Self-voice suppression device and method

Abstract

This application relates to a self-speech suppression device and method. The self-speech suppression method includes: receiving a first audio signal and a second audio signal from a first microphone and a second microphone, respectively; receiving the first audio signal through an optimized filter and generating a third audio signal based on the optimal filter coefficients; subtracting the second audio signal from the third audio signal to obtain an error signal; converting the error signal and the omnidirectional signal into an error frequency domain signal and an omnidirectional frequency domain signal, respectively, wherein the omnidirectional signal is the first audio signal, the second audio signal, or a linear combination thereof; calculating the amplitude ratio of the error frequency domain signal and the omnidirectional frequency domain signal in each frequency band; and determining the compression ratio of each frequency band based on the amplitude ratio of each frequency band.

Description

Technical Field

[0001] This invention relates to a self-speech suppression device and method, and particularly to a self-speech suppression device and method based on spectrum analysis. Background Technology

[0002] Self-speech suppression is an essential function of hearing aids (or assistive hearing devices), and its effectiveness directly determines the market acceptance of hearing aids. Generally, without special processing, hearing aids amplify all recorded sounds. When the wearer speaks, their self-speech is conducted through the skull to the ear canal, creating a feeling of blockage, and / or is transmitted through the air to the hearing aid, where it is amplified and makes the wearer feel uncomfortable due to excessive volume. The former (the feeling of blockage) can be resolved by releasing sound pressure, while the latter (the discomfort due to excessive volume) often forces the user to lower the hearing aid volume, resulting in limited actual hearing aid effectiveness.

[0003] If a hearing aid is equipped with a self-speech suppression function, it can suppress self-speech when it is detected, thus increasing the overall volume and achieving a better hearing aid effect.

[0004] In one known technique for self-speech suppression, an additional bone conduction sensor is required to detect the presence or absence of self-speech, enabling suppression when self-speech is detected. However, this additional bone conduction sensor increases the cost of the hearing aid, and its placement within the hearing aid cavity affects reception and complicates the cavity design.

[0005] Other known techniques for self-speech suppression typically require additional in-ear microphones, computationally intensive blind source separation (BSS) algorithms, computationally and storage-intensive AI-based speaker recognition technologies, extensive binaural information exchange, the use of two or more adaptive filters, dynamic gain adjustment or beamforming, or a combination of these. However, these known techniques suffer from drawbacks such as increased cost, high computational and storage requirements, large amounts of binaural information exchange, potential algorithm convergence difficulties, and / or slow response times.

[0006] Therefore, it is necessary to develop a self-speech suppression algorithm that can be calculated using only the signal output from the external earphone (or external microphone) that is essential in general hearing aids or wireless headphones. At the same time, the algorithm must be simple enough to have extremely low computational and storage requirements, so that it can be widely applied to most hearing aids or wireless headphones on the market. Summary of the Invention

[0007] The present invention aims to provide a self-speech suppression device comprising: a first microphone, a second microphone, an optimized filter, a subtractor, a frequency domain converter, a spectrum comparator, and a compression ratio determiner. The first microphone outputs a first audio signal, and the second microphone outputs a second audio signal. The optimized filter is communicatively connected to the first microphone to receive the first audio signal and generate a third audio signal based on the optimized filter coefficients. The subtractor is communicatively connected to the second microphone and the optimized filter to subtract the second audio signal from the third audio signal to obtain an error signal. The frequency domain converter is communicatively connected to the subtractor to convert the error signal and the omnidirectional signal into an error frequency domain signal and an omnidirectional frequency domain signal, respectively, wherein the omnidirectional signal is either the first audio signal, the second audio signal, or a linear combination thereof. The spectrum comparator is communicatively connected to the frequency domain converter to calculate the amplitude ratio of the error frequency domain signal and the omnidirectional frequency domain signal in each frequency band. The compression ratio determiner is communicatively connected to the spectrum comparator to determine the compression ratio of each frequency band based on the amplitude ratio of each frequency band.

[0008] The present invention also proposes a self-speech suppression method comprising: receiving a first audio signal and a second audio signal from a first microphone and a second microphone, respectively; receiving the first audio signal through an optimized filter and generating a third audio signal based on the optimal filter coefficients; subtracting the second audio signal from the third audio signal to obtain an error signal; converting the error signal and the omnidirectional signal into an error frequency domain signal and an omnidirectional frequency domain signal, respectively, wherein the omnidirectional signal is the first audio signal, the second audio signal, or a linear combination thereof; calculating the amplitude ratio of the error frequency domain signal and the omnidirectional frequency domain signal in each frequency band; and determining the compression ratio of each frequency band based on the amplitude ratio of each frequency band.

[0009] This invention proposes a method for analyzing the proportion of self-speech contained in each frequency band, and determining the amount of suppression required for each frequency band based on the analysis results. This invention only requires the audio output from the external microphones (i.e., the first microphone and the second microphone) that are essential for general hearing aids or wireless headphones to perform the calculation. At the same time, the computational and storage requirements of the algorithm are extremely low, so it can be widely applied to most hearing aids or wireless headphone models on the market.

[0010] To make the above features and advantages of the present invention more apparent and understandable, specific embodiments are described below in conjunction with the accompanying drawings. Attached Figure Description

[0011] A better understanding of embodiments of the present invention will be obtained from the following detailed description taken in conjunction with the accompanying drawings. It should be noted that, according to industry standard practice, the features are not shown to scale. In fact, the dimensions of the features may be arbitrarily increased or decreased for clarity of discussion.

[0012] Figure 1 This is a schematic diagram of a self-speech suppression device according to an embodiment of the present invention.

[0013] Figure 2 This is a system block diagram of a signal processor according to an embodiment of the present invention.

[0014] Figure 3 This is a flowchart of a self-speech suppression method according to an embodiment of the present invention.

[0015] Symbol Explanation

[0016] 10: Self-speech suppression device

[0017] 100: Signal Processor

[0018] 110: Optimization Filter

[0019] 120: Similarity Comparator

[0020] 130: Subtractor

[0021] 140: Frequency Domain Converter

[0022] 150: Spectrum Comparator

[0023] 160: Compression Ratio Determiner

[0024] 170: Signal Processing Unit

[0025] 180: Multiplier

[0026] h*Mic1(t): Third audio value

[0027] h(n): Optimal filter coefficients

[0028] Mic1: First microphone

[0029] Mic2: Second microphone

[0030] Mic(t): Omnidirectional signal

[0031] Mic1(t): First audio value

[0032] Mic2(t): Second audio frequency

[0033] MIC[k]: Omnidirectional frequency domain signal

[0034] OV_err(t): Error signal

[0035] OV_ERR[k]: Error frequency domain signal

[0036] OVD_Flag: Self-voice flag

[0037] r a [k]: Amplitude ratio

[0038] r[k]: Compression ratio

[0039] S1, S2, S3, S4, S5, S6, S7: Steps

[0040] X[k]: Frequency domain signal to be suppressed

[0041] Y[k]: Suppressed frequency domain signal Detailed Implementation

[0042] The embodiments of the present invention will now be discussed in detail. However, it will be understood that the embodiments provide many applicable concepts that can be implemented in a wide variety of specific contexts. The discussed and disclosed embodiments are for illustrative purposes only and are not intended to limit the scope of the invention.

[0043] Figure 1 This is a schematic diagram of a self-speech suppression device 10 according to an embodiment of the present invention. The self-speech suppression device 10 includes a first microphone Mic1, a second microphone Mic2, and a signal processor 100. In this embodiment, the first microphone Mic1 and the second microphone Mic2 are external microphones. In this embodiment, the signal processor 100 is, for example, a central processing unit, microprocessor, microcontroller, digital signal processor, application-specific integrated circuit, or other electronic components with computing capabilities. In this embodiment, the signal processor 100 can be connected to the first microphone Mic1 and the second microphone Mic2 via wired or wireless means. The first microphone outputs a first audio Mic1(t), and the second microphone outputs a second audio Mic2(t). The signal processor 100 receives the first audio Mic1(t) from the first microphone Mic1 and the second audio Mic2(t) from the second microphone Mic2.

[0044] Figure 2 This is a system block diagram of a signal processor 100 according to an embodiment of the present invention. The signal processor 100 includes an optimization filter 110, a similarity comparator 120, a subtractor 130, a frequency domain converter 140, a spectrum comparator 150, a compression ratio determiner 160, a signal processing unit 170, and a multiplier 180.

[0045] The optimized filter 110 is communicatively connected to a first microphone Mic1 (not shown) to receive the first audio Mic1(t) and generate a third audio h*Mic1(t) based on the optimal filter coefficients. Specifically, the optimized filter 110 is a filter with the optimal filter coefficients already applied, and the optimized filter 110 filters the first audio Mic1(t) using the optimal filter coefficients to generate the third audio h*Mic1(t). It is worth mentioning that the optimized filter 110 uses the optimal filter coefficients fixedly, so the optimized filter 110 is not an adaptive filter that needs to continuously update the filter coefficients. Accordingly, the self-speech suppression method of the embodiment of the present invention has a fast response speed and does not have the problem of algorithm non-convergence.

[0046] It should be noted that the signal processor 100 obtains the optimal filter coefficients beforehand through a training phase. The training phase process is as follows: First, the signal processor 100 receives first sample audio and second sample audio from the first microphone Mic1 and the second microphone Mic2. Next, the signal processor 100 performs voice activity detection (VAD) based on the first sample audio and the second sample audio to determine whether voice is present. If it is determined that no voice is present (voice activity flag VAD_Flag = 0), the signal processor 100 continues to receive the first sample audio and the second sample audio from the first microphone Mic1 and the second microphone Mic2. If it is determined that voice is present (voice activity flag VAD_Flag = 1), the signal processor 100 trains the optimized filter 110 based on the first audio and the second audio to find the optimal filter coefficients h(n), where the optimal filter coefficients h(n) reflect the frequency response difference between the first sample audio and the second sample audio along the acoustic path.

[0047] The aforementioned speech activity detection ensures that the stage of finding the optimal filter coefficients h(n) is performed only when speech is determined to be present, thus achieving more accurate convergence results. Specifically, speech activity detection involves performing speech activity detection on the first sample audio and / or the second sample audio. It can be set that when the speech activity detection result of the first sample audio and / or the second sample audio is positive, speech is determined to be present.

[0048] The aforementioned training phase is conducted in a quiet environment (defined in this invention as an ambient sound pressure level not exceeding 50 dB) to ensure that the first and second sample audio samples received by the first microphone Mic1 and the second microphone Mic2 originate from the mouth of the wearer of the self-speech suppression device. During the training phase, the wearer utters several self-speech phrases, resulting in several first and second sample audio samples. The signal processor 100 uses these sample audio samples to determine the optimal filter coefficients h(n). Notably, this training phase typically only needs to be performed once and can be completed within a few seconds (e.g., within ten seconds).

[0049] The aforementioned optimization filter 110 is configured to model the relative transfer function (i.e., the frequency response difference of the first and second sample audios along the acoustic path) between the first sample audio received by the first microphone Mic1 and the second sample audio received by the second microphone Mic2.

[0050] The signal processor 100 can find the optimal filter coefficients h(n) according to the objective function shown in equation (1):

[0051] min h E[Mic2-h*Mic1], (1)

[0052] Where h represents the filter coefficients, Mic1 and Mic2 correspond to the first and second sample audio samples respectively, and E represents the mathematical expectation. The optimal filter coefficient h(n) is the filter coefficient that minimizes the objective function. In other words, the optimal filter coefficient h(n) is found by optimizing the optimized filter 110 according to the objective function shown in equation (1). Here, the signal processor 100 can find the optimal filter coefficient h(n) using any search method, such as the least mean square error (LMSE) algorithm, the normalized LMSE algorithm, or the adaptive LMSE algorithm. This invention does not limit the method used to solve for the optimal filter coefficient h(n). Those skilled in the art will understand how to use the above-example algorithms to solve for a given objective function, which will not be described in detail here. Wherein, if the embodiment described herein is implemented in the time domain, then “*” in the above equation (1) represents convolution; and if the embodiment described herein is implemented in the frequency domain, then “*” in the above equation (1) represents product.

[0053] The signal processor 100 can also find the optimal filter coefficients h(n) in the following way.

[0054] The signal processor 100 can first perform mathematical operations on the first sample audio and the aforementioned relative transformation function to obtain the third sample audio. Then, it seeks the optimal filter coefficient h(n) with the goal of maximizing the similarity between the third sample audio and the second sample audio. The similarity can be defined using the cosine similarity or correlation coefficient between the third sample audio and the second sample audio. If the embodiment described here is implemented in the time domain, the aforementioned mathematical operation is a convolution; if the embodiment described here is implemented in the frequency domain, the aforementioned mathematical operation is a product.

[0055] Regarding the solution method for the optimal filter coefficient h(n), the adaptive minimum mean square error algorithm is used as an example. After the nth sample audio is read in, the error signal OV_err is calculated according to the following formula (2):

[0056] OV_err=(Mic2-h*hMic1)(n), (2)

[0057] Where h*Mic1 corresponds to the third sample audio, and the filter coefficient h is updated according to this error signal OV_err, as shown in the following equation (3):

[0058] h′=h+μ0·VAD_Flag·OV_err·Mic1, (3)

[0059] Where h′ represents the updated filter coefficients, and h and Mic1 in the above equation are N-dimensional column vectors, i.e., h = (h0, h1, ..., h N-1 ) T , Mic1=(Mic1[n],Mic1[n-1],…,Mic1[n-N+1]) T The convergence step parameter μ0 is a given positive number. When the speech activity flag VAD_Flag = 1, h′ is updated according to the above equation (3). Conversely, when the speech activity flag VAD_Flag = 0, h′ is not updated. The above is an example of solving the optimal filter coefficients h(n) in the time domain. According to the above, in an environment where the ambient sound pressure level does not exceed 50 dB (i.e., during the training phase), the optimal filter coefficients h(n) are those that minimize the error signal OV_err.

[0060] Similarly, the optimal filter coefficients can also be solved in the frequency domain in the following way, firstly by defining the discrete Fourier transform pairs according to the following equations (4), (5) and (6):

[0061]

[0062] Then, for each frame, the error signal OV_ERR[k] is calculated according to the following equation (7):

[0063] OV_ERR[k]=(MIC2[k]-H[k]·MIC1[k])(n), (7)

[0064] And based on the error signal OV_ERR[k], update the filter coefficients H[k], as shown in the following equation (8):

[0065] H[k]′=H[k]+μ0·VAD_Flag·|OV_ERR[k]|·MIC1[k], (8)

[0066] Where H[k]′ represents the updated filter coefficients. When the speech activity flag VAD_Flag = 1, H[k]′ is updated according to equation (8) above. Conversely, when the speech activity flag VAD_Flag = 0, H[k]′ is not updated. The above is an example of solving for the optimal filter coefficients in the frequency domain.

[0067] Similarity comparator 120 is communicatively connected to the second microphone Mic2 (not shown) to receive the second audio Mic2(t). Similarity comparator 120 is communicatively connected to the optimization filter 110 to receive the third audio h*Mic1(t). Similarity comparator 120 compares the similarity between the second audio Mic2(t) and the third audio h*Mic1(t) to generate a self-voice flag OVD_Flag, wherein the self-voice flag OVD_Flag is used to indicate whether the first microphone Mic1 and the second microphone Mic2 have recorded self-voice.

[0068] In embodiments of the present invention, the aforementioned similarity is cosine similarity or correlation coefficient. In other words, the similarity can be defined using the cosine similarity or correlation coefficient between the second audio Mic2(t) and the third audio h*Mic1(t). When the similarity is greater than a threshold, the self-voice flag OVD_Flag = 1 indicates that the first microphone Mic1 and the second microphone Mic2 have recorded self-voice. When the similarity is not greater than the threshold, the self-voice flag OVD_Flag = 0 indicates that the first microphone Mic1 and the second microphone Mic2 have not recorded self-voice. The aforementioned threshold can be a value determined by the designer according to actual needs, such as 0.9995 or 0.999, etc. This disclosure does not limit the value of the threshold.

[0069] In embodiments of the present invention, the similarity comparison method employed by the similarity comparator 120 is based on the consistency of phase and amplitude, and therefore is more reliable than the power comparison method that relies solely on amplitude. The similarity comparison method can be implemented in various ways; an example is given below.

[0070] First, the first audio Mic1(t) is input to the optimized filter 110 with the optimal filter coefficients applied to obtain the third audio h*Mic1(t), and the third audio is represented by sig1 in the following equation (9), and the second audio Mic2(t) is represented by sig2. In terms of phase, the cosine similarity between sig1 and sig2 satisfies the following equation (9):

[0071]

[0072] The threshold value mentioned above is, for example, 0.9995. This disclosure does not limit the value of the threshold value. In equation (9), the subscript 2 represents the L2 norm of the vector. Here, sig1·sig2 is the short-time inner product, as shown in equation (10):

[0073]

[0074] Where M is a positive integer. Furthermore, in terms of amplitude, the amplitude of sig1 should be comparable to that of sig2, for example, satisfying equation (11) as follows:

[0075] sig1·sig2>γsig2·sig2, (11)

[0076] The threshold γ can be set to 0.999, and this disclosure does not limit the value of γ.

[0077] Specifically, the similarity comparator 120 aims to detect whether the wearer of the self-speech suppression device 10 is uttering their own speech, and can suppress the audio accordingly to provide the wearer with a better user experience.

[0078] Specifically, in some known self-speech detection methods, signal analysis is usually performed directly on the audio received by the microphone to determine whether self-speech has occurred. In contrast, this invention uses an optimized filter 110 with optimal filter coefficients and a similarity comparator 120 to determine whether self-speech has occurred. Therefore, this invention can better improve the accuracy of self-speech detection, reduce the amount of computation, and shorten the response time.

[0079] In an embodiment of the present invention, when the self-voice suppression device 10 is worn on the head of the wearer, the first microphone Mic1 and the second microphone Mic2 are both located in the same ear of the wearer (e.g., Figure 1 (The left ear of the wearer shown). In embodiments of the present invention, it should be noted that, as Figure 1 As shown, when the self-speech suppression device 10 is worn on the wearer's head, the first microphone Mic1 is closer to the wearer's mouth than the second microphone Mic2. This is because, relative to the second microphone Mic2, the first audio output by the first microphone Mic1 needs to pass through the optimization filter 110 before it can be correlated with the second audio output by the second microphone Mic2 for signal processing (e.g., similarity comparison). Therefore, the first microphone Mic1 being closer to the wearer's mouth than the second microphone Mic2 can better compensate for the time delay introduced by the optimization filter 110.

[0080] Subtractor 130 is communicatively connected to a second microphone Mic2 (not shown) to receive a second audio Mic2(t), and subtractor 130 is communicatively connected to an optimization filter 110 to receive a third audio h*Mic1(t). Subtractor 130 subtracts the second audio Mic2(t) from the third audio h*Mic1(t) to obtain an error signal OV_err(t) (for example, refer to equation (2) above).

[0081] Specifically, when the self-voice flag OVD_Flag = 1, the second audio Mic2(t) and the third audio h*Mic1(t) are very similar signals, therefore the error signal OV_err(t) will be a signal with a relatively small amplitude. Conversely, when the self-voice flag OVD_Flag = 0, the second audio Mic2(t) and the third audio h*Mic1(t) are no longer similar, therefore the error signal OV_err(t) will be a signal with a relatively large amplitude. The amplitude mentioned here as relatively small or relatively large is relative to the first audio Mic1(t) or the second audio Mic2(t) at the same moment.

[0082] Frequency domain converter 140 is communicatively connected to a first microphone Mic1 and / or a second microphone Mic2 (not shown) to receive a first audio Mic1(t) and / or a second audio Mic2(t). Frequency domain converter 140 is also communicatively connected to a subtractor 130 to receive an error signal OV_err(t). Frequency domain converter 140 is used to convert the error signal OV_err(t) and the omnidirectional signal Mic(t) respectively (e.g., via a fast Fourier transform (FFT)) into an error frequency domain signal OV_ERR[k] and an omnidirectional frequency domain signal MIC[k]. The omnidirectional signal Mic(t) is either the first audio Mic1(t), the second audio Mic2(t), or a linear combination of the first audio Mic1(t) and the second audio Mic2(t).

[0083] The spectrum comparator 150 is communicatively connected to the frequency domain converter 140 to receive the error frequency domain signal OV_ERR[k] and the omnidirectional frequency domain signal MIC[k], and calculates the amplitude ratio (or power ratio) r of the error frequency domain signal OV_ERR[k] and the omnidirectional frequency domain signal MIC[k] in each frequency band (or each frequency band). a [k]. The spectrum comparator 150 is used to compare the power of the error frequency domain signal OV_ERR[k] and the omnidirectional frequency domain signal MIC[k] in a frequency-band manner. In other words, the spectrum comparator 150 is used to perform spectrum analysis on the error frequency domain signal OV_ERR[k] and the omnidirectional frequency domain signal MIC[k].

[0084] For example, the spectrum comparator 150 calculates the amplitude ratio r according to the following equation (12). a [k]:

[0085]

[0086] In other words, the spectrum comparator 150 performs frequency analysis on the ratio of the error frequency domain signal OV_ERR[k] to the omnidirectional frequency domain signal MIC[k] to obtain the frame-averaged value of the power spectrum component of the ratio of the error frequency domain signal OV_ERR[k] to the omnidirectional frequency domain signal MIC[k].

[0087] The spectrum comparator 150 can also calculate the amplitude ratio r in the following way. a [k].

[0088] Whenever the error frequency domain signal OV_ERR[k] and the omnidirectional frequency domain signal MIC[k] are obtained, their absolute values ​​are calculated as an exponential moving average according to the following equations (13) and (14):

[0089] smooth_OV_ERR[k]=β·smooth_OV_ERR[k]+(1-β)·‖OV_ERR[k]‖, (13)

[0090] smooth_MIC[k]=β·smooth_MIC[k]+(1-β)·‖OV_ERR[k]‖, (14)

[0091] For example, β is 0.7.

[0092] Next, the amplitude ratio r is calculated according to the following formula (15). a [k].

[0093]

[0094] Compression ratio determiner 160 communicates with spectrum comparator 150 to receive amplitude ratio r of each frequency band.a [k], and based on the amplitude ratio r of each frequency band a [k] determines the compression ratio r[k] for each frequency band.

[0095] Compression ratio determiner 160 communicates with similarity comparator 120 to receive self-voice flag OVD_Flag. Compression ratio determiner 160 determines the compression ratio based on the ratio r of self-voice flag OVD_Flag to the amplitude of each frequency band. a The compression ratio r[k] for each frequency band is determined by [k]. Specifically, the compression ratio determiner 160 determines the compression ratio r[k] based on the ratio r[k] of the self-voice flag OVD_Flag and the amplitude of each frequency band. a [k] is used to determine the proportion of self-speech presence in each frequency band, and the compression ratio (i.e., compression ratio r[k]) of each frequency band is calculated accordingly.

[0096] In an embodiment of the present invention, the compression ratio determiner 160 determines the amplitude ratio r of each frequency band. a [k] is smoothed to obtain the compression ratio r[k] for each frequency band. The above smoothing process can be achieved through the following operation: when the self-voice flag OVD_Flag indicates that the first microphone Mic1 and the second microphone Mic2 have recorded self-voice (i.e., self-voice flag OVD_Flag = 1), the compression ratio r[k] is adjusted towards the amplitude ratio r a [k] approaches, conversely, when the self-voice flag OVD_Flag indicates that the first microphone Mic1 and the second microphone Mic2 have not recorded self-voice (i.e., self-voice flag OVD_Flag = 0), the compression ratio r[k] approaches 1.

[0097] This causes the compression ratio r[k] to be related to the amplitude ratio r a The way [k] approaches is shown in equation (16):

[0098] r[k]=α·r[k]+(1-α)·r a [k], (16)

[0099] The method by which the compression ratio r[k] approaches 1 is shown in equation (17):

[0100] r[k]=α·r[k]+(1-α)·1, (17)

[0101] Where α is a value less than and close to 1, for example, α = 0.9.

[0102] Specifically, the compression ratio r[k] is between the amplitude ratio r aThe value between [k] and 1, and the compression ratio r[k] is a continuously adjusted value. The preset value of the compression ratio r[k] is a value set by the designer based on actual needs. When the self-voice flag OVD_Flag = 1, because there is self-voice, the audio should be compressed more, thus causing the compression ratio r[k] to move towards the amplitude ratio r a [k] approaches; when the self-voice flag OVD_Flag = 0, since there is no self-voice, the audio should be compressed less, thus causing the compression ratio r[k] to approach 1.

[0103] The signal processing unit 170 is communicatively connected to a first microphone Mic1 and a second microphone Mic2 (not shown) to receive a first audio signal Mic1(t) and a second audio signal Mic2(t). It can selectively perform beamforming on the first audio signal Mic1(t) and the second audio signal Mic2(t), and selectively perform noise reduction, followed by time-domain to frequency-domain conversion (e.g., via Fast Fourier Transform, FFT) to obtain the frequency domain signal X[k] to be suppressed. In other words, the frequency domain signal X[k] to be suppressed is obtained by selectively performing beamforming and noise reduction on the first audio signal Mic1(t) and the second audio signal Mic2(t), followed by time-domain to frequency-domain conversion.

[0104] Multiplier 180 is communicatively connected to compression ratio determiner 160 to receive the frequency domain signal X[k] to be suppressed, and multiplies the frequency domain signal X[k] to be suppressed with compression ratio r[k] to generate the suppressed frequency domain signal Y[k]. In other words, compression ratio r[k] is multiplied by the audio to be suppressed to obtain the suppressed audio, thus completing the self-speech suppression operation.

[0105] Figure 3This is a flowchart of a self-speech suppression method according to an embodiment of the present invention. In step S1, the signal processor 100 receives a first audio signal Mic1(t) and a second audio signal Mic2(t) from the first microphone Mic1 and the second microphone Mic2, respectively. In step S2, the optimization filter 110 receives the first audio signal Mic1(t) and generates a third audio signal h*Mic1(t) based on the optimal filter coefficients h(n). In step S3, the subtractor 130 subtracts the second audio signal Mic2(t) from the third audio signal h*Mic1(t) to obtain an error signal OV_err(t). In step S4, the frequency domain converter 140 converts the error signal OV_err(t) and the omnidirectional signal Mic(t) into an error frequency domain signal OV_ERR[k] and an omnidirectional frequency domain signal MIC[k], respectively. In step S5, the spectrum comparator 150 calculates the amplitude ratio r of the error frequency domain signal OV_ERR[k] and the omnidirectional frequency domain signal MIC[k] in each frequency band. a [k]. In step S6, the compression ratio determiner 160 determines the compression ratio based on the amplitude ratio r of each frequency band. a [k] determines the compression ratio r[k] for each frequency band. In step S7, the multiplier 180 multiplies the frequency domain signal X[k] to be suppressed with the compression ratio r[k] to generate the suppressed frequency domain signal Y[k].

[0106] In embodiments of the present invention, as described above, the first microphone Mic1 and the second microphone Mic2 are both located in the same ear of the wearer, thus the self-speech suppression device 10 proposed in this invention can operate independently in one ear. On the other hand, in other embodiments of the present invention, a binaural information exchange function can be added, that is, another set of self-speech suppression devices can be configured in the other ear of the wearer. These two sets of self-speech suppression devices can operate collaboratively, and it can be required that self-speech is only detected when both sets of self-speech suppression devices determine that self-speech is present. This will improve the accuracy of self-speech suppression in binaural hearing aids.

[0107] Based on the above, this invention proposes a method for analyzing the proportion of self-speech contained in each frequency band, and determining the amount of suppression required for each frequency band based on the analysis results. This invention only requires the use of the external microphones (i.e., the first microphone and the second microphone) that are essential for general hearing aids or wireless headphones to perform signal analysis. It also has the advantages of high accuracy, low computational load, and short response time, and can operate independently in one ear. It can be widely applied to most hearing aids or wireless headphone models on the market.

[0108] The foregoing has outlined the features of several embodiments, thus enabling those skilled in the art to better understand the implementation of the present invention. Those skilled in the art should understand that the present invention can be readily used as a basis to design or modify other processes and structures, thereby achieving the same objectives and / or advantages as the embodiments described herein. Those skilled in the art should also understand that these equivalent constructions do not depart from the concept and scope of the present invention, and that various changes, substitutions, and modifications can be made without departing from the concept and scope of the present invention.

Claims

1. A self-speech suppression device, comprising: A first microphone, used to output a first audio signal; A second microphone for outputting a second audio signal; An optimized filter is communicatively connected to the first microphone to receive the first audio and generate a third audio based on an optimized filter coefficient. A subtractor is communicatively connected to the second microphone and the optimization filter to subtract the second audio signal from the third audio signal to obtain an error signal; A frequency domain converter is communicatively connected to the subtractor to convert the error signal and an omnidirectional signal into an error frequency domain signal and an omnidirectional frequency domain signal, respectively, wherein the omnidirectional signal is the first audio signal or the second audio signal or a linear combination thereof; A spectrum comparator, communicatively connected to the frequency domain converter, calculates the amplitude ratio of the error frequency domain signal to the omnidirectional frequency domain signal in each frequency band; and A compression ratio determiner is communicatively connected to the spectrum comparator to determine a compression ratio for each frequency band based on the amplitude ratio of each band.

2. The self-speech suppression device as claimed in claim 1, wherein the compression ratio determiner performs a smoothing process on the amplitude ratio of each frequency band to obtain the compression ratio of each frequency band.

3. The self-speech suppression device as described in claim 2, further comprising: A similarity comparator is communicatively connected to the second microphone and the optimization filter to compare the similarity between the second audio and the third audio to generate a self-speech flag, wherein the self-speech flag is used to indicate whether the first microphone and the second microphone have recorded self-speech. The compression ratio determiner determines the compression ratio of each frequency band based on the ratio of the self-voice flag to the amplitude of each frequency band.

4. The self-speech suppression device as claimed in claim 3, wherein the smoothing process comprises: When the self-voice flag indicates that the first microphone and the second microphone have recorded self-voice, the compression ratio is brought closer to the amplitude ratio. and When the self-voice flag indicates that the first and second microphones have not recorded self-voice, the compression ratio is brought closer to 1.

5. The self-speech suppression device as described in claim 4, The formula that makes the compression ratio approach the amplitude ratio is as follows: r[k]=α·r[k]+(1-α)·r a [k]; The formula that brings the compression ratio closer to 1 is as follows: r[k] = α·r[k] + (1-α)·1; Where r a [k] represents the amplitude ratio, r[k] represents the compression ratio, and α is a value less than and close to 1.

6. The self-speech suppression device as described in claim 1, further comprising: A multiplier, communicatively connected to the compression ratio determiner, multiplies a frequency domain signal to be suppressed by the compression ratio to generate a suppressed frequency domain signal.

7. The self-speech suppression device as claimed in claim 6, wherein the frequency domain signal to be suppressed is obtained by selectively subjecting the first audio and the second audio to beamforming or not, and selectively subjecting to noise suppression or not, and then converting them from the time domain to the frequency domain.

8. The self-speech suppression device of claim 1, wherein the optimized filter is trained in an environment with an ambient sound pressure level not exceeding 50 dB, and the optimal filter coefficients are such that the error signal has a minimum value.

9. The self-speech suppression device of claim 3, wherein the similarity is a cosine similarity or a correlation coefficient, wherein when the similarity is greater than a threshold, the self-speech flag indicates that the first microphone and the second microphone have recorded self-speech.

10. A self-speech suppression method, comprising: Receive a first audio signal and a second audio signal from a first microphone and a second microphone, respectively; The first audio is received through an optimized filter and a third audio is generated according to an optimal filter coefficient. An error signal is obtained by subtracting the second audio signal from the third audio signal. The error signal and an omnidirectional signal are respectively converted into an error frequency domain signal and an omnidirectional frequency domain signal, wherein the omnidirectional signal is the first audio signal or the second audio signal or a linear combination thereof; Calculate the amplitude ratio of the error frequency domain signal to the omnidirectional frequency domain signal in each frequency band; and The compression ratio of each frequency band is determined based on the amplitude ratio of each band.

Images