Speaking valve and processing system and method thereof

Abstract

The invention discloses a speaking valve and a processing system and method thereof, and the method comprises the steps: dividing a voice signal collected by a voice collection module into a plurality of frequency bands, and carrying out the independent denoising processing of the plurality of frequency bands: the plurality of frequency bands comprise a first frequency band and a second frequency band, the first frequency band comprises airflow noise, and the second frequency band comprises airflow noise; the second frequency band comprises environmental noise, and the frequency of the first frequency band is smaller than that of the second frequency band; respectively removing the airflow noise of the first frequency band and the environmental noise of the second frequency band to obtain a denoised first frequency band and a denoised second frequency band; fusing the denoised signals of the first frequency band and the denoised signals of the second frequency band to obtain a final denoised voice signal; and acoustic feature extraction is carried out according to the final de-noised voice signal, and the pronunciation condition is acquired according to the acoustic features. According to the embodiment of the invention, airflow noise and environmental noise interference can be removed more accurately, the definition of the voice signal can be reserved to the maximum extent, and quantitative and objective pronunciation conditions can be obtained.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of monitoring, in particular to a speaking valve and a processing system and method thereof. BACKGROUND

[0002] In the clinical rehabilitation management of patients with severe tracheostomy, speech rehabilitation and swallowing function assessment still mainly rely on the subjective auditory judgment of medical staff, and there is a lack of special tools that can realize objective and quantitative assessment. This situation leads to the fact that key speech behaviors such as the duration, frequency and volume of the patient's pronunciation cannot be accurately recorded, and the real-time dynamic changes of vocal cord vibration are difficult to be scientifically analyzed, which not only restricts the objective judgment of the effect of speech rehabilitation training, but also is not conducive to optimizing the individualized training plan according to the actual situation of the patient, and to a certain extent, affects the rehabilitation process. SUMMARY

[0003] The present application provides a speaking valve and a processing system and method thereof, which are used to realize objective and quantitative assessment of the pronunciation of patients.

[0004] The present application provides a speech processing method based on a speaking valve, wherein a voice collecting module is arranged on the cover of the speaking valve, and the processing method comprises the following steps:

[0005] The voice signal collected by the voice collecting module is divided into a plurality of frequency bands, and independent denoising processing is performed on the plurality of frequency bands respectively:

[0006] The plurality of frequency bands include a first frequency band and a second frequency band, the first frequency band includes airflow noise, the second frequency band includes environmental noise, and the frequency of the first frequency band is less than the frequency of the second frequency band; the airflow noise of the first frequency band and the environmental noise of the second frequency band are removed respectively to obtain a denoised first frequency band and a denoised second frequency band;

[0007] The signals of the denoised first frequency band and the denoised second frequency band are fused to obtain a final denoised voice signal;

[0008] Acoustic feature extraction is performed according to the final denoised voice signal, and the pronunciation condition is obtained according to the acoustic feature.

[0009] Further, the removing of the airflow noise of the first frequency band comprises:

[0010] The airflow noise of the first frequency band is periodically removed by an adaptive periodic filter to obtain the denoised first frequency band.

[0011] Further, before the airflow noise of the first frequency band is periodically removed by the adaptive periodic filter to obtain the denoised first frequency band, the method further comprises the following steps:

[0012] remove the airflow noise below the preset frequency in the first frequency band through a band-pass filter.

[0013] Further, the removing airflow noise periodically from the signal of the first frequency band through the adaptive periodic filter to obtain a denoised first frequency band comprises:

[0014] periodically modeling the airflow noise, the periodically modeling comprising predicting airflow noise variation by detecting periodic changes of respiration;

[0015] removing airflow noise periodically from the signal of the first frequency band according to the airflow noise variation and the adaptive periodic filter.

[0016] Further, the adjusting the adaptive periodic filter according to the airflow noise variation further comprises:

[0017] adopting an adaptive filtering algorithm to estimate the power spectral density of the airflow noise, and dynamically adjusting the coefficients of the adaptive periodic filter according to the estimation result.

[0018] Further, the adjusting the adaptive periodic filter according to the airflow noise variation further comprises:

[0019] adopting a weighted sliding window method to dynamically adjust the weights of the adaptive periodic filter.

[0020] Further, the removing the environmental noise of the second frequency band comprises:

[0021] adopting a spectral subtraction method to estimate the environmental noise and track the variation of the environmental noise in real time;

[0022] first adjusting the coefficients of the adaptive filter according to the variation of the environmental noise, and removing the environmental noise of the second frequency band through the adaptive filter after the first adjustment to obtain a preliminarily denoised second frequency band;

[0023] adopting an adaptive filtering algorithm to estimate the power spectral density of the airflow noise, and secondly adjusting the coefficients of the adaptive filter dynamically according to the estimation result, and removing the environmental noise of the second frequency band through the adaptive filter after the second adjustment to obtain a finally denoised second frequency band.

[0024] Further, the fusing the signal of the denoised first frequency band and the denoised second frequency band to obtain a finally denoised speech signal comprises:

[0025] Error signals are calculated for the de-noised first frequency band and the de-noised second frequency band respectively, and new de-noised first and second frequency bands are obtained by adjusting the de-noising weights corresponding to the de-noised first and second frequency bands according to the error signals, and the signals of the new de-noised first and second frequency bands are fused to obtain a final de-noised speech signal.

[0026] Further, the acoustic features include short-time energy, pitch period and zero-crossing rate, and the pronunciation conditions include maximum pronunciation duration and maximum counting ability.

[0027] The obtaining of the pronunciation conditions according to the acoustic features includes:

[0028] The maximum pronunciation duration is obtained by analyzing the short-time energy and the pitch period, and the maximum counting ability is obtained by analyzing the short-time energy and the zero-crossing rate.

[0029] The speech valve also includes a voice collection and processing assembly.

[0030] One end of the cover is provided with a one-way valve, and the other end is provided with an airflow passage, the airflow passage includes a breathing passage and an oxygen passage, the breathing passage is in communication with the one-way valve, and the middle part of the cover is provided with an oxygen inlet interface, and the oxygen passage is in communication with the oxygen inlet interface.

[0031] The voice collection and processing assembly includes a voice collection module and a voice processing module connected with each other, and the voice collection module is arranged on the surface of the cover and close to the one-way valve.

[0032] The voice processing module is used for acquiring the voice signal collected by the voice collection module and performing the voice processing method in any one of the above embodiments.

[0033] From the above technical solutions, the present application has the following advantages:

[0034] The voice signal of the embodiment includes airflow noise of low frequency and environmental noise of medium-high frequency, the frequency of the airflow noise is lower than that of the environmental noise, the voice signal is divided into a first frequency band and a second frequency band, the frequency of the first frequency band is lower than that of the second frequency band, the first frequency band includes the airflow noise of low frequency, and the second frequency band includes the environmental noise, independent de-noising is performed on the first frequency band and the second frequency band to obtain a de-noised first frequency band and a de-noised second frequency band, the problem that part of the frequency band noise is not completely removed when a single de-noising method is used for full frequency band processing is overcome, the airflow noise and the environmental noise interference can be more accurately removed, the intelligibility of the voice signal can be maximally preserved, the signals of the de-noised first frequency band and the de-noised second frequency band are fused after de-noising to obtain a final de-noised voice signal, and the pronunciation conditions can be quantified and objective through acoustic feature extraction of the final de-noised voice signal. BRIEF DESCRIPTION OF DRAWINGS

[0035] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor under the premise of the drawings.

[0036] Figure 1 A schematic diagram of the overall structure of a speaking valve provided by the embodiment of the present application is shown in the figure.

[0037] Figure 2 A schematic diagram of the working process of a speaking valve provided by the embodiment of the present application is shown in the figure.

[0038] The figure legend is as follows: 1, outer cover; 2, one-way valve; 3, air flow channel; 31, breathing channel; 32, oxygen channel; 4, speech acquisition and processing assembly; 41, speech acquisition module; 42, speech processing module; 5, signal transmitter; 6, laryngeal vibration sensor; 7, turbine; 8, oxygen inlet interface. DETAILED DESCRIPTION

[0039] In order to make the purposes, features and advantages of the present application more obvious and easy to understand, the technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the embodiments described below are only some of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the present application.

[0040] The terms "first", "second", "third", "fourth" and the like in the specification of the present application and the drawings (if any) are used to distinguish similar objects, and do not necessarily indicate a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device that includes a series of steps or units does not necessarily limit to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0041] In clinical practice, in order to assist the tracheostomy patients in voice, the medical staff usually installs a speaking valve on the neck trachea of the patient. On this basis, if an independent voice acquisition and processing component is additionally arranged on the neck of the patient, the physical burden of the larynx of the patient will be directly increased, which may cause discomfort and even affect the safety of breathing, and does not comply with the clinical nursing principle of critical patients.

[0042] Therefore, the inventors put forward an improved design idea of the existing speaking valve: integrating the voice acquisition and processing component on the speaking valve, so that the improved speaking valve has the voice signal acquisition and processing capability while retaining the core function of assisting in voice, and assisting the medical staff in evaluating the voice of the patient. However, the existing speaking valves generally do not have the voice signal acquisition and processing capability, and there is no voice processing system and matching method that can be directly adapted to the speaking valve.

[0043] It should be noted that the inventors improve the "speaking valve" developed by the applicant, and realize that the pronunciation duration and pronunciation conditions of the patient are objectively recorded by arranging the voice acquisition and processing component on the speaking valve, which is beneficial to the evaluation of the voice rehabilitation training effect and the optimization of the training scheme.

[0044] Please refer to Figure 1 , Figure 1 A speaking valve provided by an embodiment of the present application comprises an outer cover 1 and a voice acquisition and processing component 4.

[0045] One end of the outer cover 1 is provided with a one-way valve 2, and the other end is provided with an airflow passage 3, the airflow passage 3 comprising a breathing passage 31 and an oxygen passage 32, the breathing passage 31 being in communication with the one-way valve 2, and the middle part of the outer cover 1 being provided with an oxygen inlet interface 8, the oxygen passage 32 being in communication with the oxygen inlet interface 8.

[0046] The voice acquisition and processing component 4 comprises a voice acquisition module 41 and a voice processing module 42 connected with each other, and the voice acquisition module 41 is arranged on the surface of the outer cover 1 and close to the one-way valve 2.

[0047] The voice processing module 42 is used for acquiring the voice signal collected by the voice acquisition module 41 and performing the voice processing method of the following embodiments.

[0048] It can be understood that, in specific implementation, on the one hand, the cover 1 of the speaking valve is located on the surface of the neck of the patient, is arranged close to the vibration conduction path of the vocal cords of the patient, the voice acquisition module 41 is arranged on the cover 1, and voice signals generated by vibration of the vocal cords can be directly captured. The voice signals are collected by the voice acquisition module 41 after being transmitted to the cover 1 through soft tissues of the neck. On the other hand, since the oxygen channel 32 only occupies part of the space in the cover 1, the voice acquisition module 41 is arranged away from the space in this embodiment, which is beneficial to reducing the influence of airflow noise of the oxygen channel 32 on the voice acquisition module 41.

[0049] It should be noted that, with the progress of oxygen therapy equipment, high-flow oxygen therapy (HFNC) gradually becomes an important means for treating respiratory failure and maintaining stable oxygen concentration, and is especially suitable for patients with diseases such as chronic obstructive pulmonary disease. The existing speaking valve only meets the voice function requirement, and generally does not design an oxygen therapy interface, and cannot be directly connected with the HFNC oxygen therapy equipment, so that patients who need to receive oxygen therapy and use the voice valve at the same time face the double difficulties of insufficient oxygen supply and inability to normally sound. The patient needs to frequently switch between the voice valve and the oxygen therapy equipment, which interrupts the oxygen therapy. For patients who need oxygen therapy, they may be exposed to a low-oxygen environment, increasing the risk of hypoxia and respiratory burden, and there is a great safety hazard, and it is easy to cause dryness of the airway of the patient, damage to the mucous membrane, and increased risk of sputum blockage. The interface and channel for oxygen therapy designed in this embodiment can be directly connected with the HFNC oxygen therapy equipment, so that the speaking valve meets the voice function requirement and can also receive oxygen therapy.

[0050] It should be noted that the voice acquisition module 41 arranged on the cover 1 is more conducive to evaluating the pronunciation of the patient.

[0051] 1) Away from high-speed airflow: The high-flow oxygen of HFNC forms a turbulent noise (low frequency, periodicity) in the trachea or inside the valve. If the microphone is placed in the airflow path, it will be overwhelmed. The cover 1 is located away from the main airflow, reducing noise.

[0052] 2) Close to the vibration source of the vocal cords: The cover 1 is located on the surface of the neck, close to the vibration conduction path of the vocal cords (transmitted through the soft tissues of the neck), and can directly capture the sound wave signals of the vibration of the vocal cords, rather than airflow noise.

[0053] 3) Reduce signal attenuation: The sound in the trachea needs to pass through the artificial airway and secretions, and the signal collected on the surface of the neck is more complete, which is conducive to calculating MPT and MCA.

[0054] The embodiment of the present application provides a voice processing method based on a speaking valve. The cover of the speaking valve is provided with a voice acquisition module. The processing method comprises the following steps.

[0055] The voice signal collected by the voice collection module is divided into multiple frequency bands, and the multiple frequency bands are independently denoised respectively:

[0056] The multiple frequency bands include a first frequency band and a second frequency band, the first frequency band includes airflow noise, and the second frequency band includes environmental noise, the frequency of the first frequency band is lower than the frequency of the second frequency band; the airflow noise in the first frequency band and the environmental noise in the second frequency band are removed respectively to obtain a denoised first frequency band and a denoised second frequency band;

[0057] The signals of the denoised first frequency band and the denoised second frequency band are fused to obtain a final denoised voice signal;

[0058] Acoustic features are extracted according to the final denoised voice signal, and pronunciation conditions are obtained according to the acoustic features.

[0059] It should be noted that the airflow noise is mainly generated when the airflow in the patient's breathing process passes through the device, and is usually manifested as low-frequency noise, with a frequency generally concentrated below 40Hz and obvious periodicity, for example, the patient's inhalation and exhalation cycle. The environmental noise comes from other equipment in the hospital (such as the operation sound of medical equipment), personnel conversation, etc., which is usually distributed in a wide frequency range and has no obvious periodicity, and may partially overlap with the frequency of the patient's voice signal, especially in the medium and high frequency ranges.

[0060] It can be understood that in specific implementation, the voice signal includes low-frequency airflow noise and medium-high frequency environmental noise, which will affect the accuracy of acoustic feature extraction and the objectivity of pronunciation condition acquisition if not removed.

[0061] The frequency of the airflow noise is lower than that of the environmental noise. By dividing the voice signal into a first frequency band and a second frequency band, the frequency of the first frequency band is lower than that of the second frequency band, wherein the first frequency band includes low-frequency airflow noise, and the second frequency band includes medium-frequency and high-frequency environmental noise. Independent denoising of the first frequency band and the second frequency band obtains a denoised first frequency band and a denoised second frequency band, which overcomes the problem that a single denoising method does not completely remove the noise in part of the frequency band when processing the whole frequency band. It can more accurately remove airflow noise and environmental noise interference, and can maximize the clarity of the voice signal. After denoising, the signals of the denoised first frequency band and the denoised second frequency band are fused to obtain a final denoised voice signal. Through acoustic feature extraction of the final denoised voice signal, a quantitative and objective pronunciation condition can be obtained.

[0062] In a more specific embodiment, dividing the voice signal collected by the voice collection module into multiple frequency bands includes:

[0063] The voice signal is divided into multiple frequency bands using Fourier transform, band-pass filtering, or wavelet transform.

[0064] In a more specific embodiment, removing the airflow noise of the first frequency band comprises:

[0065] Periodically removing the airflow noise of the first frequency band from the signal of the first frequency band by the adaptive periodic filter to obtain a de-noised first frequency band.

[0066] It can be understood that in specific implementation, based on the periodicity of the airflow noise, the adaptive periodic filter can dynamically match the noise period, and the airflow de-noising effect is more accurate and more stable.

[0067] In a more specific embodiment, before periodically removing the airflow noise of the first frequency band from the signal of the first frequency band by the adaptive periodic filter to obtain a de-noised first frequency band, it further comprises:

[0068] Removing the airflow noise below the preset frequency in the first frequency band by the band-pass filter.

[0069] It can be understood that in specific implementation, since the airflow noise of low frequency is generally concentrated below 40 Hz, in this embodiment, a band-pass filter is used, and the medium frequency of the band-pass filter is set to 40 Hz, so that the signal below 40 Hz in the low frequency band is removed, and the signal above 40 Hz is retained; based on the characteristics of the airflow noise having low frequency and narrow range, this embodiment first performs band-pass filtering on the airflow noise, preliminarily de-noises, reduces the calculation pressure of the subsequent adaptive filter, and improves the operation efficiency.

[0070] It should be noted that the band-pass filter is used to optimize the low-frequency characteristics of the airflow noise. The design of the band-pass filter should set the low-frequency cutoff point at about 40 Hz to remove the airflow noise, but retain the low-frequency part of the speech signal (such as the low-frequency resonance of speech).

[0071] Design method of the band-pass filter:

[0072] 1) Define the cutoff frequency: in this case, the low-frequency cutoff is 40 Hz, and the high-frequency cutoff is the upper limit of the patient's speech signal (usually set to about 4000 Hz). Suppose we design a band-pass filter from 40 Hz to 4000 Hz.

[0073] 2) Calculate the filter coefficients: calculate the coefficients of the filter by the window function method or the least square method.

[0074] When designing an FIR filter by the window function method, the impulse response of the filter can be calculated by the following formula:

[0075]

[0076] where ω h n and ω ln are the digital frequencies of the high and low frequency cutoff points, n is the discrete-time index of the impulse response.

[0077] 3) Apply filter: Once the filter coefficients are calculated, the speech signal can be filtered by a convolution operation.

[0078] In a more specific embodiment, the periodic removal of the airflow noise from the signal of the first frequency band by the adaptive periodic filter obtains a denoised first frequency band, which includes:

[0079] Periodically modeling the airflow noise, the periodic modeling includes predicting the airflow noise variation by detecting the periodic variation of the breathing;

[0080] Periodically removing the airflow noise from the signal of the first frequency band according to the airflow noise variation combined with the adaptive periodic filter.

[0081] It can be understood that in specific implementation, the embodiment explicitly determines the periodic characteristics of the airflow noise through periodic modeling, so that the filter can specifically remove the airflow noise and reduce the residual airflow noise.

[0082] In a more specific embodiment, adjusting the adaptive periodic filter according to the airflow noise variation further includes:

[0083] An adaptive filtering algorithm is used to estimate the power spectral density of the airflow noise, and the coefficients of the adaptive periodic filter are dynamically adjusted according to the estimation result.

[0084] It can be understood that in specific implementation, the embodiment estimates the energy distribution (i.e., the power spectral density) of the current airflow noise in real time, judges the intensity of the airflow noise at different frequencies, and adjusts the coefficients of the adaptive periodic filter in real time. In the place where the airflow noise is strong, the suppression of the filter is increased; in the place where the airflow noise is weak, the suppression is weakened, so as to ensure that the useful signal is not lost.

[0085] It should be noted that the adaptive filtering algorithm is an LMS (Least Mean Square Error) algorithm.

[0086] The LMS algorithm is a commonly used adaptive filtering method, which optimizes the filter coefficients by minimizing the error between the noise and the target signal. The LMS algorithm is very simple and has high computational efficiency, and is suitable for real-time processing. Its role in the talking valve:

[0087] 1) Adaptive adjustment: The LMS algorithm can automatically adjust the filter coefficients according to the real-time noise variation to adapt to the periodic characteristics of the airflow noise.

[0088] 2) Low delay: The LMS algorithm is simple and has high real-time performance, and is suitable for real-time noise estimation and processing in the talking valve.

[0089] 3) can effectively remove airflow noise while enhancing the speech signal.

[0090] LMS algorithm process:

[0091] 1) Initialize filter: set the initial coefficient of the filter w0.

[0092] 2) Error calculation: at each time n, the input signal is x(n), and the output of the filter is y(n).

[0093] Calculate output: ,

[0094] where, is the current filter coefficient, and x(n) is the input signal.

[0095] Calculate error: ,

[0096] where d(n) is the desired output (usually a noise-free signal), and e(n) is the error signal.

[0097] Update coefficients: update the filter coefficients according to the error signal to adjust the filter to minimize the error:

[0098] ,

[0099] where μ is the step size parameter, which determines the speed of the update.

[0100] In a more specific embodiment, adjusting the adaptive periodic filter to periodically remove airflow noise from the signal in the first frequency band according to changes in airflow noise further comprises:

[0101] Using a weighted sliding window method to dynamically adjust the weight of the adaptive periodic filter.

[0102] It can be understood that in specific implementation, the short-term changes of the signal are captured in real time through the sliding window, and different weights are assigned to the data at different times within the window (such as higher weight for recent data and lower weight for long-term data). This makes the filter more sensitive to the latest changes in the signal, while avoiding the dramatic fluctuations in filter parameters caused by abnormal data at a single time, making the filter output smoother and further improving the stability and timeliness of the filter.

[0103] It should be noted that conventional noise filtering algorithms (such as Wiener filtering) are generally designed to handle environmental noise, but for respiratory airflow noise, especially low-frequency noise, it may lead to misidentification or loss of speech signals.

[0104] In a more specific embodiment, removing environmental noise in the second frequency band comprises:

[0105] The spectral subtraction method is used to estimate the ambient noise to track the change of the ambient noise in real time;

[0106] The adaptive filter is adjusted according to the change of the ambient noise to remove the ambient noise of the second frequency band to obtain a de-noised second frequency band.

[0107] It can be understood that the spectral subtraction method continuously tracks the intensity and frequency characteristics of the ambient noise to provide a "real-time updated noise template" for the filter, avoiding the lag of the noise reduction effect. The adaptive filter dynamically adjusts the filter parameters according to the noise information provided by the spectral subtraction method, and only removes the noise frequency band, thereby maximizing the clarity of the voice.

[0108] It should be noted that the spectral subtraction method can effectively reduce the background noise and ensure that the voice signal can be clearly extracted under a high noise level. The spectral subtraction noise calculation formula is as follows:

[0109] Spectral subtraction noise: When Y(t, f) becomes negative, it is set to zero to ensure non-negativity.

[0110] In a more specific embodiment, adjusting the adaptive filter according to the change of the ambient noise to remove the ambient noise of the second frequency band further comprises:

[0111] An adaptive filtering algorithm is used to estimate the power spectral density of the airflow noise, and the coefficients of the adaptive periodic filter are dynamically adjusted according to the estimation result.

[0112] It can be understood that the energy distribution (i.e. power spectral density) of the current ambient noise is calculated in real time to determine the intensity of the ambient noise at different frequencies, and the adaptive periodic filter coefficients are adjusted in real time. In the high ambient noise intensity, the filter suppression intensity is increased; in the weak ambient noise, the suppression is weakened to ensure that the useful signal is not lost.

[0113] In a more specific embodiment, the spectral subtraction method is used to eliminate the fixed frequency band noise first, and the residual noise is processed by the LMS algorithm.

[0114] It can be understood that the spectral subtraction method (frequency domain) is used for static noise (such as mechanical noise), and the spectral subtraction formula is used to eliminate the fixed frequency band noise. LMS algorithm (time domain): used for dynamic noise (such as airflow periodic change), real-time update of filter coefficients .

[0115] If the residual noise energy after spectral subtraction is lower than the threshold, LMS is skipped; if it is higher than the threshold, LMS is started to enhance the suppression. That is, spectral subtraction first "filters the big head": the frequency and energy of static noise (such as medical mechanical equipment) are relatively fixed, which belongs to "predictable" noise. Using spectral subtraction to directly subtract its frequency spectrum in the frequency domain can quickly and efficiently eliminate most of the fixed noise, which is equivalent to "cutting off" the most obvious noise source first.

[0116] LMS "cleans up the small tail" after: dynamic noise (such as medical staff conversation or foot sound) has real-time fluctuations in frequency and energy, which cannot be predicted in advance. The LMS algorithm can update the filter coefficients in real time in the time domain, accurately track and suppress this dynamic residual noise, which is equivalent to a second cleanup of "the fish that slips through the net". The combination of spectral subtraction and LMS algorithm can efficiently eliminate static noise first, and then accurately suppress dynamic noise with LMS algorithm, so as to achieve comprehensive noise cleaning while maximizing the retention of useful signals.

[0117] In a more specific embodiment, the signals of the de-noised first frequency band and the de-noised second frequency band are fused to obtain a final de-noised voice signal, which includes:

[0118] The error signals of the de-noised first frequency band and the de-noised second frequency band are calculated respectively, and the de-noising weights corresponding to the de-noised first frequency band and the de-noised second frequency band are adjusted according to the error signals to obtain new de-noised first frequency band and new de-noised second frequency band. The signals of the new de-noised first frequency band and the new de-noised second frequency band are fused to obtain a final de-noised voice signal.

[0119] It can be understood that in specific implementation, if the de-noised first frequency band or the first frequency band error is large and the residual noise is much, the de-noising weight is increased; if the error is large and the voice distortion is serious, the de-noising weight is reduced. By adjusting the weight, the de-noising degree of the frequency band with strong noise can be increased, and the de-noising weight of the frequency band with low noise energy can be reduced to reduce the de-noising degree, so as to avoid the "one-size-fits-all" processing with fixed parameters, so as to dynamically balance the noise suppression and voice fidelity in different frequency bands, and finally realize the optimal solution of "less noise residual and low voice distortion".

[0120] It should be noted that the frequency spectrum energy of the voice signal is unevenly distributed, and the noise in different frequency bands is completely different. In the first frequency band, the main frequency of the voice is the pitch frequency, and the noise mainly includes airflow noise, which is periodic and stable environmental noise. Excessive de-noising can easily lead to distortion of the voice, so if the first frequency band error is large, the de-noising weight is preferably reduced; the second frequency band includes complex environmental noise of burst or irregular noise, and insufficient de-noising will leave obvious noise, so if the first frequency band error is large, the de-noising weight is preferably increased.

[0121] In a more specific embodiment, the weight distribution includes:

[0122] The error signals are calculated for the two noises respectively, and the filter weights are distributed by a weighted least mean square (WLMS) algorithm to preferentially suppress the frequency band that has greater influence on the MPT / MCA calculation.

[0123] In a more specific embodiment, the acoustic features include: short-time energy, pitch period and zero-crossing rate, and the vocalization conditions include: maximum vocalization duration and maximum counting ability.

[0124] The vocalization conditions are obtained according to the acoustic features, including:

[0125] The maximum vocalization duration is obtained by analyzing the short-time energy and the pitch period, and the maximum counting ability is obtained by analyzing the short-time energy and the zero-crossing rate.

[0126] It can be understood that in specific implementation, the embodiment obtains the maximum phonation time (MPT) and the maximum counting ability (MCA) of the patient's voice signal, and transmits the MPT and MCA data to an external monitoring device or other user terminal to display to medical personnel, thereby assisting in voice rehabilitation evaluation: for example, the maximum vocalization duration of a normal adult male is ≥24.8s, and that of an adult female is ≥16.7s; the maximum counting ability of a normal adult male is ≥18s, and that of an adult female is ≥12s. If the monitored value is less than the normal value, it indicates that the patient's vocal function has certain obstacles, and the corresponding vocal function training can be given adaptively.

[0127] It should be noted that the feature extraction algorithm based on the speaking valve is specifically optimized for the vocalization characteristics of tracheostomy patients.

[0128] The vocalization characteristics of tracheostomy patients are:

[0129] 1) Weak voice: due to airflow bypassing the tracheostomy, the vocal cord vibration energy is low, and the signal-to-noise ratio is poor.

[0130] 2) Unstable fundamental frequency: incomplete vocal cord closure leads to large pitch fluctuations (±50 Hz).

[0131] 3) High frequency missing: airflow bypasses the oral cavity, loses oral cavity resonance, and high frequency (>2000 Hz) energy is significantly attenuated.

[0132] It should be noted that the acoustic features include: short-time energy, pitch period and zero-crossing rate, and the specific calculation method is as follows:

[0133] Short-Time Energy (STE): Short-time energy is a measure of signal energy intensity, reflecting the degree of signal change. In speech signals, parts with higher energy are usually related to pronunciation and speech activity.

[0134] Short-time energy E(t) calculation formula:

[0135]

[0136] Where x(t+n) represents the sample value of the signal in the time window [t,t+N], and N is the window length.

[0137] 2) Zero-Crossing Rate (ZCR): Zero-crossing rate refers to the number of times the signal crosses zero in a unit of time, which is an indicator of the smoothness of the speech signal. There is a significant difference in zero-crossing rate between the silent part (silence) and the voiced part (pronunciation).

[0138] Zero-crossing rate ZCR(t) calculation formula:

[0139]

[0140] Where sign(x) returns the sign function of the signal, i.e. x>0 returns 1, x<0 returns -1, and zero-crossing is when the sign of two consecutive samples is different.

[0141] 3) Pitch Period: Pitch period is the periodicity characteristic of speech signal, usually related to pronunciation. Calculating the pitch period can help evaluate the stability of pronunciation. Calculation method: Pitch period is usually estimated by the autocorrelation function (ACF), i.e. calculating the similarity between the signal and its delayed version, and estimating the period through its peak value.

[0142] Pitch period R x (t) calculation formula:

[0143]

[0144] In addition, the MPT and MCA calculation methods are as follows:

[0145] 1) Maximum Pronunciation Time (MPT): Maximum pronunciation time refers to the longest time the patient can sustain pronunciation, which reflects the patient's speech duration ability. By analyzing short-time energy (STE) and pitch, the duration of speech can be detected.

[0146] Maximum pronunciation time (MPT) calculation process:

[0147] Short-time energy (STE) and pitch are used to detect the active part of the speech signal;

[0148] The duration of the speech signal from start to end is calculated;

[0149] If the energy of the signal is below a set threshold (e.g. a silent segment), the timer is stopped and the maximum phonation time (MPT) is obtained.

[0150] Formula: MPT = max(duration of sustained sound),

[0151] Where the duration of sustained sound is determined by the part of the signal where the short-time energy is above a certain threshold and the pitch is stable.

[0152] It can be appreciated that MPT is used to capture phonatory defects of decreased phonatory endurance, automatically segmenting the sustained part by double thresholds of short-time energy and pitch, and has the clinical significance of directly reflecting the lung-glottal coupling efficiency and guiding the respiratory muscle training.

[0153] 2) Maximum counting ability (MCA): Maximum counting ability refers to the longest time that a patient can count continuously. This value is usually related to the rhythm, sustainability and phonatory ability of speech. The duration of each number pronunciation can be determined by the changes in short-time energy and zero-crossing rate.

[0154] Calculation process:

[0155] Different number pronunciations are separated according to short-time energy and zero-crossing rate;

[0156] The longest duration of consecutive number pronunciations is calculated.

[0157] Formula: MCA = max(duration of counting phase),

[0158] The duration of counting phase is the duration of the counting phase, and by short-time energy E(t) and zero-crossing rate ZCR(t), the speech active segment of the patient when counting can be identified, so as to calculate the maximum counting time.

[0159] It can be appreciated that MCA is used to capture phonatory-breathing coordination disorders, automatically segmenting syllables by the interval between peaks of short-time energy and zero-crossing rate, and has the clinical significance of revealing phonatory-breathing synchronization defects and setting targets for rhythm training.

[0160] In a more specific embodiment, the speech acquisition module is a high-sensitivity microphone.

[0161] In a more specific embodiment, the voice signal is pre-processed before noise reduction, specifically:

[0162] When the patient speaks, the sound waves generated by the vibration of the vocal cords are transmitted to the microphone through the upper respiratory tract. The microphone converts the sound pressure signal into an electrical signal, which is amplified by the preamplifier and then sampled by the analog-to-digital converter (ADC) to obtain a digital audio signal. In order to improve the signal-to-noise ratio, the system performs noise reduction on the collected voice signal, for example, by digital filtering to remove the interference of environmental noise and respiratory airflow noise.

[0163] The embodiment of the application also provides a processing system for the above-mentioned voice processing module 42, which comprises a processor and a signal transmitter 5, the input end of the processor is connected with the voice collection module 41, the output end of the processor is connected with the input end of the signal transmitter 5, and the output end of the signal transmitter 5 is connected with the user terminal.

[0164] It can be understood that, in specific implementation, the pronunciation situation is transmitted to the user terminal through the signal transmitter 5 by the processor, to assist medical staff in evaluating the pronunciation situation of the patient.

[0165] It should be noted that the above-mentioned sensors and modules are connected to the built-in microprocessor or control unit, and real-time data acquisition, processing and storage are performed. The system can display relevant parameters and voice / vocal cord data, so as to facilitate medical staff to comprehensively evaluate and guide rehabilitation. The device has compact structure, adopts low-power electronic components and rechargeable power supply design, is suitable for bedside use, and reduces cost.

[0166] In a more specific embodiment, as shown in Figure 1 the vocal cord vibration detection module is connected with the processor.

[0167] The vocal cord vibration detection module is attached to the neck to capture the patient's vocal cord vibration characteristics, detect the vocal cord vibration frequency and intensity, evaluate the vocal cord activity, understand the patient's swallowing function, and transmit the data to the integrated control module or external monitoring device to display to medical personnel. The laryngeal vibration sensor 6 hardware consists of a sensing and controller, connected by a serpentine interconnect. The sensing part is located above the Adam's apple, with a size of 0.7 cm (length) x 0.9 cm (width) x 0.3 cm (thickness). This position can capture high-quality sound signals because it is close to the sound source and can obtain clearer laryngeal vibration information. According to previous high-quality literature, the laryngeal vibration sensor 6 performs well in dynamic motion and complex noise environments, can accurately detect laryngeal events, and is not affected by background noise, with high practicality and reliability. For example, the characteristic of normal swallowing is the presence of multiple sharp peaks within a relatively short 0.2 seconds. If a patient is observed to have a continuous low-frequency component during speech and swallowing, it can be determined that the patient's vocal cord activity is impaired and may have problems with swallowing function. Then, based on the data collected above, the patient can be given corresponding vocal cord activity training to enhance the patient's swallowing function.

[0168] In a more specific embodiment, a rotating turbine 7 is provided in the respiratory passage 31, and the rotating turbine 7 is connected to the processor. Figure 1

[0169] In a more specific embodiment, a high-precision flow rate sensor (such as a rotating turbine 7) is built into the airflow passage 3 of the speaking valve, which detects and records the patient's peak inspiratory flow rate signal in real time, and transmits the data to the integrated control module (MCU) or external monitoring device for evaluation of respiratory capacity and adjustment of gas supply.

[0170] Measurement principle: When the patient coughs or exhales forcefully, the high-speed airflow passes through the passage of the speaking valve, and the blades of the rotating turbine 7 sensor rotate. The rotational speed of the rotating turbine 7 is proportional to the airflow speed, and the airflow speed can be calculated by detecting the rotational speed of the turbine 7 through a photoelectric or magneto-electric sensor. For example, the rotating turbine 7 generates a certain number of pulse signals per revolution, and the number of pulses per unit time represents the flow rate. This rotating turbine 7 sensor has high sensitivity and linearity, and can capture the peak flow rate at the moment of coughing in real time. For example, the normal inspiratory peak flow rate is 30-40 L / min, and if the patient's actual inspiratory flow rate is 45 L / min, the high-flow oxygen required for adjustment should be > 45 L / min to ensure stable oxygen supply.

[0171] The algorithm of the process mainly includes signal acquisition, filtering, peak extraction, and data transmission.

[0172] ​(1) Flow signal acquisition

[0173] First, the flow sensor converts the air flow speed into an electrical signal, generating a signal Q(t) that varies over time, where t is time and Q(t) is the instantaneous speed of the air flow. The electrical signal output by the sensor is typically based on turbine speed or differential pressure measurement.

[0174] (2) Signal preprocessing (filtering)

[0175] In order to remove possible noise or non-physiological signals (such as device errors or environmental interference), the original signal can be filtered. A band-pass filter is used to process the signal, retaining the effective frequency range:

[0176] Q filtered(t) = filter(Q(t)), where filter represents the band-pass filtering operation, removing high-frequency noise and low-frequency interference.

[0177] (3) Peak extraction

[0178] From the filtered signal Qfiltered(t), the flow sensor calculates the peak inspiratory flow rate by measuring the instantaneous change in air flow. The algorithm is usually based on integrating or linearly fitting the flow rate over time to obtain the peak flow rate. The specific algorithm is:

[0179]

[0180] tpeak ∈ [t0, t1], where Vpeak represents the peak inspiratory flow rate, Qt represents the air flow rate, tpeak is the time point corresponding to the peak, and the interval [t0, t1] represents the effective inspiratory time range.

[0181] (4) Peak inspiratory flow rate evaluation and feedback

[0182] Based on the detected peak inspiratory flow rate Vpeak, it is determined whether the patient needs to adjust the oxygen supply. For example, if the sensor detects that Vpeak is greater than 30 L / min (normal range), the oxygen flow rate can be adjusted to be greater than 30 L / min to ensure that the patient always receives sufficient air supply.

[0183] As shown in Figure 2 , the embodiment of the present application provides a process for a speaking valve. The process integrates the working methods of the speech acquisition and processing component, the laryngeal vibration sensor, the turbine, and the oxygen inlet interface device, and the specific content is as follows:

[0184] System initialization setting:

[0185] The MCU performs self-checking, the sensors are zeroed, and the interface module detects the HFNC connection state.

[0186] 1. Inhale phase (0ms - 500ms)

[0187] Parallel data stream:

[0188] Flow rate sensor outputs Q(t) in real time;

[0189] MCU calculates Vpeak every 50ms;

[0190] If Vpeak is greater than the threshold, the MCU immediately prompts to increase the HFNC flow to ensure stable oxygen supply.

[0191] 2. Voice phase (500ms - 3000ms, triggered by the user)

[0192] Passage: dual-channel valve body is used;

[0193] Inhale passage - for patient to inhale;

[0194] Voice passage - for vocal cord vibration;

[0195] Parallel data: let the patient speak and say "ah" for as long as possible;

[0196] Voice module:

[0197] Microphone collects the patient's voice data and transmits it to the central processor. The central processor performs noise reduction processing (LMS) according to the voice characteristics of different patients, further extracts features (STE / ZCR / MFCC) and evaluates MPT / MCA, and then delivers the processed data to the MCU for matching with normal values in the model library to display the patient's pronunciation;

[0198] Laryngeal vibration sensor:

[0199] Collects data on the patient's laryngeal vibration during speech and transmits it to the central processor for processing, matching with normal values in the model library, swallowing pattern recognition, and then delivery to the MCU;

[0200] Synergy point:

[0201] If the choking algorithm marks "choking risk", the MCU suspends the MPT timer and sends instructions to the terminal within 200ms

safety interruption

[0202] If MPT is less than the normal value, the MCU pushes a training prompt to the terminal after the cycle ends.

[0203] 3. Exhale / swallowing phase (may overlap)

[0204] Flow rate sensor detects exhale flow rate for next inhale prediction;

[0205] The vibration sensor continues to monitor the peak within the 0.2s window to confirm the completion of swallowing;

[0206] The MCU summarizes the data of this round: Vpeak, MPT, MCA, swallowing score, and then packs and uploads.

[0207] 4. End of cycle and abnormal handling:

[0208] If any sensor is lost, the MCU immediately closes the analysis, keeps the oxygen supply, and performs audible and light alarms.

[0209] The above, the above examples are only used to illustrate the technical solutions of the present application, but not limit it; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing examples, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A speech processing method based on speech valves, characterized in that, A voice acquisition module is installed on the outer cover of the speaking valve, and the processing method includes: The voice signal acquired by the voice acquisition module is divided into multiple frequency bands, and each frequency band is subjected to independent noise reduction processing: The multiple frequency bands include a first frequency band and a second frequency band. The first frequency band includes airflow noise, and the second frequency band includes ambient noise. The frequency of the first frequency band is lower than the frequency of the second frequency band. The airflow noise of the first frequency band and the ambient noise of the second frequency band are removed respectively to obtain a denoised first frequency band and a denoised second frequency band. The signals from the first and second denoised frequency bands are fused to obtain the final denoised speech signal. Acoustic features are extracted from the final denoised speech signal, and the pronunciation is obtained based on the acoustic features.

2. The speech processing method based on speaking valves according to claim 1, characterized in that, The removal of airflow noise in the first frequency band includes: The denoised first frequency band is obtained by periodically removing airflow noise from the signal in the first frequency band using an adaptive periodic filter.

3. The speech processing method based on speaking valves according to claim 2, characterized in that, Before obtaining the denoised first frequency band by periodically removing airflow noise from the signal in the first frequency band using an adaptive periodic filter, the method further includes: Airflow noise below a preset frequency in the first frequency band is removed by using a bandpass filter.

4. The speech processing method based on speaking valves according to claim 2 or 3, characterized in that, The step of periodically removing airflow noise from the signal in the first frequency band using an adaptive periodic filter to obtain a denoised first frequency band includes: Periodic modeling of airflow noise includes predicting changes in airflow noise by detecting periodic variations in breathing. Based on the changes in airflow noise, the adaptive periodic filter is used to periodically remove airflow noise from the signal in the first frequency band.

5. The speech processing method based on speaking valves according to claim 4, characterized in that, The step of adjusting the adaptive periodic filter according to the airflow noise change further includes: An adaptive filtering algorithm is used to estimate the power spectral density of airflow noise, and the coefficients of the adaptive periodic filter are dynamically adjusted based on the estimation results.

6. The speech processing method based on speaking valves according to claim 5, characterized in that, The step of adjusting the adaptive periodic filter according to the airflow noise change further includes: The weights of the adaptive periodic filter are dynamically adjusted using a weighted sliding window method.

7. The speech processing method based on speaking valves according to claim 1, characterized in that, The removal of environmental noise in the second frequency band includes: The spectral subtraction method is used to estimate environmental noise and track changes in environmental noise in real time. The coefficients of the adaptive filter are adjusted for the first time based on the changes in ambient noise. The ambient noise in the second frequency band is removed by the adaptive filter after the first adjustment to obtain the preliminary denoised second frequency band. An adaptive filtering algorithm is used to estimate the power spectral density of airflow noise, and the coefficients of the adaptive filter are dynamically adjusted a second time based on the estimation results. The environmental noise in the second frequency band is removed by the second-adjusted adaptive filter to obtain the final denoised second frequency band.

8. The speech processing method based on speaking valves according to claim 1, characterized in that, The signals from the first and second denoised frequency bands are fused to obtain the final denoised speech signal, which includes: Error signals are calculated for the first denoised frequency band and the second denoised frequency band respectively. Based on the error signals, the denoising weights corresponding to the first denoised frequency band and the second denoised frequency band are adjusted to obtain a new first denoised frequency band and a new second denoised frequency band. The signals of the new first denoised frequency band and the new second denoised frequency band are fused to obtain the final denoised speech signal.

9. The speech processing method based on speaking valves according to claim 1, characterized in that, The acoustic characteristics include: short-time energy, fundamental period, and zero crossover rate; the articulation conditions include: maximum articulation duration and maximum counting ability. The step of obtaining the pronunciation information based on the acoustic features includes: The maximum vocal duration is obtained by analyzing short-time energy and fundamental frequency, and the maximum counting ability is obtained by analyzing short-time energy and zero crossover rate.

10. A speaking valve, characterized in that, include: Outer casing and voice acquisition and processing components; One end of the outer cover is provided with a one-way valve, and the other end is provided with an airflow channel. The airflow channel includes a breathing channel and an oxygen channel. The breathing channel is connected to the one-way valve. The middle part of the outer cover is provided with an oxygen inlet port, and the oxygen channel is connected to the oxygen inlet port. The voice acquisition and processing component includes a voice acquisition module and a voice processing module connected to each other. The voice acquisition module is disposed on the surface of the outer cover and close to the one-way valve. The voice processing module is used to acquire the voice signal acquired by the voice acquisition module and execute the voice processing method according to any one of claims 1-9.

Images