Signal processing method applied to fully implanted bone conduction hearing aid device

Abstract

The invention discloses a signal processing method applied to a fully-implanted bone conduction hearing aid device, and relates to the technical field of signal processing, and the method comprises the steps: firstly introducing a synchronous sampling clock to control a bone conduction microphone to collect a mixed signal containing heartbeat, pulse, vibrator vibration and voice, carrying out the separation and extraction of a two-pass filtering module, and then carrying out the processing of the mixed signal; and activating the oscillator-microphone interference model and inputting an oscillator driving signal, carrying out differential interference suppression on heartbeat and pulse signals, and finally enhancing the extracted signal and compensating and outputting a voice signal in combination with the oscillator driving signal. According to the invention, the technical problem that the mixed signal collected by the embedded bone conduction microphone is affected by interference and the interference is difficult to accurately separate and suppress to extract the high-quality voice signal is solved, and the accurate separation and interference suppression of the mixed signal are achieved. The technical effect of effectively improving the extraction quality and definition of the voice signal in the fully-implanted bone conduction hearing aid device is achieved.

Description

Technical Field

[0001] The present invention relates to the field of signal processing technology, and in particular to a signal processing method applied to a fully implantable bone conduction hearing aid device. Background Art

[0002] The signals collected by embedded bone conduction microphones often contain mixed signals such as heartbeat, pulse, vibrator vibration and voice. In the existing technology, traditional signal processing methods have difficulty in effectively separating and suppressing interference signals when faced with such mixed signals, resulting in poor voice signal quality. The inventors found that traditional methods are unable to accurately handle various types of interference in mixed signals, such as the interference of vibrator vibration on voice signals, the mixing of physiological signals and voice signals, etc., because they do not design special processing procedures for the special working environment and signal characteristics of implantable devices. As a result, the extracted voice signals are not clear and intelligible enough, making it difficult to meet the requirements of hearing aids for high-quality voice output. Summary of the Invention

[0003] This application solves the technical problem that the mixed signal collected by the embedded bone conduction microphone is affected by interference, making it difficult to accurately separate and suppress interference to extract high-quality voice signals. This application introduces a synchronous sampling clock to collect mixed signals, including heartbeat, pulse, oscillator vibration, and voice signals. Through dual-pass filtering separation, oscillator-microphone interference model interference suppression, signal enhancement and voice compensation, etc., this application constructs a multi-scale time delay prediction and dynamic feedback mechanism, thereby accurately separating the various components in the mixed signal, effectively improving the voice signal extraction quality and anti-interference ability in fully implantable bone conduction hearing aid devices.

[0004] In response to the above technical problems, the present application proposes a signal processing method for a fully implantable bone conduction hearing aid device, wherein the method includes: introducing a synchronous sampling clock, controlling the embedded bone conduction microphone to perform signal acquisition, and establishing a signal data set, wherein the signal data set is a mixed signal set including a heartbeat signal, a pulse signal, a vibrator vibration signal, and a voice signal; after configuring a dual-pass filtering module, sending the signal data set to the dual-pass filtering module, performing signal separation and extraction using the dual-pass filtering channel, and establishing a separation and extraction result; activating the vibrator-microphone interference model, and using the vibrator drive signal in the separation and extraction result as an input feature, inputting it into the vibrator-microphone interference model; performing differential interference suppression on the heartbeat signal and pulse signal in the separation and extraction result according to the output result, and establishing an extracted signal; after performing signal enhancement on the extracted signal, using the signal enhancement result and the vibrator drive signal to compensate for the voice signal and then output it.

[0005] This application proposes one or more technical solutions, which have at least the following technical effects:

[0006] This application introduces a synchronous sampling clock to control the embedded bone conduction microphone to collect mixed signals (including heartbeat, pulse, oscillator vibration and voice signals), and constructs an adaptive dynamic feedback mechanism through separation by a dual-pass filtering module, interference suppression by a oscillator-microphone interference model, multi-scale time delay differential processing and signal enhancement compensation, so as to accurately separate the effective components in the mixed signal and realize high-quality extraction and anti-interference optimization of voice signals in a fully implantable bone conduction hearing aid device. It achieves the technical effect of accurate separation and interference suppression of mixed signals, and effectively improves the extraction quality and clarity of voice signals in a fully implantable bone conduction hearing aid device. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0008] Figure 1 1 is a flow chart of a signal processing method for a fully implantable bone conduction hearing aid provided in an embodiment of the present application.

[0009] Figure 2 This is a flow chart of constructing an interference model in a signal processing method for a fully implantable bone conduction hearing aid provided in an embodiment of the present application. DETAILED DESCRIPTION

[0010] The present application provides a signal processing method for a fully implantable bone conduction hearing aid device, which is used to solve the technical problem that the mixed signal collected by the embedded bone conduction microphone is affected by interference, making it difficult to accurately separate and suppress interference to extract high-quality voice signals.

[0011] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only some of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0012] It should be noted that any variations of the terms "include" and "have" are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or are inherent to these processes, methods, products or devices.

[0013] like Figure 1As shown, a signal processing method applied to a fully implantable bone conduction hearing aid device, wherein the method includes:

[0014] Step A100: Introduce a synchronous sampling clock to control the embedded bone conduction microphone to perform signal acquisition and establish a signal data set. The signal data set is a mixed signal set including a heartbeat signal, a pulse signal, a vibrator vibration signal and a voice signal.

[0015] In the embodiment of the present application, the synchronous sampling clock is a clock source used to uniformly control the timing of signal acquisition.

[0016] Specifically, first, a synchronous sampling clock is introduced, such as a 44.1kHz fixed-frequency crystal oscillator, to synchronously trigger the analog-to-digital conversion process of the embedded bone conduction microphone with clock pulses. When the microphone is close to the surface of the skull, the synchronous clock controls it to continuously collect acoustic vibration signals at equal time intervals. The heartbeat signal generates periodic mechanical vibrations due to heartbeats, the pulse signal forms pressure fluctuations accompanied by vasoconstriction and relaxation, the oscillator vibration signal is driven by the oscillator inside the hearing aid device to generate regular mechanical waves, and the voice signal is conducted to the microphone through the skull. After sampling, these signals form a time domain mixed signal set. For example, the heartbeat signal amplitude is about 0.1-0.3mV, the pulse signal amplitude is 0.2-0.5mV, the oscillator vibration signal amplitude is 0.5-1.2mV, and the voice signal amplitude is 0.3-0.8mV. Each signal presents different periodic characteristics in the time domain.

[0017] After clock synchronization, the collected mixed signal's frequency components accurately correspond to the inherent frequency bands of each signal: heartbeat / pulse signals are concentrated in the low-frequency range of 0.5-3 Hz, oscillator vibration signals are primarily distributed in the mid-frequency range of 50-500 Hz, and speech signals are concentrated in the mid- and high-frequency range of 300-3400 Hz. This dataset provides a consistent temporal sequence and clear spectral characteristics for subsequent dual-pass filtering and separation.

[0018] By introducing a synchronous sampling clock to control the timing of the embedded bone conduction microphone, the acquisition of a mixed data set containing heartbeat, pulse, oscillator vibration and voice signals is achieved, ensuring the signal time domain synchronization and spectral accuracy, and providing high-quality data input for subsequent processing steps such as dual-pass filtering separation and interference suppression.

[0019] Step A200: After configuring the dual-pass filtering module, the signal data set is sent to the dual-pass filtering module, and signal separation and extraction are performed using the dual-pass filtering channel to establish a separation and extraction result.

[0020] In the embodiment of the present application, the dual-pass filtering module is a signal processing unit integrating high-pass and low-pass parallel channels.

[0021] Optionally, when configuring a dual-pass filtering module, parallel high-pass and low-pass filtering channels are integrated. The high-pass channel uses a second-order Butterworth filter with a cutoff frequency of 50 Hz, constructed in hardware using an operational amplifier and an RC network. The low-pass channel uses a fourth-order Chebyshev filter with a cutoff frequency of 3 Hz, achieving frequency characteristics through a multi-stage LC filter network. After module configuration is complete, a mixed-signal data set, controlled by a synchronous sampling clock, is transmitted to the module input via the SPI bus, where the internal multiplexer distributes the signals by channel.

[0022] When the signal enters the high-pass channel, speech signals above 50Hz are output from the first output terminal, with a measured amplitude of 0.3-0.8mV; oscillator vibration signals between 50-500Hz are output from the second output terminal, with an amplitude of 0.5-1.2mV. During low-pass channel processing, heartbeat and pulse signals below 3Hz are output from the first and second output terminals, with amplitudes of 0.1-0.3mV and 0.2-0.5mV, respectively.

[0023] Next, activate the high-pass filter channel and the low-pass filter channel in the dual-pass filter channel, separate the voice signal and the vibrator vibration signal from the first and second output ends of the high-pass channel, and extract the heartbeat signal and the pulse signal from the first and second output ends of the low-pass channel. The specific steps are described in detail in A210-A220.

[0024] Through the collaborative processing of parallel high-pass and low-pass channels, the voice, oscillator vibration, heartbeat, and pulse signals in the mixed signal set are accurately separated by frequency band, and four independent signals are output, providing pure input signals for subsequent oscillator interference model suppression, signal enhancement and other steps, and solving the interference problem caused by the overlap of mixed signal frequency bands from the hardware filtering level.

[0025] Step A300: activating a vibrator-microphone interference model, and using the vibrator driving signal in the separation and extraction result as an input feature to input the vibrator-microphone interference model.

[0026] In one embodiment of the present application, first, the vibrator drive signal (50-500Hz, amplitude 0.5-1.2mV) separated by double-pass filtering is preprocessed: converted to the time-frequency domain by short-time Fourier transform (STFT), with a frame length of 25ms and a frame shift of 10ms, to obtain a 256-dimensional complex feature vector. The vector is input into the initialized LSTM interference model, which contains 3 hidden layers and 128 neurons in each layer. The model dynamically adjusts the weight according to the individual transfer characteristics (the skull frequency domain response obtained in the guided training phase) and the environmental feedback window (100ms sliding window). During the forward propagation of the model, the attention mechanism automatically assigns weights to each frequency band: in a noisy environment, such as when the ambient sound energy is >80dB, the weight of the 300-2000Hz voice band is increased by 30%, while the suppression of the 50-200Hz vibrator low-frequency interference is enhanced.

[0027] The model output is a predicted oscillator interference signal (time domain waveform), whose delay parameters are dynamically adjusted via a multi-scale delay prediction module: a sliding search is performed within the 1-50ms range with a 1ms step size, the cross-correlation function between the drive signal and the microphone signal is calculated, and the delay corresponding to the peak point is used as the optimal estimate. The predicted interference signal and the mixed signal collected by the microphone are adaptively canceled. The adaptive algorithm uses a normalized LMS with a step size of μ = 0.05 and converges after 200 iterations. The final output speech signal is bandpass filtered (300-3400Hz) and gain compensated (dynamic range compression ratio 2:1), improving the signal-to-noise ratio and speech clarity index.

[0028] By activating the LSTM-based vibrator-microphone interference model, the separated and extracted vibrator driving signal is converted into a time-frequency feature vector input model. The prediction parameters are dynamically adjusted in combination with individual transfer characteristics and environmental feedback. After multi-scale time delay estimation and adaptive cancellation, the vibrator interference is accurately suppressed. The output result is a speech signal with significantly improved purity, which effectively solves the problem of interference of vibrator vibration on speech extraction in fully implantable bone conduction hearing aids.

[0029] Step A400: performing differential interference suppression on the heartbeat signal and the pulse signal in the separation and extraction result according to the output result to establish an extraction signal.

[0030] Specifically, a multi-scale time delay prediction window is constructed to identify the transmission delay between the oscillator drive signal and the heartbeat and pulse signals. The predicted interference signal is time-aligned based on the delay parameters and a differential operation is performed to achieve targeted suppression of the oscillator low-frequency interference in the physiological signal. The specific steps are described in detail in A410-A420.

[0031] After differential interference suppression, the following steps are performed to create the extracted signal: First, the differentially processed heartbeat and pulse signals are bandpass filtered using a fourth-order Butterworth filter. The filter frequency range is set to 0.5-2Hz for the heartbeat signal and 0.7-3Hz for the pulse signal, effectively blocking the residual frequency band of the oscillator interference. The filtered signals are converted to digital signals using an ADC, and the amplitude range is normalized to the range [-1, 1] to eliminate amplitude fluctuations caused by individual differences.

[0032] Subsequently, dynamic range compression (compression ratio 2:1) was performed on the normalized signal, compressing the dynamic range of the heartbeat signal from 20dB to 10dB, and the pulse signal from 25dB to 12.5dB. This prevents distortion of strong signals while improving the discernibility of weak signals. Observation with an oscilloscope showed that the time domain waveform smoothness of the compressed signal was significantly improved, and the heartbeat cycle and pulse fluctuation characteristics were clearly discernible, providing a low-noise, highly discernible extracted signal for subsequent signal enhancement.

[0033] By accurately defining the frequency band of physiological signals through band-pass filtering, unifying the amplitude range through normalization processing, and optimizing the dynamic characteristics of signals through dynamic range compression, the heartbeat and pulse signals are further purified based on differential interference suppression, laying a high-quality data foundation for subsequent processing steps such as waveform enhancement.

[0034] Step A500: After performing signal enhancement on the extracted signal, the speech signal is compensated using the signal enhancement result and the oscillator driving signal and then output.

[0035] Optionally, first, the heartbeat signal and pulse signal in the extraction results are band-pass filtered respectively, the band-pass filtering results are normalized and then dynamic range compression is performed, and then the signal is smoothed using a sliding window and then waveform enhancement is performed to enhance the extracted signal. The specific steps are described in detail in A510-A530.

[0036] Next, speech compensation is performed, and the enhanced physiological signal and the oscillator drive signal (50-500Hz, amplitude 0.5-1.2mV) are input into the compensation model: the physiological signal is used to identify the user's physiological state (such as resting / exercise) to adjust the compensation strategy, and the oscillator drive signal is used to predict the oscillator interference waveform through the LSTM model. For example, when the ambient sound energy is >80dB, the model automatically increases the gain of the 300-2000Hz voice band by 30%, while suppressing the 50-200Hz oscillator interference. After adaptive filtering and cancellation, the speech SNR is further improved to 28-32dB, and the speech clarity index (RASTI) is improved from 0.45 to 0.82. After 200 iterations of training, the output compensated speech signal has a word recognition rate of over 92% in a 65dB white noise environment.

[0037] By extracting high-purity physiological features through signal enhancement and combining them with the oscillator driving signal to build a dynamic compensation model, we can achieve precise cancellation of oscillator interference in the speech signal and optimize the frequency band gain. Compared with traditional methods, this method improves the speech SNR and word recognition rate, and effectively solves the impact of physiological noise and oscillator interference on speech quality in fully implantable devices.

[0038] Furthermore, step A200 in the method provided in the embodiment of the present application includes:

[0039] A210: Activate the high-pass filter channel in the dual-pass filter channel, use the high-pass filter channel to perform signal separation on the signal data set, output the voice signal from the first output end of the high-pass filter channel, and output the vibrator vibration signal from the second output end.

[0040] A220: Activate the low-pass filter channel in the dual-pass filter channel, use the low-pass filter channel to perform signal analysis on the signal data set, output a heartbeat signal from a first output end of the low-pass filter channel, and output a pulse signal from a second output end.

[0041] Specifically, when activating the dual-pass filter channel, the high-pass filter channel is first activated through the hardware control circuit: this channel uses a Butterworth second-order filter circuit, and the cutoff frequency is set to 50Hz. When the mixed signal (digital signal after 44.1kHz sampling) is input, the voice signal above 50Hz (300-3400Hz) is output from the first output end, and the measured time domain amplitude is 0.3-0.8mV, while the vibrator vibration signal of 50-500Hz is output from the second output end, with an amplitude of 0.5-1.2mV.

[0042] The low-pass filter channel is simultaneously activated, using a fourth-order Chebyshev filter network with a cutoff frequency of 3Hz. After analyzing the mixed signal, heartbeat signals below 3Hz (0.5-2Hz) are output from the first output terminal (amplitude 0.1-0.3mV), and pulse signals (0.7-3Hz) are output from the second output terminal (amplitude 0.2-0.5mV). Oscilloscope testing shows that the voice signal output by the high-pass channel has no overlap with the time domain waveform of the oscillator vibration signal, and the low-frequency components of the heartbeat / pulse signal extracted by the low-pass channel have a purity of over 95%.

[0043] By activating the high-pass and low-pass filter channels in parallel and using filters with different cutoff frequencies to divide the mixed signal data set into frequency bands, the voice signal and oscillator vibration signal are separated from the first / second output end of the high-pass channel, and the heartbeat signal and pulse signal are extracted from the first / second output end of the low-pass channel, thus achieving precise frequency band separation and independent output of the mixed signal.

[0044] Further, if Figure 2 As shown, step A300 in the method provided in the embodiment of the present application includes:

[0045] A310: Capturing individual transfer properties during the guided training phase.

[0046] A320: After calling the universal vibrator-microphone interference model, the universal vibrator-microphone interference model is initialized and configured according to the transfer characteristics.

[0047] A330: Configuring an environment feedback window, collecting environment context variables in the environment feedback window, and using the environment context variables to compensate for the initialized general vibrator-microphone interference model to establish the vibrator-microphone interference model.

[0048] In the embodiment of the present application, individual transfer characteristics refer to differences in sound conduction characteristics caused by differences in skull structure (such as bone density and thickness) between different users. The environmental feedback window is a sliding time window.

[0049] Optionally, first, use the built-in three-axis accelerometer and pressure sensor of the hearing aid device to capture individual transfer characteristics: let the user emit a standard test tone, such as a 200-8000Hz sweep frequency signal, and simultaneously record the time domain response of the vibrator driving current and the microphone acquisition signal. After FFT conversion, the frequency domain transfer function of the individual skull is obtained, and its peak value difference in the 300-500Hz frequency band can reach ±12dB.

[0050] Next, the universal oscillator-microphone interference model is called. The model is a pre-trained LSTM network with three hidden layers and 128 neurons in each layer. The individual transfer function is used as the initial weight parameter to initialize the model so that the 256 nodes in the model input layer correspond one-to-one to the frequency domain feature points of the transfer function.

[0051] Next, an environmental feedback window is configured. This sliding window has a length of 100ms and a step size of 20ms. Environmental context variables are collected using ambient light sensors, air pressure sensors, and a microphone array. The rate of change of light intensity reflects whether the user is moving, air pressure fluctuations indicate ambient noise levels, and the spectral characteristics of ambient sound distinguish between quiet and noisy scenes. These variables are quantized to form a 16-dimensional feature vector, which is used to compensate for the initialized model. In noisy environments (i.e., ambient sound energy >80dB), the model's attention mechanism increases the weight of the speech signal by 30% and enhances interference suppression in the 500-2000Hz frequency band. In quiet environments, the high-frequency gain is reduced. The model is trained using an adaptive gradient descent algorithm with an initial learning rate of 0.001 and weight updates every 500ms of signal processing. The training objective is to minimize the mean squared error between the predicted and actual interference. After 200 iterations, the error converges to below 0.05mV.

[0052] By capturing the individual transfer characteristics (cranial frequency domain response) in the guided training stage and initializing the general interference model, combined with the context variables collected by the environmental feedback window for dynamic compensation, a vibrator-microphone interference model that can adapt to individual differences and environmental changes is constructed, the interference prediction error is reduced, the speech intelligibility is improved in noisy environments, and the interference suppression effect is significantly optimized.

[0053] Further, the step A400 in the method provided by the embodiment of the application comprises:

[0054] A410: constructing a multi-scale time delay prediction window according to the vibrator driving signal, performing mutual correlation peak point identification of the heartbeat signal and the pulse signal on the multi-scale time delay prediction window respectively, and establishing a transfer delay.

[0055] A420: performing differential interference suppression of the corresponding heartbeat signal and pulse signal respectively by using the transfer delay and the output result.

[0056] In the embodiment of the application, the time delay prediction window is a 1-50 ms sliding time window constructed based on the vibrator driving signal, which is used to segment the signal to calculate the time correlation of the vibrator driving signal and the heartbeat / pulse signal. The mutual correlation peak point is the time point with the highest correlation in the mutual correlation function of the vibrator driving signal and the physiological signal (heartbeat / pulse), which corresponds to the maximum value of the time domain matching degree of the two signals. The transfer delay is the conduction time difference of the vibrator driving signal to the physiological signal determined by the mutual correlation peak point.

[0057] Specifically, first, a multi-scale time delay prediction window is constructed according to the vibrator driving signal (50-500 Hz, time domain amplitude 0.5-1.2 mV), the window range is set to 1-50 ms, which can cover the maximum possible time delay of cranial conduction, and the signal segment is segmented by sliding with a step of 1 ms. The mutual correlation function of the heartbeat signal (0.5-2 Hz) and the pulse signal (0.7-3 Hz) in each window with the vibrator driving signal is calculated respectively, and the transfer delay is determined by finding the peak point. For example, when the mutual correlation coefficient of the heartbeat signal and the driving signal reaches the maximum value 0.85 at 12 ms, the transfer delay of the heartbeat signal is determined to be 12 ms; the mutual correlation coefficient of the pulse signal reaches 0.82 at 18 ms, and the delay is determined to be 18 ms. The process accelerates the mutual correlation calculation by FFT, and the single-frame processing time is controlled to be less than 5 ms.

[0058] Using the above transfer delay, the output result (predicted vibrator interference signal, time domain waveform) of the vibrator-microphone interference model is time-aligned according to the delay parameter, and the differential operation is performed on the heartbeat signal: heart_clean heart pred [t-τ heart ], where τ heart ​​is 12ms; the pulse signal is processed in the same way, τ pulse is 18ms. Among them, S heart_clean It is the pure heartbeat signal after differential interference suppression and removal of oscillator interference, which is the low-noise physiological signal that we ultimately want to obtain. heart It is the original collected heartbeat signal mixed with the interference of the vibrator, including the noise introduced by the vibrator vibration through skull conduction. pred is the oscillator interference signal predicted by the oscillator-microphone interference model. t is a time variable that represents the instantaneous moment of the signal in the time domain and is used to locate the time dimension coordinate of the signal. heart It is the transmission delay of the heartbeat signal (such as 12ms), that is, the time difference between the oscillator drive signal and the heartbeat signal acquisition path, which is obtained by identifying the cross-correlation peak point. pulse It is the transmission delay of the pulse signal (such as 18ms), which is used to align the timing of the pulse signal and the predicted interference signal to support the accuracy of the differential operation.

[0059] By constructing a multi-scale time delay prediction window and identifying the cross-correlation peak points, the transmission delay between the oscillator drive signal and the heartbeat and pulse signals is accurately established. After time alignment of the predicted interference signal based on the delay parameters, a differential operation is performed. This achieves targeted suppression of the oscillator's low-frequency interference in the physiological signal, reduces the residual noise of the heartbeat / pulse signal, and provides low-interference pure physiological signal input for the subsequent signal enhancement step.

[0060] Furthermore, step A500 in the method provided in the embodiment of the present application includes:

[0061] A510: Perform band-pass filtering on the heartbeat signal and the pulse signal in the extraction result respectively.

[0062] A520: After normalizing the bandpass filtering results, dynamic range compression is performed.

[0063] A530: Performs waveform enhancement after signal smoothing using a sliding window.

[0064] In the embodiment of the present application, bandpass filtering is a filtering method that allows signals in a specific frequency band to pass through while attenuating signals in other frequency bands, and is used to perform frequency band screening on heartbeat signals and pulse signals after differential interference suppression.

[0065] Specifically, first, when a fourth-order Butterworth bandpass filter is used to process the heartbeat and pulse signals, the flat amplitude-frequency response characteristics within the passband are utilized to set a passband frequency band of 0.5-2Hz for the heartbeat signal and a passband frequency band of 0.7-3Hz for the pulse signal. A recursive filtering algorithm is used to allow the target frequency band signal to pass without distortion. At the same time, noise components below the lower limit frequency, such as below 0.5Hz for the heartbeat signal and below 0.7Hz for the pulse signal, and above the upper limit frequency, such as above 2Hz for the heartbeat signal and above 3Hz for the pulse signal (such as low-frequency interference of the vibrator vibration) are attenuated. This filtering process can reduce the residual noise amplitude of the heartbeat signal and improve the signal-to-noise ratio and the signal-to-noise ratio of the pulse signal, thereby effectively realizing the frequency band separation of physiological signals and interference noise.

[0066] Subsequently, the filtering results are normalized to the amplitude range of [-1,1] to eliminate amplitude fluctuations caused by individual skull conduction differences. Dynamic range compression is then performed using a 2:1 compression ratio to compress the dynamic range of the heartbeat signal from 20dB to 10dB, and the pulse signal from 25dB to 12.5dB, thereby improving the recognizability of weak signals.

[0067] Finally, a 50ms Hanning sliding window is used to smooth the signal and reduce time domain fluctuations. On this basis, waveform enhancement is performed. Signal periods are established for the smoothed heartbeat signal and pulse signal respectively. Autocorrelation amplification is performed based on the period. Then, correlation alignment is performed between the two and correlation authentication is performed based on this. Finally, waveform enhancement is completed based on the correlation authentication results and the autocorrelation amplification processing results. The specific steps are described in detail in A531-A533.

[0068] By accurately defining the frequency band of physiological signals through band-pass filtering, normalizing the amplitude range to a unified value, optimizing the dynamic characteristics of the signal through dynamic range compression, and performing sliding window smoothing, the SNR of the extracted heartbeat and pulse signals is improved compared to traditional methods, and the residual noise is reduced, providing high-purity and high-recognition physiological signal input for subsequent voice signal compensation, effectively ensuring the signal processing accuracy of the fully implantable bone conduction hearing aid device.

[0069] Furthermore, step A530 in the method provided in the embodiment of the present application includes:

[0070] A531: establish signal cycles for the heartbeat signal and pulse signal after signal smoothing.

[0071] A532: Perform autocorrelation amplification processing on the signal according to the signal period.

[0072] A533: Perform correlation alignment of heartbeat signals and pulse signals, perform correlation authentication based on the correlation alignment results, and complete waveform enhancement using the correlation authentication results and autocorrelation amplification processing results.

[0073] Specifically, the researchers first determined the signal period of the heartbeat and pulse signals after sliding window smoothing. The researchers used the autocorrelation function to calculate the time domain peak intervals of the signals. For heartbeat signals, the maximum autocorrelation peak (correlation coefficient > 0.9) was identified within the 0.8-1.2 second range, determining its period as 1 second. For pulse signals, the period was determined to be 0.7 seconds within the 0.6-0.9 second range. This process leverages the periodicity of physiological signals and, verified by FFT spectrum analysis, achieved a period identification accuracy exceeding 95%.

[0074] Subsequently, autocorrelation amplification processing is performed according to the periodic parameters. Based on the time domain characteristics of the heartbeat signal (period 1 second) and the pulse signal (period 0.7 seconds), the correlation of the signals under different time delays is calculated through the autocorrelation function to determine the periodic peak position of the signal. In the specific operation, the amplitude of the autocorrelation function of the heartbeat signal is amplified by 2-3 times, and the pulse signal is amplified by 1.5-2 times, so that the periodic fluctuation amplitude of the heartbeat signal is increased from 0.1-0.3mV to 0.3-0.8mV, and the pulse signal is increased from 0.2-0.5mV to 0.5-1.0mV. This amplification process strengthens the periodic component of the signal and synchronously suppresses non-periodic noise, so that the noise amplitude is further reduced from 0.02-0.05mV to 0.01-0.03mV. After oscilloscope testing, the period stability of the amplified heartbeat signal is improved to more than 95%, and the waveform regularity of the pulse signal is significantly enhanced.

[0075] Next, the heartbeat and pulse signals are aligned. After determining through cross-correlation that the pulse signal lags behind the heartbeat signal by a physiological delay of approximately 0.15 seconds, the two are time-aligned: using the starting point of the heartbeat signal's cycle as a reference, the pulse signal is shifted along the time axis by 0.15 seconds. The dynamic time warping (DTW) algorithm is then used to perform nonlinear matching on the waveform to address the cycle offset problem caused by individual differences. During the association authentication phase, the cross-correlation coefficient between the two is calculated to be above 0.85 (the standard threshold under healthy physiological conditions is 0.8), confirming that it meets the physiological correlation characteristic that the pulse fluctuation lags behind the heartbeat. This alignment process not only eliminates timing deviations but also enhances the correlation between the signals through joint optimization, making the composite features of the heartbeat and pulse signals more prominent, providing high-precision physiological signal input for subsequent speech compensation.

[0076] By establishing signal periodicity to enhance time domain features, autocorrelation amplification to improve period stability, and correlation alignment to utilize physiological correlation, the SNR of heartbeat and pulse signals is improved and the waveform distortion rate is reduced. Ultimately, high-precision waveform enhancement of physiological signals is achieved through multi-dimensional processing, providing high-fidelity physiological feature input for subsequent speech signal compensation.

[0077] Furthermore, step A600 in the method provided in the embodiment of the present application includes:

[0078] A610: Obtain output verification data from the user, and establish scenario test feedback based on the output verification data.

[0079] A620: Establish feedback compensation according to the scenario test feedback, and perform signal processing optimization according to the feedback compensation.

[0080] In one embodiment, the user's output verification data is first obtained through the microphone array and accelerometer of the hearing aid device, including speech signal waveforms, ambient noise spectra, and user head movement trajectories in different scenarios. When the user is in a noisy environment (ambient sound energy > 80dB), the system records the speech signal's distortion parameters (such as harmonic distortion rate > 15%) and noise interference characteristics (energy proportion of the 500-2000Hz frequency band > 40%). In a quiet environment (< 50dB), data such as speech formant shift (such as the first formant F1 shift > 50Hz) is collected, forming a scenario test feedback database containing more than 200 sets of samples.

[0081] Based on scenario test feedback, a three-dimensional feedback compensation model was established: the X-axis represents the ambient noise level (30-100dB), the Y-axis represents the user's motion state (resting, walking, running), and the Z-axis represents the energy distribution of the speech band (200-8000Hz). When the user is detected walking and the ambient noise level is 75dB, the model automatically increases the gain of the 300-3400Hz speech band by 25%, while simultaneously implementing a -15dB attenuation on the 50-500Hz oscillator interference band. Using an adaptive filter (order 128), the weight coefficients are adjusted in real time, increasing the compensated speech signal-to-noise ratio (SNR) to 28-32dB. After 200 training iterations, the compensation parameter convergence error in complex scenarios was less than 5%, and the final output speech signal achieved a word recognition rate exceeding 90% in an 80dB traffic noise environment.

[0082] This content constructs scenario test feedback by obtaining multi-dimensional output verification data, establishes a dynamic compensation model based on the feedback, and optimizes signal processing parameters, achieving adaptive adjustment of the speech compensation strategy, thereby improving the speech SNR and word recognition rate of the fully implantable bone conduction hearing aid device in different environments, significantly improving the stability of the hearing aid effect and user experience.

[0083] In summary, the signal processing method for a fully implantable bone conduction hearing aid provided in the embodiments of the present application has the following technical effects:

[0084] This application introduces a synchronous sampling clock to control the embedded bone conduction microphone to collect mixed signals including heartbeat, pulse, vibrator vibration and voice. After separation and extraction by the dual-pass filtering module, the vibrator-microphone interference model is activated and the vibrator drive signal is input. The heartbeat and pulse signals are differentially suppressed based on the model output, and then processed through bandpass filtering, normalization, dynamic range compression and waveform enhancement. Finally, the enhanced physiological signal and vibrator drive signal are used to compensate the output of the voice signal, realizing signal processing of a fully implantable bone conduction hearing aid device, effectively improving the voice clarity and signal quality, achieving the technical effect of precise separation and interference suppression of mixed signals, and effectively improving the extraction quality and clarity of the voice signal in the fully implantable bone conduction hearing aid device.

[0085] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

[0086] Obviously, those skilled in the art may make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application is intended to include these modifications and variations.

Claims

1. A signal processing method for a fully implantable bone conduction hearing aid device, characterized in that: The method comprises: A synchronous sampling clock is introduced to control the embedded bone conduction microphone to perform signal acquisition and establish a signal data set, wherein the signal data set is a mixed signal set including a heartbeat signal, a pulse signal, a vibrator vibration signal, and a voice signal; After configuring the dual-pass filtering module, the signal data set is sent to the dual-pass filtering module, and the signal separation and extraction is performed using the dual-pass filtering channel to establish a separation and extraction result; activating a vibrator-microphone interference model, and using the vibrator driving signal in the separation and extraction result as an input feature to input the vibrator-microphone interference model; Perform differential interference suppression on the heartbeat signal and the pulse signal in the separation and extraction result according to the output result to establish an extraction signal; After the extracted signal is enhanced, the speech signal is compensated using the signal enhancement result and the oscillator driving signal and then output.

2. The signal processing method for a fully implantable bone conduction hearing aid according to claim 1, wherein: The method of performing signal separation and extraction using a dual-pass filtering channel to establish a separation and extraction result includes: activating a high-pass filter channel in the dual-pass filter channel, performing signal separation on the signal data set using the high-pass filter channel, outputting a voice signal from a first output terminal of the high-pass filter channel, and outputting a vibrator vibration signal from a second output terminal; A low-pass filter channel in the dual-pass filter channel is activated, and signal analysis of the signal data set is performed using the low-pass filter channel. A heartbeat signal is output from a first output end of the low-pass filter channel, and a pulse signal is output from a second output end.

3. The signal processing method for a fully implantable bone conduction hearing aid according to claim 1, wherein: Before activating the vibrator-microphone interference model, the method includes: In the guided training phase, the transfer characteristics of individuals are captured; After calling the universal vibrator-microphone interference model, initializing and configuring the universal vibrator-microphone interference model according to the transfer characteristics; An environment feedback window is configured, environment context variables are collected in the environment feedback window, and the environment context variables are used to compensate the initialized universal vibrator-microphone interference model to establish the vibrator-microphone interference model.

4. The signal processing method for a fully implantable bone conduction hearing aid according to claim 3, wherein: The step of performing differential interference suppression on the heartbeat signal and the pulse signal in the separation and extraction result according to the output result includes: Constructing a multi-scale delay prediction window based on the vibrator driving signal, identifying the cross-correlation peak points of the heartbeat signal and the pulse signal in the multi-scale delay prediction window, and establishing a transmission delay; The transmission delay and the output result are used to perform differential interference suppression on the corresponding heartbeat signal and pulse signal respectively.

5. The signal processing method for a fully implantable bone conduction hearing aid according to claim 1, wherein: The step of enhancing the extracted signal comprises: performing band-pass filtering on the heartbeat signal and the pulse signal in the extraction result respectively; After normalizing the bandpass filtering results, dynamic range compression is performed; After signal smoothing using a sliding window, waveform enhancement is performed.

6. The signal processing method for a fully implantable bone conduction hearing aid according to claim 5, wherein: The waveform enhancement step includes: Establishing signal cycles for the heartbeat signal and pulse signal after signal smoothing respectively; performing autocorrelation amplification processing of the signal according to the signal period; Perform correlation alignment of heartbeat signals and pulse signals, perform correlation authentication based on the correlation alignment results, and complete waveform enhancement with the correlation authentication results and autocorrelation amplification processing results.

7. The signal processing method for a fully implantable bone conduction hearing aid according to claim 1, wherein: The method of compensating the speech signal using the signal enhancement result and the oscillator driving signal and then outputting the compensating speech signal comprises: Obtaining user output verification data, and establishing scenario test feedback based on the output verification data; Feedback compensation is established based on the scenario test feedback, and signal processing optimization is performed based on the feedback compensation.

Images