If the system is embedded in an earbud, the speaker may be the earbud's speaker, such as speaker 106. If the system is associated with another device, such as a personal user device connected to such an earbud, such as device 120, the speaker may be speaker 126 of the mobile device. The system may include a display, such as display 128 of a personal user device, which may also be used to provide feedback to the user and receive data from the user. In embodiments where the wearable device has its own display (e.g., a smartwatch), feedback may be provided on that display. The system may include other hardware modules capable of providing feedback, such as lights, vibration hardware, etc.

[0056] During training, different types of feedback can be provided to assist the user throughout the process. For example, visual feedback can be provided. Visual feedback can be offered using a personal user device's display or other types of visual cues, such as lights. For instance, color codes can be used to indicate the status of the process. In a specific example, gray can be used by default, turning green when a posture is detected. This allows the user to clearly see that the system is recognizing a posture. As another example of visual feedback, the system can be configured to display, for example, a graph of bioelectrical signals measured by electrodes on a personal user device's display. Figure 3 Screen 320 displays an example, showing a graph of signal 302 as the user performs a gesture. The specific image of the bioelectrical signal measured by the electrodes on the screen can help the user detect what happens when the gesture is performed in different ways, such as with more or less force. As another example of visual feedback, the system can display an indication of the intensity of the simulated measured bioelectrical signal, such as an instrument used for this purpose. An example of an instrument for feedback signal 303 is also shown... Figure 3 As shown on screen 320.

[0057] Another type of feedback the system can provide is audio feedback. An example of audio feedback is a beep, pop, or any sound emitted by the system when it detects a gesture. In specific embodiments, the sound can be weak or strong, depending on the intensity of the gesture, similar to sonar.

[0058] Another type of feedback the system can provide is haptic feedback. An example of haptic feedback is a simple vibration when the system detects a gesture. In specific embodiments, the vibration can be weak or strong, depending on the intensity of the gesture.

[0059] As previously mentioned, the electrodes can be located on the outer surface of the wearable device. The wearable device can be an earpiece. The system can include other wearable devices, such as... Figure 1 Wearable devices, such as a second earpiece, are included. The second earpiece may also be equipped with electrodes, such as electrodes 112 and 113, to measure bioelectrical signals in the manner described above with respect to the first earpiece. The electrodes of the second earpiece may also be located on the outer surface of the second earring in the manner described herein with respect to the first earpiece. In these cases, the gesture signal in the bioelectrical signal can be identified using a combination of data measured by the electrodes of the first earpiece and data measured by the electrodes of the second earpiece. The second earpiece (second earpiece) 110 may include components that are the same as or similar to those of the first earpiece of system 100. For example, the second earpiece may include components such as a processing block 114, a memory 115, and a speaker 116.

[0060] In specific embodiments of the invention, one or more electrodes can independently measure bioelectrical signals. Measurements can be performed relative to a reference electrode. In specific embodiments, multiple measurements can be analyzed to determine the system's common-mode signal (e.g., multiple measurements can be summed and averaged). Multiple measurements can be performed relative to a single reference electrode or multiple reference electrodes. In specific embodiments, the system can be designed to generate a signal opposite to the common-mode signal and feed this signal back into the wearer's body to cancel the common-mode signal. Common-mode signals and opposite signals can be continuously measured and generated to improve system performance. The signal used to cancel the common-mode signal can be fed back using an electrode, which may be referred to as the driving right leg electrode.

[0061] In specific embodiments, one or more different elements of the posture recognition system (e.g., two separate earpieces or earpieces) may include different numbers and configurations of electrodes. In embodiments with two electrodes, one electrode may be a reference electrode and the other may be a measuring electrode. In embodiments with more than two electrodes, a third electrode may be another measuring electrode, a right leg drive electrode, or a ground electrode. In specific embodiments of the invention, the system may include two elements, each including three electrodes, for a total of six. In these embodiments, one electrode on each of the two elements may serve as a reference electrode for the other electrodes on that element. In specific embodiments of the invention, two different elements may be connected together or connected via a wireless communication link. For example, two earpieces may each include two electrodes, and the two earpieces may be connected together. In this embodiment, a wire may be used to allow the two devices to function as a single measurement system, with measurements from each device performed together to produce more accurate readings. In a system consisting of two connected earpieces and at least three electrodes, there are several alternative configurations, including: two measuring electrodes in the first ear and one reference electrode in the second ear; one measuring electrode and one reference electrode in the first ear and one measuring electrode in the second ear; one measuring electrode and one driving right leg electrode in the first ear and one reference electrode in the second ear; one measuring electrode and one reference electrode in the first ear and one driving right leg electrode in the second ear; one reference electrode and one driving right leg electrode in the first ear and one measuring electrode in the second ear; one measuring electrode and one ground electrode in the first ear and one reference electrode in the second ear; one measuring electrode and one reference electrode in the first ear and one ground electrode in the second ear; and one reference electrode and one ground electrode in the first ear and one measuring electrode in the second ear.

[0062] Figure 4An example of embodiment 400 is included, which comprises two earpieces 401 and 402 connected by a wire 403. In the embodiment using a wire to connect the two earpieces, the system comprising the two earpads and all the electrodes they contain can be considered a single electrical system, with only one reference electrode used for the entire system. As previously described, each channel can receive a differential signal between a measurement electrode and a reference electrode. All channels can be processed synchronously by the same ADC, and the signal processing and prediction steps can take the signals from all channels as input. By considering all channels simultaneously, signal processing quality (e.g., noise removal) and prediction can be improved.

[0063] Generally, the greater the distance between the measuring electrode and the reference electrode, the better the signal. Therefore, this setup allows for a reference electrode in one ear and one or more measuring electrodes in the other ear, which can produce better signal quality compared to having both measuring and reference electrodes in the same ear. However, for example, when the electrodes are mounted in true wireless earbuds (TWS), there may be no wired connection between the two earbuds. In these embodiments, each earbud can be considered an independent electrical system. Therefore, in a particular embodiment, one or more measuring electrodes and their reference electrode are located in the same ear, which may lead to a degraded signal quality. Figure 4 An example of embodiment 450 is also shown, which includes a single earplug system with electrodes 405 and 406 located on the same ear. When a second earplug is also used in this configuration, the electrodes of each earplug will provide separate measurements of bioelectrical signals.

[0064] In embodiments where multiple earplugs provide independent measurements, various recombination solutions can be implemented to combine data from different earplugs. In a specific embodiment of the invention, predictive recombination can be used. With this solution, the signal processing and prediction steps described above can be run independently on each earplug, and predictions from two earplugs can be combined in various ways. For example, if one earplug predicts a probability of 0.60 for jaw clenching detection, while other earplugs predict a probability of 0.80, these probabilities can be averaged, and it can be assumed that the probability of jaw clenching detection is 0.70. The advantage of this solution is that, since each earplug is an independent system, the subject can enjoy the system's functionality by wearing only one earplug.

[0065] In specific embodiments of the invention, feature recombination can be used. This solution involves independently computing these features on each earbud when using a machine learning model based on pre-computed features (e.g., support vector classifier, logistic regression, etc.), and then combining them. These features can be combined by averaging the values ​​of a given feature across the two earbuds. For example, if feature A has a value of 3.0 for one earbud and a value of 5.0 for the other, then feature A with a value of 4.0 can be passed to the model for prediction. Alternatively, features can be combined by passing exactly the same features to the model. For example, features A1 and A2 obtained from individual earbuds can be passed to the model, in which case the model can be trained to process twice the number of features.

[0066] In a specific embodiment of the invention, signal reconstruction can be used. This solution involves recombining the signals from the two earbuds before the signal processing step, or after signal processing and before the prediction step. In a specific embodiment of the invention, the ADCs of the two earbuds may be out of sync (i.e., the discrete values ​​measured by each ADC may not be recorded at exactly the same time, and this delay may vary depending on the intended use of the devices). In this case, some preprocessing steps, such as signal shifting and resampling, can be performed to maximize the correlation between the signal windows observed from each earbud.

[0067] The reconstruction steps described above can be performed via embedded software in the earbuds or external electronic devices (e.g., personal user devices such as smartphones or laptops). If reconstruction is performed on the earbuds, and both earbuds are worn together, one earbud can act as the "master earbud," collecting predictions / features / signals from the other earbuds to perform reconstruction. Regardless of whether reconstruction is performed on the earbuds or on an external device, the system is able to determine whether both earbuds are being used and adjust accordingly.

[0068] In a specific embodiment of the invention, the system is configured to determine whether a bioelectrical signal is present in data from a first set of electrodes (e.g., from a first earplug) and whether it is present in data from a second set of electrodes (e.g., from a second earplug). If the bioelectrical signal is present in both the first and second data sets, the recognition of the posture signal within the bioelectrical signal can be combined with a stored feature model; if the bioelectrical signal is present in only one of the first and second data sets, only one of the first and second data sets can be used, along with the stored feature model. Thus, if one earplug is not used or loses contact with some electrodes, the system can still perform its intended function using the remaining data.

[0069] In specific embodiments of the invention, it is advantageous to collect multiple sample signals on either side of the wireless link (e.g., two wireless earbuds connected via Bluetooth) to avoid sending raw data from one ear to the other via the communication connection. In specific embodiments of the invention, instead of sending raw data, local processing can be performed on each side and combined for better signal detection prediction. Predictions from the "processed signals" can be combined, for example, through a voting system or averaging. Data from both sides can be combined for a single prediction by providing a probability (not 0 or 1) for each side. The probabilities can be added, and then a threshold can be applied. Combining preprocessed signals (i.e., raw data) can also be performed by determining the average of the signals or training a model to make predictions using two (or more) electrodes as input.

[0070] In a specific embodiment of the invention, the system can be configured to sample bioelectrical signals using a sliding window. The system can use the data in the sliding window to identify posture signals within the bioelectrical signals. The sliding window can have a predefined period, such as one to three seconds. The sliding window can have a predefined sampling interval, such as from 0.1 seconds to 0.5 seconds.

[0071] Electrode signals can be sampled at different times. A sampler can be used for a series of samples. The sampler can be applied in the time domain (convolution) or the frequency domain (Fast Fourier Transform, coefficient multiplication, and Inverse Fast Fourier Transform). This series of samples can span a time period of 1 to 3 seconds (e.g., a 2-second time window). The 1 to 3-second window range is a trade-off between computational resources, accuracy, and response time. In fact, a longer observation window (e.g., 20 seconds) can provide a clear jaw bite signal, but it is too slow (lagging) and computationally too expensive. However, a 2-second time window is a very long response time (lagging), which is unacceptable in most cases. To reduce detection latency, the sample window can be "slid" slightly, for example, 0.2 seconds. In other words, every 0.2 seconds, the older 0.2-second sample is deleted, the sample is moved 0.2 seconds, and then the next 0.2-second sample is added. Sampling can be performed using a trained artificial intelligence (AI) model.

[0072] In a specific embodiment of the invention, the system can continuously record EMG data, for example, measuring a new bioelectrical signal value for each channel every 1 / Fs second (Fs is the sampling frequency). For example, for Fs = 250 Hz, a value is measured every 4 ms. To perform signal processing and prediction operations, the system may require a fixed-length input window. For example, when Fs = 250 Hz, the system can use a fixed-size window of N = 500 data points per channel as input, corresponding to a signal duration of T = 500 / 250 = 2 s. Since the system continuously records the object's EMG data, updating the time window is meaningful so that it includes the latest data recorded by the system but still contains N data points. One solution for this is to use a sliding window. Figure 5 This includes an example of a double mandibular clenching signal 500 measured by electrodes. Figure 5 Includes an example of a sliding window of length T that is updated at fixed time intervals S, suitable for signal 500.

[0073] A long window size (Tlarge) allows observation of a phenomenon (such as the double mandibular clenching shown in the figure), as well as what occurs before and after a longer time period. This can improve the accuracy of systems designed to detect such phenomena, as it helps avoid confusion with other phenomena that may have very similar footprints over shorter time windows. On the other hand, a long window size may mean that signal processing and prediction operations require more processing time and / or power, as the number of operations performed by the processing unit may be higher.

[0074] If the hardware (such as an embedded system with limited computing power and memory) has limitations on computing power, then processing time can increase, as can the latency before the system returns a prediction. This can be a problem if the system needs to respond as quickly as possible (e.g., if the system is used to play / pause music, there shouldn't be too long a wait between executing jaw posture and playing / pausing the music). Selecting hardware capable of allocating a significant amount of computing power to make the processing time fall within an appropriate range over a large window is a challenge, especially when size and battery consumption are limited.

[0075] Reference Figure 5In the specific embodiment described, the system can calculate a new signal processing and prediction operation every S seconds. If S is large, there may be some additional delay before the system detects the phenomenon. For example, if S = 3s, the phenomenon may be detected within a maximum of 3s after the actual execution of the gesture. Therefore, a short time interval S can reduce the system's response time. On the other hand, a shorter time interval S also means that the system may need to calculate the signal processing and prediction operation more frequently, which may be limited by the stated potential computing power and battery consumption. For the purposes of this disclosure, it is estimated that using a T-window between 1 and 3 seconds and a time interval S between 0.1 and 0.5 seconds is a good range. This is not a limitation of the invention, as other ranges may be used depending on the specific capabilities, desired results, and limitations of the particular system.

[0076] In specific embodiments of the present invention, the system can be configured to operate in a specific mode. For example, the system can be configured to operate in a low-power high-sensitivity mode and / or a high-power high-precision mode. Depending on specific factors such as events or system states, the system can be configured to switch between modes. For example, the system can be configured to detect potential gesture signals in a low-power high-sensitivity mode and trigger a high-power high-precision mode upon detection of a potential gesture signal.

[0077] Predictive models with good performance (accuracy and recall) can be quite complex, thus requiring significantly more processing power compared to very basic predictive models (e.g., comparing signal amplitude to a predefined threshold). Continuously running such a complex model (for each input time window) can negatively impact battery life and is generally not useful, as users may not perform the gesture of interest for extended periods. To mitigate this problem, specific embodiments of the present invention relate to a system that can continuously run a simpler model (in a low-power, high-sensitivity mode) with lower accuracy but higher recall (i.e., the model can predict a large number of false positives but validate a small number of false negatives). Then, when this simpler model predicts a positive, a more complex model (in a high-power, high-accuracy mode) can be run to optimize the prediction. By doing so, the impact on battery life can be significantly limited.

[0078] Simpler models (low-power, high-sensitivity modes) can be implemented in several ways. The model can contain very simple operations, such as thresholding. For example, if the maximum voltage amplitude observed within the signal window is higher than T, a more complex prediction model (high-power, high-precision mode) can be triggered. The model can serve as a shorter input window, thus significantly reducing the number of computational operations.

[0079] For more complex models (high-power, high-precision modes), they can be trained offline, for example in... Figure 5The model operates on a sliding window or other signal window. This model can be used for high precision and high recall to provide the best possible predictions. To maximize the performance of this model, signals from both ears can be combined as previously described in this invention. Furthermore, the model can be run on consecutive sliding windows and average prediction probabilities. If all time windows contain this phenomenon, the average prediction probability can reduce the variance of the predictions, which may improve the results. A potential drawback of this solution is that it can increase the response time between the time point of gesture execution and the time point of triggering a positive prediction. For example, if we consider predictions for three consecutive 2-second windows every 0.2 seconds, averaging these three predictions is likely to increase the confidence of the overall prediction, but the waiting time before triggering control will be 2 x 0.2s = 0.4 seconds.

[0080] More complex models can be trained to detect phenomena within a given time window. This way, if the same phenomenon appears in consecutive windows, nothing prevents the system from triggering a positive prediction for all those windows. Several methods can mitigate this problem. For example, if the algorithm used provides a way to locate key descriptive features (such as peak detection), it can detect whether phenomena detected in consecutive windows correspond to the same implementation. As another example, a buffer period can be defined for which a positive prediction is not triggered after a phenomenon has been positively detected.

[0081] In specific embodiments of the invention, the system may further include filters, such as filters 109 and 119, for filtering bioelectrical signals measured by electrodes. The filters may include at least a high-pass filter response. In specific embodiments of the invention, the high-pass filter response may have a cutoff frequency below 90 Hz. In specific embodiments of the invention, the cutoff frequency of the high-pass filter response may be between 50 Hz and 90 Hz. The cutoff frequency can be described from the stopband (below the cutoff frequency) to the passband (above the cutoff frequency) of the high-pass filter. In specific embodiments, the cutoff frequency may be the -3 dB frequency of the filter response, meaning that at the cutoff frequency, the filter has attenuated the signal to -3 dB relative to the filter response in the passband. The filter may alternatively or in combination further include at least one notch filter response to attenuate the signal. The attenuation may be within the passband of the high-pass filter response. For example, the filter may include a notch filter that attenuates signals at frequencies of 100 Hz and / or 120 Hz. In specific embodiments of the invention, at least one notch filter response may have a notch including at least one of 100 Hz and 120 Hz. Filters can be analog, digital, or mixed-signal filters.

[0082] EMG signals mostly fall within the 50-500Hz frequency range, with a greater emphasis on the 50-150Hz range. To observe these phenomena, a suitable sampling frequency can be chosen for the ADC. According to the Nyquist-Shannon theorem, a sampling frequency Fs allows observation of phenomena within the range of 0Hz to Fs / 2Hz without aliasing. Therefore, the highest possible sampling frequency can be chosen. On the other hand, a high sampling frequency may mean high memory and computation time required to process the signal window. A good trade-off is to choose a sampling frequency no lower than 125Hz (otherwise it would be difficult to observe a suitable EMG signal) rather than higher than 300Hz (because most desired signals are likely below 150Hz). In a specific embodiment of the invention, the electrode signal can be sampled at around 250Hz, at least 125Hz, and below 500Hz. This is a trade-off between not being too high to limit the dataset (below 500Hz) and not being too low to detect Nyquist-based EMG signals (at least 125Hz).

[0083] EMG systems can be affected by surrounding electromagnetic perturbations. The most common source is likely the domestic power grid, which generates 50Hz and its harmonics in Europe and 60Hz and its harmonics in the United States. This noise may overlap with the EMG frequency range. Specific embodiments of the invention use filters to remove this noise, thus preventing contamination of the EMG signal.

[0084] Filters can be implemented in a variety of ways. For example, filters can be implemented using analog components (R, L, C components, with or without active gain operational amplifiers) that filter analog signals. As another example, filters can be implemented using digital functions that take the discretized and digitized signal from the ADC output as input. There are many libraries that implement such filters, broadly categorized into FIR (Finite Impulse Response) and IIR (Infinite Impulse Response) filters, and they offer various strategies for performing precondition operations on the signal window (e.g., padding, FFT, etc.).

[0085] Specific embodiments of the invention consider only frequencies above 60 Hz, for example, using high-pass filters with cutoff frequencies between 65 Hz and 80 Hz, depending on how well the filter attenuates near the cutoff frequency. This allows for the elimination of disturbances caused by 50 Hz or 60 Hz noise. To mitigate the effects of harmonics (e.g., 100 Hz, 120 Hz), specific embodiments of the invention use notch filters, which are essentially band-stop filters with a very narrow suppression band centered at a given frequency. This leaves most signals at other frequencies unaffected.

[0086] The filters used here can be digital filters of various forms. For example, a coefficient window can be applied using convolution of the signal in the time domain, or a coefficient window can be applied using point-by-point multiplication of the signal in the frequency domain. In any case, the more precise the desired filtering, the longer the coefficient window. For example, if using convolution in the time domain and applying a 100Hz notch filter with a 20dB suppression band of 1Hz (i.e., the signal attenuates by 20dB or more between 99.5Hz and 100.5Hz), the window may be much longer than if the same filter were applied but with a 3Hz suppression band. These considerations can have an impact. For example, if the filter's coefficient window is significantly longer than the length of the signal window being analyzed, the result of the filtering operation may be suboptimal. Furthermore, a long filter coefficient window can imply a high computational load, which can cause problems in highly constrained embedded systems.

[0087] In specific embodiments of the invention, a high-pass filter can be applied between 60Hz and 80Hz, either as an alternative to or in combination with a notch filter. For example, a 60Hz high-pass filter (removing 50Hz), a 65Hz high-pass filter (removing both 50Hz and 60Hz), or an 80Hz high-pass filter (to avoid any leakage if the filter's time window is low and computational power is limited). A high-pass filter (60Hz-80Hz) can also be combined with one or more harmonic notch filters (100Hz, 120Hz, 150Hz, 180Hz, etc.).

[0088] In specific embodiments of the invention, the system can be configured to detect system events or states, such as connected and / or idle states. The system can also be configured to transmit raw data of bioelectrical signals or derivatives thereof upon detection, for uploading the raw data or derivatives thereof, for example, for training algorithms. The raw data or derivatives thereof can then be processed on a server to clean or label the data for training. This data can then be used to retrain the model and pushed to a network of wearable devices using older versions of the model. The derivatives described herein can also be used locally to train models on wearable devices without necessarily requiring remote transmission.

[0089] Derivatives of raw data can include training labels implicitly derived from the raw data and device usage. For example, raw data can be used to perform automatic baseline adjustment (i.e., passively calibrating the signal based on the evolution of measured normal or average activity). In this case, the derivative would identify the data as average baseline data. As another example, event-independent data collected can be processed using unsupervised learning methods (e.g., clustering to identify potential positive and negative examples, manually reviewing, and adding them to the training set with the correct labels). As yet another example, data prior to using "pause music" or "play music" on an accompanying interface (e.g., the touchscreen of a smartphone paired with the model device) might contain an incorrect negative example (the command was not triggered by jaw clenching). In this case, a derivative of the raw data might be a label marking it as a "positive" example in the training set. As yet another example, data prior to using a similar interface element on an accompanying interface overlaying the most recent interface signal generated by the pose recognition system might contain incorrect positive examples. In this case, a derivative of the raw data could be a label marking the relevant raw data as a "negative" example in the training set. Similarly, using "backup" input on an earbud haptic control with gesture recognition components can generate derivatives of the raw data in a similar form. In a further example, the model can identify raw data associated with highly uncertain inferences and label it as training data using an active learning method.

[0090] Raw data can be unprocessed data from bioelectrical signals recorded by electrodes. Raw data can be anonymized and sampled before transmission. It can be uploaded to an external device or server for processing. In this way, external systems can improve "general" detection models based on large amounts of anonymized user data, such as default-stored feature models. Actively learned parameters can be included (e.g., signals recorded around the detection event). For example, actively learned parameters could be other UI signals, such as a false detection of jaw clenching if the user uses another interface (touchscreen or haptics) to perform a gesture-triggered action, such as play / pause.

[0091] The neural interface can be configured to store raw data and derivatives locally directly in the earpiece, and can be synchronized offline with external devices when available bandwidth and / or connectivity is available. Raw data recordings used for offline sharing during idle periods (even partial signal) can be used (from time to time) to train algorithms later (for millions of users).

[0092] The system can collect data and share it with external electronic devices, such as personal user devices (e.g., smartphones, remote servers, etc.). The system includes memory components (e.g., flash memory) for storing raw data (e.g., data acquired at the output of an ADC). In this case, the memory components can be selected to meet size and consumption constraints. This data can be sent to external devices during some predefined synchronization periods (e.g., when the device is charging, when the occupied memory reaches a certain threshold, etc.). Synchronization can be performed wirelessly (e.g., Bluetooth, BLE, WiFi) or via wired connections (e.g., USB-C, Micro-USB).

[0093] In specific embodiments of the present invention, the system can also wirelessly and continuously stream raw data or its derivatives to an external electronic device. In this case, various strategies can be applied together or separately, depending on the power consumption caused by such streaming and the device's power consumption limitations. For example, when the battery is above a certain threshold, only data is streamed; only X minutes of data are streamed every Y minutes, and so on.

[0094] In specific embodiments of the invention, the system can collect not only raw data but also potential "annotations" about that data. Annotations can be examples of the derivatives described above. For instance, the system can provide a method for the user to specify the existence of erroneous positive or negative predictions. For example, using wireless headphones, the user can configure combinations of haptic controls to explicitly indicate that a music play or pause command has been erroneously activated.

[0095] In a specific embodiment of the invention, the raw data can be used to adjust the model online for a specific user using a passive calibration procedure. In another specific embodiment, the system can be configured to determine the average power of a bioelectrical signal when no posture signal is recognized, and to normalize the bioelectrical signal using the determined average power. For example, the average power B of the bioelectrical signal can be measured every X minutes. Assuming the model is trained at an average power level A, the power of the bioelectrical signal can be multiplied by a coefficient A / B to normalize it in real time.

[0096] In a specific embodiment of the invention, the stored feature model is a classifier. The classifier can be implemented as a hardware module, such as a processing block implementing a neural network. In this way, the stored features can be embedded in the classifier, for example, associated with network weights. The classifier can also be implemented in software, for example, as a function called by the processing block. In this way, the stored features can be embedded in the classifier, for example, associated with variables of the function.

[0097] The classifier can be a binary classifier that outputs "yes" or "no" to a query (e.g., whether an input pose signal corresponds to a stored feature). The classifier can also determine the probability that a potential match exists between the input pose signal and the stored features. Furthermore, the classifier can determine the probability that which stored feature is most likely to correspond to the input signal.

[0098] The classifier can also embed a set of false positive gesture signals. These signals can be detected from postures that are not intended to be captured by the system, such as chewing, speaking, swallowing, or head lateral movement. In this way, the classifier can provide information not only related to the postures the system "detects" but also to other postures. This allows the classifier to output not only that the signal is not a "jaw clenching signal" but also that a specific action being performed, such as "speaking." This approach resists false positives by collecting specific datasets of natural head movements that otherwise resemble a jaw clenching posture. The raw data and its derivatives may include user-specific false positive gesture signals. These signals can be explicitly mined from the user (e.g., by providing them with instructions to speak or swallow and labeling the recorded data) or implicitly (e.g., by visually detecting false negatives from a microphone used to detect the user's speech, or by identifying false negatives similar to those in a database extracted from the general population using other methods described above).

[0099] The predictive model described above, employing a complex model (high-power, high-precision mode), can detect jaw clenching postures with very high accuracy (i.e., few false positives) and very high recall (i.e., few false negatives). A significant challenge in achieving this is avoiding confounding the correct jaw clenching events the system will use (whether double-jaw, triple-jaw, or long-jaw clench) with other common postures performed by the object. In particular, some postures are so common that the object may perform them very frequently without intending to use the system to control electronic devices. Some of these postures may involve masseter muscle activity (e.g., chewing, speaking, swallowing), making them good candidates for false positive detection.

[0100] A good strategy to avoid such errors is to collect as much data on these postures as possible so that the model can understand the differences (sometimes subtle) between these postures and the expected jaw clenching posture. If the model uses well-designed features (e.g., rule-based models with thresholds defined on signal amplitude, "classic" machine learning models such as SVC, Rogers regression, etc.), these features need to be designed to contain enough information for the model to distinguish between different postures (e.g., peak counters where a double jaw clench shows two peaks, for example, in a 2-second time window, while chewing shows more peaks (separated by regular time intervals)).

[0101] If the model takes a signal window as input without any prior feature engineering (e.g., a neural network), it can then learn subtle differences between these postures during the training phase. Furthermore, the data used to train the model can significantly impact the latency between a user performing a jaw-clenching posture and the action being triggered on an electronic device.

[0102] As an example, if the model can only be trained using a time window centered on the jaw clenching signal (e.g.) Figure 6 If signal 600 is present, then it is very likely that it will not be able to detect a jaw bite occurring on one side of the window (e.g., as shown in signal 610). However, as previously described in this disclosure, when data is recorded continuously, the first time window containing a given jaw bite event may look like the example of signal 610, with the event shifted to the right. In that case, it may be necessary to wait for more windows until the jaw bite event is centered like signal 600, which may result in increased latency.

[0103] On the other hand, an event cannot be easily detected as soon as it appears in the window, as this could increase the chance of confusing it with other events and triggering false positives. For example, signal 610 might correspond to an object that has just started chewing; in this case, waiting longer would help avoid confusion, since the input window would look like the window in signal 620.

[0104] In a specific embodiment of the invention, the system is configured to monitor the contact condition of the electrodes. The system may be configured to generate an alarm signal upon detecting a malfunction in the contact condition of any electrode. The alarm signal may be a direct indication of poor or missing contact, such as an indication on a display, or providing feedback disclosed herein via a speaker or other means. The alarm signal may also be a control command used to trigger specific actions within the system, such as stopping music playback or ending a call.

[0105] EMG signal quality depends on continuous contact between the electrode and the skin. Without contact, EMG signals cannot be transmitted from the body to electronic devices. If the contact area changes, the electrode impedance also changes, which can directly affect the measured signal by producing electrode motion artifacts.

[0106] Specific embodiments of the present invention relate to monitoring electrode contact quality, which can be done in various ways. For example, a very low-intensity current can be injected into one electrode, and the recordings of the other electrodes can be observed. This is a standard procedure for measuring electrode-skin interface impedance. If the impedance is too high or varies too much, poor contact quality may exist. As another example, the signal measured by the electrodes can be observed. If the electrode-skin contact varies greatly (e.g., due to significant electrode movement), the signal may show a correlated pattern at low frequencies (e.g., if a user walks at 2 steps per second and the earpiece with the embedded electrodes is slightly loose, there may be a strong signal component around 2 Hz). Observing these patterns can serve as an indication of poor electrode-skin contact.

[0107] When the electrodes are no longer in contact with the skin, the output signal of the analog amplifier may diverge to one of its extreme values ​​(or alternately). This is sometimes called lead-off detection. Furthermore, the digital signal at the ADC output may reach its extreme value (and possibly other values). For example, this can be used to detect whether an earplug has been removed and activate some command accordingly (e.g., trigger an alarm, pause music, etc.). Jaw clench detection can be disabled if there is poor contact (due to the signal associated with poor electrode contact).

[0108] Figure 7 Examples of personal head-wearable devices in the form of earplugs according to specific embodiments of the present invention are included. Earplug 700 includes two electrodes 701 and 702 located on the outer surface of the earplug. In this example, the electrodes are located on the ear tip 703 of the earplug, and when the earplug is worn, the electrodes will be placed in the ear canal. In this way, the electrodes will be located within the ear canal when worn.

[0109] As previously described in this disclosure, bioelectrical signals are measured by measuring the differential voltage between a measuring electrode and a reference electrode. To obtain the best possible signal-to-noise ratio (SNR), it is advantageous to place the measuring electrode as close as possible to the desired signal source, and the reference electrode in a position that minimizes signal collection from the source while still being close enough to the measuring electrode to capture a noise component as similar as possible to the noise component captured by the measuring electrode. The theoretically ideal setup includes the measuring electrode recording M = signal + noise, the reference electrode recording R = noise, and the differential signal D = MR = signal. In practice, the noise components captured by the measuring and reference electrodes can differ, and the reference electrode may also capture some signal. In this way, it can be viewed as the measuring electrode recording M = signal 1 + noise 1, the reference electrode recording R = signal 2 + noise 2, and the differential signal D = MR = (signal 1 - signal 2) + (noise 1 - noise 2).

[0110] In specific embodiments of the invention, the electrodes are located inside the ear, as described in reference earplug 700. The electrodes can be located within the ear canal, inferior nasal concha, superior nasal concha, earlobe, tragus, or any other location around the ear. If the system is designed to operate in a "monoear" manner (each earplug can operate independently), both the measuring electrode and the reference electrode are located on the same earplug. This can mean that the distance between the measuring electrode and the reference electrode is significantly limited; in typical EMG systems, the reference electrode is usually much smaller than the ear.

[0111] Therefore, a key factor in improving the SNR of a single-ear setup is maximizing the distance between the measuring and reference electrodes. This disclosure illustrates several non-limiting arrangements for electrode placement. The figure shows an arrangement associated with earplug 700, where both electrodes are placed on the ear tip 703 of the earplug. This arrangement typically allows for good contact between the electrodes and the skin, as the ear tip maintains pressure on the ear canal, but the electrodes are very close together.

[0112] The diagram shows another configuration of earplugs 710 and 720, with one or both electrodes located on the earplug housing 723 (electrodes 711 and 712 in earplug 710, electrode 721 in earplug 720). Ensuring good and continuous contact between the electrodes and the skin using this configuration can be more complex, especially since the housing size is typically fixed and may not be customizable to the user's ear shape and size. Therefore, the contact area between the earplug housing and the ear may vary from object to object. However, on the other hand, using the housing to place the electrodes allows full utilization of the earplug's entire size and maximizes the distance between the electrodes (e.g., placing the measuring electrode at position 712 and the reference electrode at position 721).

[0113] To address the shortcomings of these solutions, in specific embodiments of the invention, portions of the earplug, such as ear tip 733 and ear wing 734, representing earplug 730, can be used to place electrodes. In these embodiments, one electrode may be located on the ear wing of the earplug, and another electrode on the ear tip. Soft conductive materials can be used to manufacture the conductive ear tip and ear wing (at least partially). When these portions are made of soft and flexible materials, and when they are worn so that they apply pressure to the skin of the ear canal (ear tip) or superior nasal concha (ear wing), good and consistent electro-skin contact can be maintained. These portions of the earplug can be of different sizes (as shown as a block in earplug 730, or individually), which allows for a degree of customization to the ear size and shape of a particular individual, thereby improving contact quality. These components of the earplug are replaceable, meaning they can be replaced while worn, thus not affecting the durability of the overall system. Furthermore, this makes cleaning these portions easier. The ear wing electrode may include one or more gold / conductive electrodes, such as small plugs, which can be of various sizes and apply pressure to the skin to ensure electrode contact with the skin.

[0114] To ensure an electrical connection between the detachable soft electrodes and the electronics contained within the earbud housing, a system can be used where conductive elements (which may be made of metal or other conductive materials) can serve as connectors. For example... Figure 8 As shown, the connector may have a portion protruding from the housing surface and a portion extending into the AFE inside the housing. 800 shows a connector 801 for ear wings and ear tips. 810 and 820 show variations of the ear wing and ear tip connector with hooks 815 for holding the ear wing or ear tip in a predetermined position.

[0115] The materials for electrodes (such as eartip and wing electrodes) can be selected from a variety of candidate materials with certain properties. For example, the material can be a good electronic conductor or a good transducer for converting ionic current into electronic current, so that the resulting electrode-skin interface can have an acceptable impedance (e.g., less than 1 M Ohm for electrodes of a typical size considered). For example, those good conductivity properties can be obtained by using a conductive gel between the electrode and the skin, but using such a gel may not be suitable, so the selected materials can achieve good conductivity by using them to construct so-called “dry electrodes.”

[0116] In specific embodiments of the invention, the electrode material can also be waterproof so that sweat does not degrade it too quickly, and so that electrodes made of these materials can be easily washed with water, for example, to remove earwax that may accumulate during normal use. Furthermore, this material is comfortable enough to be comparable to existing non-conductive earplugs and ear wings currently integrated into earplugs, so that users can wear earplugs for several hours without pain or discomfort.

[0117] Candidate materials may include composite polymers, such as silicone resins (e.g., PDMS, copolyesters, etc.) mixed with electronically conductive particles (e.g., carbon black, CNT, graphite, Ag, etc.) and / or ionicly conductive particles (e.g., Ag / AgCl, etc.). Candidate materials may also include intrinsically conductive polymers, including polymers that possess conductive properties due to their original composition (e.g., PEDOT:PSS) mixed with additives to impart water resistance, stretchability, or adhesive properties (e.g., waterborne polyurethane-WPU, D-sorbitol). Candidate materials may also include metal-coated fabrics (e.g., using silver and silver chloride coatings) and other suitable materials. The examples provided herein are not limiting to a limited range of candidate materials.

[0118] In specific embodiments of the present invention, the electrode may include the surface of a composite polymer. The composite polymer may be a rubber polymer mixed with conductive particles. The rubber polymer may be a soft silicone resin. The conductive particles may include at least one selected from carbon black, carbon nanotubes, graphite, silver, and silver chloride. In specific embodiments of the present invention, the electrode includes an intrinsically conductive polymer. The intrinsically conductive polymer may have a conductive original composition and waterproof additives.

[0119] As described above in this disclosure, the system according to specific embodiments of the invention can provide different kinds of feedback in different situations. Feedback can be auditory, visual, and / or tactile. Feedback can be provided if the user wants to know whether the measurement signal is good enough, and this can trigger EMG technology. Feedback can also be provided if the user wants verification that the posture has been correctly identified and the associated action has been performed. Many other non-limiting situations are also disclosed herein, where feedback may be an important feature.

[0120] In specific embodiments of the invention, the system can provide feedback related to signal quality detection. For example, one or more feedbacks can be provided to indicate good or poor signal quality. This detection can be triggered according to various conditions. For example, feedback can be continuously provided when the headphones are worn. In these cases, the quality of the EMG signal is continuously measured whenever contact is detected, and feedback is triggered appropriately.

[0121] Feedback can also be provided when the headphones are worn, following specific events. Signal quality is likely to be consistently poor while a user is wearing the headphones. However, providing too much feedback over time can be frustrating, potentially necessitating limiting the timing of signal quality detection. Feedback can also be provided before the headphones are worn, such as if the system detects that the user is about to or has just placed the headphones in their ears, for example, when the user removes the headphones from the case, or when a contact is detected that wasn't previously detected. In these cases, signal quality detection can be activated for the next 1-5 minutes.

[0122] If a user has just received feedback but hasn't taken any action, no further feedback will be provided for the next 2-10 minutes. When a user has just triggered gesture, haptic, or voice control on the device, signal quality detection may need to be activated for the next 1-5 minutes. This is especially true if the user triggers haptic / voice control after the system has detected gesture control, as this likely indicates a false positive due to poor signal quality.

[0123] Different types of feedback can be provided. For example, when the headphones are worn before they were worn, audio feedback can be provided. For instance, if good signal quality is detected, the feedback could be a standard beep, beep, or any type of sound; otherwise, it would be muted. In this case, another type of audio feedback could be a beep, beep, or any signal quality-specific sound. Audio feedback can also be provided when the headphones are worn, for example, if the signal quality changes, audio feedback can be provided via a standard beep, beep, or any type of sound, and / or when the signal quality changes via a beep, beep, or any type of sound specific to the signal quality.

[0124] Tactile feedback can also be provided, for example, when the headphones are worn moments before they were worn, a standard vibration is used if good signal quality is detected, except otherwise, and / or signal quality-specific vibrations (e.g., long vibrations indicate good signal quality, short vibrations indicate poor signal quality). Feedback can also be provided when the headphones are worn, for example, a standard vibration is used if the signal quality changes; if the signal quality changes, a signal quality-specific vibration is used (e.g., long vibrations indicate good signal quality, short vibrations otherwise).

[0125] Feedback can also be provided to confirm gesture detection. Feedback can be given to users when they perform a gesture. The gesture can be detected, but the action it triggers may not produce any direct visual, tactile, or audio consequences. For example, if a user opens a communication channel using a gesture, the user can ensure the channel is correctly opened before starting a conversation. Different types of feedback can be auditory feedback (with a beeping / beeping sound or any type of sound when the system detects a gesture) and / or tactile feedback (with a simple vibration when the system detects a gesture).

[0126] In specific embodiments of the invention, the system can be implemented to comply with biocompatibility specifications, impermeability specifications, or other specifications. For example, dust and water resistance can be measured via IP codes. Sold headphones may display IPXY resistance, where X represents "physical particle resistance" (dust) and Y represents "liquid particle resistance" (water). Not all headphones have "dust" or "water" resistance indicators.

[0127] While the specification has been described in detail with reference to specific embodiments of the invention, it should be understood that those skilled in the art, upon understanding the foregoing, can readily conceive of modifications, variations, and equivalents to these embodiments. For example, although the example of an earplug has been used in this invention, any wearable device, especially one positioned close to the user's head, can be used, including eyeglasses, contact lenses, earrings, necklaces, hats, and other wearable devices. Although many examples of jaw clenching recognition have been given, the concepts disclosed herein are equally applicable to recognizing other postures, including hand, arm, and finger movements, eye and eyelid movements, and other postures. These and other modifications and variations of the invention can be practiced by those skilled in the art without departing from the scope of the invention, the scope of which is set forth in more detail in the appended claims.

Claims

1. A posture detection system, comprising: A personal head-mounted wearable device, configured to be worn on the user's head; A first electrode and a second electrode are disposed on the personal head-wearable device and positioned to contact the user's skin when worn, wherein at least the first electrode and the second electrode measure bioelectrical signals and generate raw data of the bioelectrical signals; One or more processors; as well as One or more non-transitory computer-readable media storing instructions that, when executed by said one or more processors, cause the system to: The bioelectrical signals are analyzed using a stored feature model of the posture signals to identify the posture signals within the bioelectrical signals. After recognizing the posture signal in the bioelectric signal, an interface signal is generated.

2. The posture detection system according to claim 1, wherein, The posture signal corresponds to a voluntary jaw-clenching posture that includes one or more jaw-clenching events, said jaw-clenching events including at least one of the following: (i) A single mandibular clenching signal; (ii) Clutching signal from both lower jaws; (iii) Tri-jaw clenching signal; and (iv) Long jaw clenching signal.

3. The posture detection system according to claim 1 further includes: User interface output (106); The one or more non-transitory computer-readable media further store instructions that, when executed by the system, cause the system to: Generate a prompt (301) to perform the posture associated with the posture signal; and The stored feature model is updated using the gesture signal to generate a modified stored feature model; The interface signal is output at the user interface output terminal (106); and The stored feature model is a default model associated with the pose.

4. The posture detection system according to claim 1, wherein: The first electrode and the second electrode are located on the outer surface of the personal head-wearable device.

5. The posture detection system of claim 1, wherein the personal head-wearable device includes a first human head-wearable device element, and the system further includes: Second personal head-wearable device components; as well as The third electrode (112) and the fourth electrode (113); The third electrode (112) and the fourth electrode (113) measure the bioelectrical signal; The first electrode and the second electrode are located on the outer surface of the first personal head wearable device element; The third electrode (112) and the fourth electrode (113) are located on the outer surface of the second personal head-wearable device element; and The identification of the posture signal in the bioelectric signal uses the following combination: (i) first data measured by the first electrode and the second electrode; and (ii) second data measured by the third electrode (112) and the fourth electrode (113).

6. The posture detection system of claim 5, wherein the one or more non-transitory computer-readable media storing instructions, when executed by the system, causes the system to: Determine whether the bioelectric signal exists in the first data and the second data; in, If the bioelectric signal is present in both the first data and the second data, the identification of the posture signal in the bioelectric signal uses a combination with the stored feature model; and If the bioelectric signal exists only in one of the first data and the second data, then the recognition of the posture signal in the bioelectric signal uses only one of the first data and the second data, and employs the stored feature model.

7. The posture detection system according to claim 1, wherein: The identification of the posture signal in the bioelectrical signal uses a combination of first data measured by the first electrode and second data measured by the second electrode; and The system is configured as follows: Determine the common-mode signal of the first data and the second data; and Feedback is provided in the opposite direction to the common-mode signal to eliminate the common-mode signal.

8. The posture detection system according to claim 1, wherein: The personal head-wearable device includes a first personal head-wearable device element and a second personal head-wearable device element wiredly connected to the first personal head-wearable device element; and The system also includes a third electrode (112) and a fourth electrode (113) located on the outer surface of the second personal head wearable device element. The identification of the posture signal in the bioelectric signal uses the following combination: (i) data measured by the first electrode and the fourth electrode; and (ii) data measured by the third electrode and the second electrode.

9. The posture detection system of claim 1, wherein the personal head-wearable device is a first human head-wearable device, and the system further comprises: The third electrode is located on the outer surface of the first personal head-wearable device; A second personal head-worn wearable device; and The fourth, fifth, and sixth electrodes are located on the outer surface of the second personal head-wearable device; The identification of the posture signal in the bioelectric signal uses the following combination: (i) using the third electrode as the first reference electrode, and data measured by the first electrode and the second electrode; (ii) using the sixth electrode as a second reference electrode, the data measured by the fourth and fifth electrodes.

10. The posture detection system according to claim 1, wherein, The one or more non-transitory computer-readable medium storage instructions, when executed by the system, cause the system to: The bioelectrical signals are sampled using a sliding window; and Using data from the sliding window, the recognition of the posture signal in the bioelectrical signal is performed; The period of the sliding window is 1 to 3 seconds; and The sampling interval of the sliding window is 0.1 to 0.5 seconds.

11. The posture detection system according to claim 1, wherein: The first electrode and the second electrode generate the raw data of the bioelectrical signal; and The one or more non-transitory computer-readable medium storage instructions, when executed by the system, cause the system to: Detect one of the following: (i) connection to the system; (ii) idle state of the system; and At the time of detection, at least one of the following is transmitted: (i) the original data of the bioelectric signal; and (ii) a derivative of the original data of the bioelectric signal.

12. The posture detection system according to claim 1, wherein, The one or more non-transitory computer-readable medium storage instructions, when executed by the system, cause the system to: When no posture signal is identified, determine the average power of the bioelectrical signal; and The bioelectric signal is normalized using the determined average power.

13. The posture detection system according to claim 1, wherein, The one or more non-transitory computer-readable medium storage instructions, when executed by the system, cause the system to: Operating in low-power, high-sensitivity mode; Detect potential attitude signals in the low-power, high-sensitivity mode; and Upon detection of the potential attitude signal, a high-power, high-precision mode is triggered.

14. The posture detection system according to claim 1, further comprising: A filter (109) is used to filter the bioelectrical signal measured by the first electrode and the second electrode; The filter (109) includes at least a high-pass filter response; and The high-pass filter response has a cutoff frequency between 50 Hz and 90 Hz.

15. The posture detection system according to claim 1, further comprising: A filter (109) is used to filter the bioelectrical signal measured by the first electrode and the second electrode; The filter (109) is configured to suppress at least one of the following: 100Hz; 120Hz; 100Hz and 120Hz.

16. The posture detection system according to claim 1, wherein: The one or more non-transitory computer-readable medium storage instructions, when executed by the one or more processors, cause the system to sample the bioelectrical signal measured by the first electrode and the second electrode using a sampling frequency; and The sampling frequency is between 125Hz and 500Hz.

17. The posture detection system according to claim 1, wherein: The stored feature model is a classifier; The classifier also embeds a set of incorrect positive pose signals; and The set of erroneous positive posture signals includes at least one of the following: (i) chewing signal; (ii) talking signal; (iii) swallowing signal; (iv) head lateral movement signal.

18. The posture detection system of claim 1, wherein the one or more non-transitory computer-readable media further stores instructions that, when executed by the system, cause the system to: Monitor the contact condition between the first electrode and the second electrode; and An alarm signal is generated when a fault is detected in the contact condition of the first electrode or the second electrode.

19. The posture detection system according to claim 1, wherein: The first electrode and the second electrode each include a surface of a composite polymer configured for skin contact; and The composite polymer is a rubber polymer mixed with conductive particles.

20. The posture detection system according to claim 19, wherein: The rubber polymer is a soft silicone resin; and The conductive particles include at least one of the following: carbon black, carbon nanotubes, graphite, silver, and silver chloride.

21. The posture detection system according to claim 1, wherein: The first electrode (102) and the second electrode (103) comprise intrinsically conductive polymers; and The intrinsically conductive polymer has a conductive original composition and a waterproof additive.

22. The posture detection system according to claim 1, further comprising: Speaker (106); The one or more non-transitory computer-readable media further stores instructions that, when executed by the one or more processors, cause the system to generate an auditory feedback signal, which is presented by the speaker when the interface signal is generated.

23. A wearable posture recognition system, comprising: First head-mounted wearable device component; A first electrode and a second electrode are located on the first head-worn wearable device element and positioned to contact the user's skin, wherein the first electrode and the second electrode measure bioelectrical signals; Second head-mounted wearable device components; A third electrode and a fourth electrode, wherein the third electrode and the fourth electrode are located on the second head-wearable device element, and wherein the third electrode and the fourth electrode measure the bioelectrical signal; One or more processors; and A non-transitory computer-readable medium storing one or more instructions, which, when executed by the system, cause the system to: Analyze the bioelectrical signals to identify posture signals within them; and An interface signal is generated when the posture signal in the bioelectric signal is identified; The identification of the posture signal in the bioelectric signal uses the following combination: (i) data measured by the first electrode and the second electrode; (ii) Data measured by the third electrode and the fourth electrode.

24. The wearable posture recognition system according to claim 23, further comprising: The fifth electrode is located on the outer surface of the first wearable device component on the human head. and The sixth electrode is located on the outer surface of the second personal head-wearable device element; The identification of the posture signal of the bioelectric signal uses the following combination: (i) data measured by the first electrode (102) and the second electrode (103) using the fifth electrode as the reference electrode for differential measurement; and (ii) data measured by the third electrode (112) and the fourth electrode (113) using the sixth electrode as the second reference electrode.

25. The wearable posture recognition system according to claim 23, further comprising: User interface output; and Wherein, the one or more non-transitory computer-readable medium storage instructions, when executed by the system, cause the system to: Generate prompts to execute the posture associated with the posture signal; The bioelectrical signals are analyzed to locate the posture signals in the bioelectrical signals using a stored feature model of the posture signals; An interface signal is generated when the posture signal in the bioelectric signal is identified; and The stored feature model is updated using the gesture signal to generate a modified stored feature model; The interface signal output is located at the user interface output terminal; and The stored feature model is a default feature model associated with the pose.

26. A posture detection method for a personal head-worn wearable device, comprising the following steps: Bioelectrical signals are measured using a first electrode and a second electrode, wherein the first electrode and the second electrode are located on the personal head-wearable device so as to contact the user's skin when worn; Using the stored feature model of the posture signal, analyze (152) the bioelectric signal to identify the posture signal in the bioelectric signal; and An interface signal (153) is generated when the posture signal in the bioelectric signal is identified.

27. The posture detection system of claim 1, wherein the personal head-worn device comprises at least one of the following: (i) Eyeglasses; (ii) Handpiece; (iii) Headphones; and (iv) Headband.

28. The posture detection system of claim 1, wherein the personal head-wearable device includes glasses, the glasses including a frame, and wherein: The first electrode is located on the temple portion of the frame; and The second electrode is located on the bridge of the nose or the second temple portion of the frame.

29. The posture detection system according to claim 28, wherein: The first electrode is located on the first side of the head; and The second electrode is located on the opposite side of the head, so that the bioelectric signal is measured as a differential signal of the head.

30. The posture detection system of claim 28, wherein the electrodes are positioned to contact at least one of: (i) Temple area; (ii) Forehead area; (iii) Nasal bridge area; (iv) The area behind the ear; and (v) The area adjacent to the masseter muscle.

31. The posture detection system of claim 28, wherein the electrode is integrated into a conductive portion of the frame formed of a polymer material containing conductive particles.

32. The posture detection system according to claim 28, further comprising: The third and fourth electrodes are disposed on the eyeglasses; The identification uses a combination of signals from the plurality of electrodes.

33. The posture detection system according to claim 28, wherein: The electrodes are located where the glasses apply pressure to the skin to improve the contact between the electrodes and the skin.

34. The posture detection system according to claim 1, wherein: The personal head-wearable device includes glasses and headphones; and The identification uses signals measured on the glasses and the headphones.

35. The posture detection system according to claim 1, wherein: The stored signature model includes a machine learning model trained to distinguish between voluntary jaw clenching postures and involuntary facial muscle activity.

36. The posture detection system according to claim 1, wherein: The electrodes are positioned away from the user's ear canal.

37. A wearable posture recognition system, comprising: The first wearable device component for a human head is configured to be located at or near the user's first ear; A second personal head-wearable device element is configured to be located at or near the user's second ear; A first electrode and a second electrode are disposed on the first personal head-wearable device element and configured to measure bioelectrical signals; and A third electrode and a fourth electrode are disposed on the second personal head-wearable device element and configured to measure the bioelectrical signal; One or more processors; and One or more non-transitory computer-readable media store instructions that, when executed by the one or more processors, cause the system to: The following combination was used to analyze the bioelectrical signals to identify posture signals: (i) First data measured by the first electrode and the second electrode; and (ii) Second data measured by the third electrode and the fourth electrode; An interface signal is generated after recognizing the posture signal in the bioelectrical signal; and Output a command signal corresponding to the identified posture signal.

38. A biosignal interface device, comprising: One or more processors; and One or more non-transitory computer-readable media store instructions that, when executed by the one or more processors, cause the system to: Operating in a low-power, high-sensitivity mode, the low-power, high-sensitivity mode is configured to detect potential gesture signals in bioelectrical signals using a first prediction model; and In response to the detection of the potential pose signal, operation is performed in a higher power, higher precision mode configured to perform pose recognition using a second prediction model with higher computational complexity than the first prediction model.

39. A pose recognition system, comprising: At least one electrode is configured to be located near at least one of the user's ears and to measure bioelectrical signals; A signal acquisition circuit is configured to receive the bioelectric signal; and One or more processors; and One or more non-transitory computer-readable media store instructions that, when executed by the one or more processors, cause the system to: Generate prompts to execute the posture associated with the posture signal; The bioelectrical signals are analyzed to identify the posture signals using a stored feature model associated with them; An interface signal is generated when the posture signal in the bioelectric signal is identified; The stored feature model is updated using the gesture signal to generate a modified stored feature model; The interface signal output is located at the user interface output terminal; and The stored feature model is a default feature model associated with the pose.

Images