Method suitable for enhancing voice of opposite side during face-to-face conversation of high-noise workshop protection earmuff

Abstract

The invention discloses a method suitable for enhancing voice of an opposite side during face-to-face conversation of a high-noise workshop protective earmuff, and the method comprises the steps: synchronously collecting noise-containing voice signals of a body and an opposite talker during face-to-face conversation in a high-noise environment through employing a microphone array, and carrying out the preprocessing of the noise-containing voice signals; acquiring input signals of a main channel and an auxiliary channel of the first-stage noise interference suppression unit; voice endpoint detection is carried out, and whether the current period is a voice period or a voice-free period is judged; a filter of the first-stage noise interference suppression unit adaptively adjusts a filter coefficient to complete cancellation of environmental noise interference; carrying out ontology voice endpoint detection, and judging whether the current period is an ontology voice period or an ontology voice-free period; and the filter coefficient is adaptively adjusted according to the judgment information, the cancellation of the voice interference of the body is completed, and an enhanced signal of the voice of the opposite side is obtained and is broadcasted by a built-in loudspeaker of the protective earmuff. According to the invention, the voice signal of the opposite side in a high-noise environment can be enhanced to a certain extent.

Description

Technical Field

[0001] The invention belongs to voice processing technology, and in particular relates to a method for enhancing the other party's voice when having face-to-face conversation with protective earmuffs in a high-noise workshop. Background Art

[0002] Noise is usually defined as unwanted noise, interference or irregular sounds in a specific environment. Long-term exposure to noise, especially high-intensity noise, can cause serious harm to human physical and mental health, and even lead to permanent hearing damage. Studies have shown that long-term exposure to high-intensity noise (such as above 85dB) may lead to noise-induced hearing loss (NIHL). NIHL is a gradually accumulated hearing loss, which usually manifests as high-frequency hearing loss in the early stage, and over time, the range of damage will gradually expand to low-frequency hearing. In addition, long-term exposure to high-noise environments may also cause mental health problems, such as anxiety, depression and other emotional disorders, as well as cardiovascular health problems, such as hypertension and heart disease. Therefore, it is particularly important to take effective protective measures for working environments with severe noise pollution. For example, in some factory workshops with high noise levels, workers must wear earplugs, earmuffs or protective headphones to avoid noise damage to hearing.

[0003] In recent years, with the continuous advancement of noise protection technology, protective earmuffs and various noise-reducing headphones have been widely used in many fields, especially in industrial sites and high-noise working environments. These protective devices effectively reduce the impact of external noise on workers and reduce the risk of hearing damage caused by noise. However, with the popularization of these protective measures, some new problems have also arisen. Specifically, although protective earmuffs can significantly shield the noise in the surrounding environment, they also hinder normal language communication between workers. In many industrial production and working environments, language communication is crucial, which not only helps to improve work efficiency, but also affects the safety of employees. Lack of effective communication may lead to miscommunication of information, increase safety hazards at work, and even cause unnecessary accidents. Therefore, how to ensure that language communication at work is not disturbed while protecting the hearing health of employees has become a technical challenge that needs to be solved urgently.

[0004] To solve this problem, the current mainstream solution is to combine protective earmuffs with headset communication modules. This solution can not only effectively isolate external noise and protect hearing safety, but also has the function of real-time voice communication, which greatly improves the communication efficiency of users during work or operation. However, this integrated design also brings some challenges, especially in terms of cost and wearing comfort. Due to the addition of a communication module, the price of such protective earmuffs is usually higher, and the weight of the earmuffs is also increased. Wearing them for a long time may cause discomfort, thus affecting the user's comfort. Therefore, how to use microphone array signal processing technology and related speech enhancement algorithms to effectively retain the other party's voice signal while wearing protective earmuffs to isolate noise, and effectively improve voice quality through noise suppression and speech enhancement technology has become an important topic that cannot be ignored in current research and practical applications. Summary of the invention

[0005] The purpose of the present invention is to provide a method for enhancing the other party's voice when having a face-to-face conversation with protective earmuffs in a high-noise workshop in order to solve the above-mentioned problems.

[0006] The technical solution to achieve the purpose of the present invention is: a method for enhancing the other party's voice when having a face-to-face conversation with protective earmuffs in a high-noise workshop, the steps are as follows:

[0007] Step 1: using a microphone array consisting of a left ear microphone, a left ear microphone, and a mouth microphone to synchronously collect noisy speech signals of the subject and the opposite talker during face-to-face conversation in a high noise environment, and preprocessing the collected signals;

[0008] Step 2: Using the speech signals of the left ear microphone channel and the right ear microphone channel obtained in step 1, fixed beam forming is performed to obtain the main channel and auxiliary channel input signals of the first-stage noise interference suppression unit;

[0009] Step 3: Use the main channel voice signal obtained in step 2 to perform voice endpoint detection to determine whether the current period is a voice period or a non-voice period;

[0010] Step 4: Using the main channel and auxiliary channel voice signals obtained in step 2 and the decision information obtained in step 3, the filter of the first-stage noise interference suppression unit adaptively adjusts the filter coefficient to cancel the environmental noise interference;

[0011] Step 5: Use the voice signal of the mouth microphone channel obtained in step 1 to perform the main voice endpoint detection to determine whether the current period is the main voice period or the non-main voice period;

[0012] Step 6. Utilize the voice signal of the mouth microphone channel obtained in step 1, the noise-suppressed voice signal obtained in step 4, and the decision information obtained in step 5, and use the noise-suppressed voice signal as the main channel input signal of the second-stage main body voice interference suppression unit, and the voice signal of the mouth microphone channel as the auxiliary channel input signal. According to the decision information, adaptively adjust the filter coefficient to cancel the main body voice interference, obtain the enhanced signal of the other party's voice, and broadcast it through the built-in speaker of the protective earmuffs.

[0013] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the program, the method for enhancing the other party's voice when having a face-to-face conversation with protective earmuffs in a high-noise workshop is implemented.

[0014] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the above-mentioned method for enhancing the voice of the other party when having a face-to-face conversation with protective earmuffs in a high-noise workshop.

[0015] A computer program product includes a computer program. When the computer program is executed by a processor, the method for enhancing the voice of the other party when having a face-to-face conversation with protective earmuffs in a high-noise workshop is implemented.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1) the present invention can be applied to protective earmuffs, and can effectively retain and enhance the voice signal of the person talking opposite in a high-noise environment, ensuring that the wearer of the earmuffs can still communicate face to face; 2) the present invention adopts a system structure composed of low-cost microphones. Compared with the protective earmuffs with integrated communication modules in the prior art, the present invention significantly reduces the production cost and has strong market competitiveness; 3) the system structure of the present invention is simple in design and easy to implement, and can provide a reliable and robust voice communication solution for wearing protective earmuffs in a high-noise environment.

[0017] The present invention is further described in detail below in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 4 is a system structure diagram of the method of the present invention.

[0019] Figure 2 Schematic diagram of a three-element microphone array for collecting noisy speech data in the present invention.

[0020] Figure 3 The time domain waveform of the noisy speech signal collected by the three microphone channels and the corresponding time-frequency spectrum diagram.

[0021] Figure 4The time domain waveform diagram and the corresponding time-frequency spectrum diagram of the noisy speech signals of the three microphone channels after preprocessing.

[0022] Figure 5 It is the time domain waveform diagram of the main channel speech signal output by the fixed beamforming unit and the corresponding time-frequency spectrum diagram.

[0023] Figure 6 A schematic diagram of the detection result of the detection control unit performing speech endpoint detection on the main channel speech signal output by the beamforming unit.

[0024] Figure 7 It is the time domain waveform of the speech signal output by the noise interference suppression unit and the corresponding time-frequency spectrum diagram.

[0025] Figure 8 This is a schematic diagram of the detection results of the detection control unit performing speech endpoint detection on the speech signal of the mouth microphone channel.

[0026] Fig. 9 It is the time domain waveform of the speech signal output by the main body speech interference suppression unit and the corresponding time-frequency spectrum diagram. DETAILED DESCRIPTION

[0027] Combination Figure 1 The present invention proposes a method for enhancing the other party's voice when having a face-to-face conversation with protective earmuffs in a high-noise workshop, comprising the following steps:

[0028] Step 1: using a microphone array consisting of a left ear microphone, a left ear microphone, and a mouth microphone to synchronously collect noisy speech signals of the subject and the opposite talker during face-to-face conversation in a high noise environment, and preprocessing the collected signals;

[0029] Furthermore, step 1 is specifically as follows:

[0030] Step 1-1, number the left ear microphone, left ear microphone, and mouth microphone as 1, 2, and 3 respectively, wherein microphones 1 and 2 are located at the left and right earmuffs of the protective earmuffs, and microphone 3 is located at the extended mouth stem connected to the protective earmuffs, ensuring that microphones 1 and 2 are symmetrically distributed on both sides of microphone 3;

[0031] Step 1-2: Using the array setting of step 1-1, collect the noisy speech signals of each channel. In the specific example of the present invention, the data sampling rate is 16kHz and the analysis period length is 16s. Pre-emphasize the collected noisy speech signals of each channel using a first-order FIR high-pass digital filter. The filter transfer function is

[0032] H(z)=1-az -1 0.9<a<1.0

[0033] In the specific example of the present invention, the pre-emphasis parameter a is set to 0.95, z -1 is the unit delay operator in the z-transform.

[0034] Step 1-3, performing short-time frame division and windowing processing on the noisy speech signals of each channel obtained by step 1-2. In a specific example of the present invention, the frame length is 20 ms, the frame shift is 10 ms, and the window function is a Hamming window;

[0035] Step 2: Using the speech signals of the left ear microphone channel and the right ear microphone channel obtained in step 1, fixed beam forming is performed to obtain the main channel and auxiliary channel input signals of the first-stage noise interference suppression unit;

[0036] Furthermore, step 2 is specifically as follows:

[0037] The fixed beamforming process is performed using the preprocessed signals of the left and right ear microphone channels obtained by the method described in step 1. The output y(n) of the fixed beamformer is defined as:

[0038]

[0039] Where M represents the number of elements in the microphone array, x i (n) represents the speech signal collected by the i-th array element, where i = 0, 1, ..., M-1. i represents the weight of the i-th filter, τ i Represents the delay estimation between the ith array element and the relative reference array element. In the specific example of the present invention, the number of microphone array elements M is set to 2, w is 1 / 2, τ i are all 0.

[0040] Step 3: Use the main channel voice signal obtained in step 2 to perform voice endpoint detection to determine whether the current period is a voice period or a non-voice period;

[0041] Furthermore, step 3 is specifically as follows:

[0042] Step 3-1: The main channel signal obtained by the method described in step 2 is subjected to noise reduction processing using an improved spectral subtraction method of multi-window spectrum estimation, specifically:

[0043] Step 3-1-1: Perform FFT on the current frame signal and calculate its amplitude spectrum |X i (k)| and the phase spectrum θ i (k) (i represents the i-th frame, k represents the k-th spectrum line), smoothing is performed between adjacent frames, specifically, M frames are taken before and after the i-th frame, and a total of 2M+1 frames are averaged. In the specific example of the present invention, M is 3, and the average amplitude spectrum is calculated as:

[0044]

[0045] Step 3-1-2, perform multi-window spectrum estimation on the current frame signal, calculate its multi-window spectrum power spectrum density P(k,i) (i represents the i-th frame, k represents the k-th spectrum line), and perform smoothing between adjacent frames. Specifically, take M frames before and after the i-th frame as the center, and average a total of 2M+1 frames. In the specific example of the present invention, M is 3, and the smoothed power spectrum density is calculated as:

[0046]

[0047] Step 3-1-3: Calculate the average power spectrum density of the noise based on the noise frame of the leading silent segment:

[0048]

[0049] Wherein, NIS is the number of frames occupied by the leading no-talk segment. In the specific example of the present invention, the number of frames NIS of the leading no-talk segment is 200; y (k,i) is the smoothed power spectral density of the i-th frame obtained by the method described in step 3-1-2.

[0050] Step 3-1-4, using the average amplitude spectrum obtained by the method described in step 3-1-1, the smoothed power spectrum density obtained by the method described in step 3-1-2, and the noise average power spectrum density value obtained by the method described in step 3-1-4, and using the spectrum subtraction relationship to calculate the gain factor, the gain factor is defined as:

[0051]

[0052] Wherein, α is the over-reduction factor, β is the gain compensation factor, and in the specific embodiment of the present invention, the over-reduction factor α is 2.8, and the gain compensation factor β is 0.001;

[0053] Step 3-1-5, using the average amplitude spectrum obtained by the method described in step 3-1-1 and the gain factor obtained by the method described in step 3-1-4, obtain the speech signal after noise reduction:

[0054]

[0055] Step 3-2: Calculate the short-time subband spectral entropy of each frame signal using the de-noised speech signal obtained by the method described in step 3-1, specifically:

[0056] Step 3-2-1: Divide the current frame signal into N b1 subbands. In the specific example of the present invention, the number of subbands N b1 is 40, calculate the subband energy of each subband, and obtain the subband energy E of the mth subband in the i-th frameb1 (m,i), 1≤m≤N b1 ;

[0057] Step 3-2-2: Using the subband energy obtained by the method described in step 3-2-1, the subband energy probability of the mth subband in the i-th frame is calculated as:

[0058]

[0059] Step 3-2-3, using the sub-band energy probability obtained by the method described in step 3-2-2, calculate the sub-band spectral entropy of the i-th frame as:

[0060]

[0061] Step 3-3: Using the short-time subband spectral entropy obtained by the method described in step 3-2, two thresholds are set, and a double threshold method is used to perform speech endpoint detection, specifically:

[0062] Step 3-3-1, set a lower threshold T2, and make a rough judgment. The signal below the T2 threshold is definitely speech, and the start and end points of the speech are outside the time points corresponding to the intersection of the threshold and the sub-band spectral entropy envelope;

[0063] Step 3-3-2, set a higher threshold T1, and search left and right respectively from the roughly judged speech start and end points obtained by the method described in step 3-3-1, and find the two points where the subband spectral entropy envelope intersects with the threshold T1, and obtain the speech start and end point positions determined by the double threshold method.

[0064] Step 4: Using the main channel and auxiliary channel voice signals obtained in step 2 and the decision information obtained in step 3, the filter of the first-stage noise interference suppression unit adaptively adjusts the filter coefficient to cancel the environmental noise interference;

[0065] Furthermore, step 4 is specifically as follows:

[0066] Step 4-1: using the voice endpoint detection result obtained by the method described in step 3, determine whether the current frame is a voice frame or a non-voice frame;

[0067] Step 4-2: Perform the first level noise interference suppression process using the main channel and auxiliary channel signals obtained by the method described in step 2 and the judgment information obtained by the method described in step 4-1, specifically:

[0068] Step 4-2-1: Filter the auxiliary channel signal, and the filter output is:

[0069] y1(n) NLMS =W1 T (n)X1(n)

[0070] Where W1(n) is the filter coefficient at the current time n, and X1(n) is the auxiliary channel input signal at the current time n;

[0071] Step 4-2-2, subtract the main channel signal from the filter output signal to obtain the output signal after noise interference suppression processing;

[0072] Step 4-2-3: If the current frame is a non-speech frame, the filter coefficients are updated. The update formula is:

[0073]

[0074] Wherein, e1(n) is the signal obtained by subtracting the main channel signal from the filter output signal, and c is a very small constant. In the specific example of the present invention, the corrected step constant μ1 is 0.64, and the constant c is 0.0001;

[0075] Step 5: Use the voice signal of the mouth microphone channel obtained in step 1 to perform the main voice endpoint detection to determine whether the current period is the main voice period or the non-main voice period;

[0076] Furthermore, step 5 is specifically as follows:

[0077] Step 5-1: Calculate the short-time subband spectral entropy of each frame signal using the speech signal of the mouth microphone channel obtained by the method described in step 1, specifically:

[0078] Step 5-1-1: Divide the current frame signal into N b2 subbands. In the specific example of the present invention, the number of subbands N b2 is 40, calculate the subband energy of each subband, and obtain the subband energy E of the mth subband in the i-th frame b2 (m,i), 1≤m≤N b2 ;

[0079] Step 5-1-2: Using the subband energy obtained by the method described in step 5-1-1, the subband energy probability of the mth subband in the i-th frame is calculated as:

[0080]

[0081] Step 5-1-3, using the sub-band energy probability obtained by the method described in step 5-1-2, calculate the sub-band spectral entropy of the i-th frame as:

[0082]

[0083] Step 5-2: Using the short-time subband spectral entropy obtained by the method described in step 5-1, two thresholds are set, and a double threshold method is used to perform speech endpoint detection, specifically:

[0084] Step 5-2-1, set a lower threshold T2', make a rough judgment, the speech below the T2' threshold is definitely the original speech, and the starting and ending points of the original speech are outside the time points corresponding to the intersection of the threshold and the sub-band spectral entropy envelope;

[0085] Step 5-2-2, set a higher threshold T1′, and search to the left and right of the roughly judged speech start and end points obtained by the method described in step 5-2-1, and find the two points where the subband spectral entropy envelope intersects with the threshold T1′, and obtain the start and end point positions of the main body speech determined by the double threshold method.

[0086] Step 6. Utilize the voice signal of the mouth microphone channel obtained in step 1, the noise-suppressed voice signal obtained in step 4, and the decision information obtained in step 5, and use the noise-suppressed voice signal as the main channel input signal of the second-stage main body voice interference suppression unit, and the voice signal of the mouth microphone channel as the auxiliary channel input signal. According to the decision information, adaptively adjust the filter coefficient to cancel the main body voice interference, obtain the enhanced signal of the other party's voice, and broadcast it through the built-in speaker of the protective earmuffs.

[0087] Furthermore, step 6 is specifically as follows:

[0088] Step 6-1, using the voice endpoint detection result obtained by the method described in step 5, determine whether the current frame is a body voice frame or a non-body voice frame;

[0089] Step 6-2: Using the voice signal of the mouth microphone channel obtained by the method described in step 1, the noise-suppressed voice signal obtained by the method described in step 4, and the judgment information obtained by the method described in step 6-1, the noise-suppressed voice signal is used as the main channel input signal of the second-stage main body voice interference suppression unit, and the voice signal of the mouth microphone channel is used as the auxiliary channel input signal to perform the second-stage main body voice interference suppression processing, specifically:

[0090] Step 6-2-1: Filter the auxiliary channel signal, and the filter output is:

[0091] y2(n) NLMS =W2 T (n)X2(n)

[0092] Where W2(n) is the filter coefficient at the current time n, and X2(n) is the auxiliary channel input signal at the current time n;

[0093] Step 6-2-2, subtract the main channel signal from the filter output signal to obtain the output signal after noise interference suppression processing;

[0094] Step 6-2-3: If the current frame is a native speech frame, the filter coefficients are updated. The update formula is:

[0095]

[0096] Wherein, e2(n) is the signal obtained by subtracting the main channel signal from the filter output signal, and c is a very small constant. In a specific embodiment of the present invention, the corrected step constant μ2 is 0.64, and the constant c is 0.0001.

[0097] Figure 2 The schematic diagram of the three-element microphone array for collecting noisy speech data is given. First, the left ear microphone, right ear microphone, and mouth microphone are marked as microphones 1, 2, and 3 respectively, and microphones 1 and 2 are symmetrically placed on the left and right earmuffs of the protective earmuffs, and a boom is installed on the protective earmuffs, and microphone 3 is placed on the boom.

[0098] Figure 3 In the figure, (a), (b) and (c) are the time domain waveforms of the noisy speech signals of the left ear microphone channel, the right ear microphone channel and the mouth microphone channel and the corresponding time-frequency spectrum diagrams. It can be seen that the speech signals of the left ear microphone channel and the right ear microphone channel are mostly submerged in the noise, especially the low-frequency part of the speech signal. Due to its special response characteristics and the fact that it is close to the mouth, the mouth microphone channel can collect a relatively pure body speech signal.

[0099] Figure 4 In the figure, (a), (b) and (c) are the time domain waveforms and corresponding time-frequency spectrograms of the noisy speech signals of the left ear microphone channel, the right ear microphone channel and the mouth microphone channel after preprocessing. It can be seen that after the pre-emphasis processing, most of the low-frequency noise of the noisy speech signal is suppressed and the speech signal is enhanced.

[0100] Figure 5 The time domain waveform of the main channel speech signal output by the fixed beamforming unit and the corresponding time-frequency spectrum diagram. It can be seen that after the noisy speech signals of the left ear microphone channel and the right ear microphone channel are fixedly beamformed, the noise signal is further suppressed and the speech signal is enhanced.

[0101] Figure 6 In the figure, (a) is the time domain waveform of the noisy speech signal input to the detection control unit, that is, the waveform of the main channel speech signal output by the beamforming unit, (b) is the speech signal waveform after the spectral subtraction improved by multi-window spectral estimation, and (c) is a schematic diagram of the corresponding spectral entropy values ​​obtained by calculating the subband spectral entropy value of each frame of the processed speech signal. It can be seen that the detection control unit accurately detects the start and end points of the speech.

[0102] Figure 7The figure is the time domain waveform of the speech signal output by the noise interference suppression unit and the corresponding time-frequency spectrum. It can be seen that the noise interference suppression unit effectively suppresses the environmental noise interference and outputs a relatively pure and undistorted speech signal.

[0103] Figure 8 In the figure, (a) is the time domain waveform of the noisy speech signal input to the detection control unit, that is, the speech signal waveform of the mouth microphone channel, and (b) is a schematic diagram of the corresponding spectral entropy value obtained by calculating the sub-band spectral entropy value of each frame of the speech signal. It can be seen that the detection control unit accurately detects the starting and ending points of the main speech.

[0104] Fig. 9 The figure is the time domain waveform of the speech signal output by the main speech interference suppression unit and the corresponding time-frequency spectrum. It can be seen that the main speech interference suppression unit effectively suppresses the main speech signal and outputs a relatively pure and undistorted other party's speech signal.

Claims

1. A method for enhancing the other party's voice when having a face-to-face conversation with protective earmuffs in a high-noise workshop, characterized in that: The steps include: Step 1: using a microphone array consisting of a left ear microphone, a left ear microphone, and a mouth microphone to synchronously collect noisy speech signals of the subject and the opposite talker during face-to-face conversation in a high noise environment, and preprocessing the collected signals; Step 2: Using the speech signals of the left ear microphone channel and the right ear microphone channel obtained in step 1, fixed beam forming is performed to obtain the main channel and auxiliary channel input signals of the first-stage noise interference suppression unit; Step 3: Use the main channel voice signal obtained in step 2 to perform voice endpoint detection to determine whether the current period is a voice period or a non-voice period; Step 4: Using the main channel and auxiliary channel voice signals obtained in step 2 and the decision information obtained in step 3, the filter of the first-stage noise interference suppression unit adaptively adjusts the filter coefficient to cancel the environmental noise interference; Step 5: Use the voice signal of the mouth microphone channel obtained in step 1 to perform the main voice endpoint detection to determine whether the current period is the main voice period or the non-main voice period; Step 6. Utilize the voice signal of the mouth microphone channel obtained in step 1, the noise-suppressed voice signal obtained in step 4, and the decision information obtained in step 5, and use the noise-suppressed voice signal as the main channel input signal of the second-stage main body voice interference suppression unit, and the voice signal of the mouth microphone channel as the auxiliary channel input signal. According to the decision information, adaptively adjust the filter coefficient to cancel the main body voice interference, obtain the enhanced signal of the other party's voice, and broadcast it through the built-in speaker of the protective earmuffs.

2. A method for enhancing the other party's voice when having a face-to-face conversation with protective earmuffs in a high-noise workshop according to claim 1, characterized in that: The specific process of step 1 is as follows: Step 1-1, number the left ear microphone, left ear microphone, and mouth microphone as 1, 2, and 3 respectively, wherein microphones 1 and 2 are located at the left and right earmuffs of the protective earmuffs, and microphone 3 is located at the extended mouth stem connected to the protective earmuffs, ensuring that microphones 1 and 2 are symmetrically distributed on both sides of microphone 3; Step 1-2: Using the array setting of step 1-1, the collected noisy speech signals of each channel are pre-emphasized using a first-order FIR high-pass digital filter. The filter transfer function is: H(z)=1-z -1 0.9<a<1.0 Where a is the pre-emphasis coefficient, z -1 is the unit delay operator in the z-transform; Step 1-3, performing short-time frame division and windowing processing on the noisy speech signals of each channel obtained by step 1-2, wherein the frame shift is half of the frame length and the window function is a Hamming window.

3. The method for enhancing the other party's voice when having a face-to-face conversation with protective earmuffs in a high-noise workshop according to claim 2, characterized in that: The specific process of step 2 is as follows: The signals of the left and right ear microphone channels obtained by the method described in step 1 are preprocessed and fixed beamformed. The output y(n) of the fixed beamformer is defined as: Where M represents the number of elements in the microphone array, x i (n) represents the speech signal collected by the i-th array element, where i = 0, 1, ..., M-1, w i represents the weight of the i-th filter, τ i Represents the delay estimate between the ith array element and the relative reference array element.

4. The method for enhancing the other party's voice when having a face-to-face conversation with protective earmuffs in a high-noise workshop according to claim 3, characterized in that: The specific process of step 3 is as follows: Step 3-1: The main channel signal obtained by the method described in step 2 is subjected to noise reduction processing using an improved spectral subtraction method of multi-window spectrum estimation, specifically: Step 3-1-1: Perform FFT on the current frame signal and calculate its amplitude spectrum |X i (k)| and the phase spectrum θ i (k) Smoothing is performed between adjacent frames. Specifically, M frames are taken before and after the i frame, and a total of 2M+1 frames are averaged. The average amplitude spectrum is calculated as: Step 3-1-2, perform multi-window spectrum estimation on the current frame signal, calculate its multi-window spectrum power spectrum density P(k,i), i represents the i-th frame, k represents the k-th spectrum line, and perform smoothing between adjacent frames. Specifically, take M frames before and after the i-th frame as the center, and average a total of 2M+1 frames. The smoothed power spectrum density is calculated as: Step 3-1-3: Calculate the average power spectrum density of the noise based on the noise frame of the leading silent segment: Where NIS is the number of frames occupied by the leading silent segment, P y (k,i) is the smoothed power spectrum density of the i-th frame obtained by the method described in step 3-1-2; Step 3-1-4, using the average amplitude spectrum obtained by the method described in step 3-1-1, the smoothed power spectrum density obtained by the method described in step 3-1-2, and the noise average power spectrum density value obtained by the method described in step 3-1-4, and using the spectrum subtraction relationship to calculate the gain factor, the gain factor is defined as: In the formula, α is the over-reduction factor, β is the gain compensation factor, P y (k,i) is the smoothed power spectrum density of the i-th frame obtained by the method described in step 3-1-2, P n (k) is the noise average power spectral density value of the noise frame of the leading silent segment obtained by the method described in step 3-1-2; Step 3-1-5, using the average amplitude spectrum obtained by the method described in step 3-1-1 and the gain factor obtained by the method described in step 3-1-4, obtain the speech signal after noise reduction: Step 3-2: Calculate the short-time subband spectral entropy of each frame signal using the de-noised speech signal obtained by the method described in step 3-1, specifically: Step 3-2-1: Divide the current frame signal into N b1 subbands, calculate the subband energy of each subband, and obtain the subband energy E of the mth subband in the i-th frame b1 (m,i), 1≤m≤N b1 ; Step 3-2-2: Using the subband energy obtained by the method described in step 3-2-1, the subband energy probability of the mth subband in the i-th frame is calculated as: Step 3-2-3, using the sub-band energy probability obtained by the method described in step 3-2-2, calculate the sub-band spectral entropy of the i-th frame as: Step 3-3: Using the short-time subband spectral entropy obtained by the method described in step 3-2, two thresholds are set, and a double threshold method is used to perform speech endpoint detection, specifically: Step 3-3-1, set a lower threshold T2, make a rough judgment, the value below the T2 threshold is speech, and the start and end points of speech are outside the time points corresponding to the intersection of the threshold and the sub-band spectral entropy envelope; Step 3-3-2, set a higher threshold T1, and search left and right respectively from the roughly judged speech start and end points obtained by the method described in step 3-3-1, and find the two points where the subband spectral entropy envelope intersects with the threshold T1, and obtain the speech start and end point positions determined by the double threshold method.

5. The method for enhancing the other party's voice when having a face-to-face conversation with protective earmuffs in a high-noise workshop according to claim 4, characterized in that: The specific process of step 4 is as follows: Step 4-1, using the voice endpoint detection result obtained by the method described in step 3, determine whether the current frame is a voice frame or a non-voice frame; Step 4-2: Perform the first level noise interference suppression process using the main channel and auxiliary channel signals obtained by the method described in step 2 and the judgment information obtained by the method described in step 4-1, specifically: Step 4-2-1: Filter the auxiliary channel signal, and the filter output is: y1(n) NLMS =W1 T (n)X1(n) Where W1(n) is the filter coefficient at the current time n, and X1(n) is the auxiliary channel input signal at the current time n; Step 4-2-2, subtract the main channel signal from the filter output signal to obtain the output signal after noise interference suppression processing; Step 4-2-3: If the current frame is a non-speech frame, the filter coefficients are updated. The update formula is: Where μ1 is the step size factor, e1(n) is the signal obtained by subtracting the main channel signal from the filter output signal, and c is a very small constant.

6. The method for enhancing the other party's voice when having a face-to-face conversation with protective earmuffs in a high-noise workshop according to claim 5, characterized in that: The specific process of step 5 is as follows: Step 5-1: Calculate the short-time subband spectral entropy of each frame signal using the speech signal of the mouth microphone channel obtained by the method described in step 1, specifically: Step 5-1-1: Divide the current frame signal into N b2 subbands, calculate the subband energy of each subband, and obtain the subband energy E of the mth subband in the i-th frame b2 (m,i), 1≤m≤N b2 ; Step 5-1-2: Using the subband energy obtained by the method described in step 5-1-1, the subband energy probability of the mth subband in the i-th frame is calculated as: Step 5-1-3, using the sub-band energy probability obtained by the method described in step 5-1-2, calculate the sub-band spectral entropy of the i-th frame as: Step 5-2: Using the short-time subband spectral entropy obtained by the method described in step 5-1, two thresholds are set, and a double threshold method is used to perform speech endpoint detection, specifically: Step 5-2-1, set a lower threshold T2', make a rough judgment, the speech below the T2' threshold is definitely the original speech, and the starting and ending points of the original speech are outside the time points corresponding to the intersection of the threshold and the sub-band spectral entropy envelope; Step 5-2-2, set a higher threshold value T1′, and search to the left and right of the roughly judged speech start and end points obtained by the method described in step 5-2-1, and find the two points where the subband spectral entropy envelope intersects with the threshold value T1′, and obtain the start and end point positions of the main body speech determined by the double threshold method.

7. A method for enhancing the other party's voice when having a face-to-face conversation with protective earmuffs in a high-noise workshop according to claim 6, characterized in that: The specific process of step 6 is as follows: Step 6-1, using the voice endpoint detection result obtained by the method described in step 5, determine whether the current frame is a body voice frame or a non-body voice frame; Step 6-2: Using the voice signal of the mouth microphone channel obtained by the method described in step 1, the noise-suppressed voice signal obtained by the method described in step 4, and the judgment information obtained by the method described in step 6-1, the noise-suppressed voice signal is used as the main channel input signal of the second-stage main body voice interference suppression unit, and the voice signal of the mouth microphone channel is used as the auxiliary channel input signal to perform the second-stage main body voice interference suppression processing, specifically: Step 6-2-1: Filter the auxiliary channel signal, and the filter output is: y2(n) NLMS =W2 T (n)X2(n) Where W2(n) is the filter coefficient at the current time n, and X2(n) is the auxiliary channel input signal at the current time n; Step 6-2-2, subtract the main channel signal from the filter output signal to obtain the output signal after noise interference suppression processing; Step 6-2-3: If the current frame is a native speech frame, the filter coefficients are updated. The update formula is: Wherein, e2(n) is the signal obtained by subtracting the main channel signal from the filter output signal, and c is a very small constant.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images