Digital hearing aid low-complexity wind noise detection method and system based on joint criterion
By using a joint criterion-based approach and dual-microphone signal processing technology, the computational complexity of wind noise detection in digital hearing aids is reduced, enabling efficient wind noise detection in outdoor environments and improving robustness and efficiency.
Patent Information
- Application Number
- CN202411479966.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-23
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-10-23
AI Technical Summary
Existing wind noise detection algorithms for digital hearing aids are highly complex and difficult to efficiently detect wind noise in complex outdoor environments.
A joint criterion-based method is adopted, which calculates the relevant information of the two microphones and uses the single microphone signal to control the calculation process of the dual microphone information. This includes steps such as signal framing, DC removal, FFT calculation, long-term steady-state energy judgment, and frequency domain correlation sequence analysis, thereby reducing the amount of computation and complexity.
It enables efficient detection of wind noise in different outdoor environments, reduces the computational load and complexity of wind noise detection algorithms, and improves the robustness and efficiency of detection.
Smart Images

Figure CN119012102A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of signal processing, and in particular to a low-complexity wind noise detection method and system for a digital hearing aid based on a joint criterion. Background Art
[0002] As the phenomenon of social aging becomes more and more serious, the number of hearing-impaired people is still increasing now and in the future. Hearing aids can conveniently help hearing-impaired people improve their hearing conditions, so that hearing-impaired patients of different ages can quickly compensate for hearing loss to varying degrees. Digital hearing aids have significant advantages over traditional analog hearing aids, such as higher sound reproduction, higher flexibility, and longer life. On the other hand, digital hearing aids can support more complex and effective hearing aid algorithms, such as the ability to identify and adapt to different environments and listening conditions, which greatly expands the functionality and practicality of hearing aids. Programmable digital hearing aids have broad development space in the field of hearing devices, so the research on digital hearing aid algorithms has a broad and far-reaching impact.
[0003] At present, the listening environment of hearing aids is complex and diverse, and the environment outdoors is even more changeable. For example, when wind noise occurs, the auditory experience of the hearing aid will be severely reduced, so it is necessary to detect the wind noise in order to suppress it. However, the current wind noise detection algorithm is generally very complex; therefore, it is necessary to design a low-complexity wind noise detection method and system for digital hearing aids based on joint criteria. Summary of the invention
[0004] The purpose of the present invention is to overcome the shortcomings of the prior art and to better and more effectively solve the problem of high complexity of wind noise detection algorithms for wind noise detection so as to suppress it. A low-complexity wind noise detection method and system for digital hearing aids based on joint criteria is provided, which realizes the function of detecting wind noise by calculating dual-microphone related information in different outdoor listening environments, and by calculating multiple statistics as criteria for detecting wind noise, it is possible to detect different degrees of wind noise in different environments. The calculation process of controlling dual-microphone information using the signal collected by a single microphone greatly reduces the calculation amount and complexity of the wind noise detection algorithm compared with the traditional dual-microphone wind noise detection algorithm.
[0005] In order to achieve the above object, the technical solution adopted by the present invention is: A low-complexity wind noise detection method for a digital hearing aid based on a joint criterion comprises the following steps: Step A, collecting the first microphone continuous original signal The continuous original signal of the second microphone collected by the second microphone After framing, the DC offset components are removed to obtain the first time domain frame signal after DC removal. And the second time domain frame signal after DC removal ; Step B, calculate the first time domain frame signal after DC removal The long-term steady-state energy, and then the first time domain frame signal after DC And the second time domain frame signal after DC removal Calculate FFT to obtain the frequency domain information of the two signals; Step C, judging whether the long-term steady-state energy exceeds a first fixed threshold value, if not, directly setting the probability of occurrence of the first wind noise to 0, and no longer calculating the frequency domain correlation sequence in the frequency domain information of the two-way signal, if exceeding the first fixed threshold value, continuing to calculate the frequency point logarithmic energy difference in the frequency domain information of the two-way signal; Step D, determining whether the frequency point logarithmic energy difference exceeds a second fixed threshold value, if not exceeding the second fixed threshold value, directly setting the probability of occurrence of the first wind noise to 0, and no longer calculating the frequency domain correlation sequence in the frequency domain information of the two-way signal; if exceeding the second fixed threshold value, continuing to calculate the average value of the frequency domain correlation sequence in the frequency domain information of the two-way signal; Step E: limiting and clipping the average value of the frequency domain correlation sequence to obtain the probability of wind noise existence, and then processing the probability of wind noise existence through a first-order infinite impulse response filter to output the second wind noise occurrence probability.
[0006] The aforementioned low-complexity wind noise detection method for a digital hearing aid based on a joint criterion, step A, the first microphone continuous original signal collected by the first microphone The continuous original signal of the second microphone collected by the second microphone After framing, the DC offset components are removed to obtain the first time domain frame signal after DC removal. And the second time domain frame signal after DC removal , where a high-pass filter is used to remove the DC offset component after framing.
[0007] The aforementioned low-complexity wind noise detection method for digital hearing aids based on joint criteria, step B, calculates the first time domain frame signal after DC removal The long-term steady-state energy, and then the first time domain frame signal after DC And the second time domain frame signal after DC removal Calculate FFT to obtain the frequency domain information of the two signals. The specific steps are as follows: Step B1, calculate the first time domain frame signal after DC removal The long-term steady-state energy is as follows: Step B11, calculate the time domain signal energy of the mth frame As shown in formula (1), (1) in, is the number of samples per frame, is the square of the sampling value of the nth sample signal; Step B12: Count the average energy of the historical N_Frame frame time domain signal at the mth frame As shown in formula (2), (2) in, is the time domain signal energy of the nth frame, The energy sum of the time domain signals from the m-N_Frame+1th frame to the mth frame, a total of N_Frame frames; Step B13, update long-term steady-state energy , as shown in formula (3), (3) in, and is the smoothing coefficient for updating the long-term steady-state energy; Step B2: the first time domain frame signal after DC And the second time domain frame signal after DC removal Calculate FFT to obtain the frequency domain information of the two signals, where the frequency domain information of the two signals includes a spectrum sequence and a power spectrum. The spectrum sequence is and , the power spectrum is and , and the power spectrum and The calculation process of is shown in formula (4) and formula (5). (4) (5).
[0008] The aforementioned low-complexity wind noise detection method for digital hearing aids based on joint criteria, step C, determines the long-term steady-state energy Whether it exceeds the first fixed threshold, if it does not exceed the first fixed threshold, the probability of the first wind noise occurring is directly set to 0, and the frequency domain correlation sequence in the frequency domain information of the two-way signal is no longer calculated; if it exceeds the first fixed threshold, the frequency point logarithmic energy difference in the frequency domain information of the two-way signal is continued to be calculated. The specific steps are as follows: Step C1, if the long-term steady-state energy Does not exceed the first fixed threshold , then directly set the probability of the first wind noise occurring is 0, as shown in formula (6), (6) Step C2, if the long-term steady-state energy Exceeding the first fixed threshold , then continue to calculate the frequency point logarithmic energy difference in the frequency domain information of the two signals , where the frequency point logarithmic energy difference in the frequency domain information of the two signals is calculated by converting the power spectrum and After converting to logarithm, the difference is calculated, as shown in formula (7): (7) in, is the first frequency point number participating in the power spectrum summation, wherein the first frequency point number represents the number of low-frequency cutoff frequency points, is the number of second frequency points participating in the power spectrum summation, where the second frequency point number represents the number of high-frequency cutoff frequency points.
[0009] The aforementioned low-complexity wind noise detection method for digital hearing aids based on joint criteria, step D, determines the frequency point logarithmic energy difference Whether it exceeds the second fixed threshold, if it does not exceed the second fixed threshold, the probability of the first wind noise occurring is directly set to 0, and the frequency domain correlation sequence in the frequency domain information of the two-way signal is no longer calculated; if it exceeds the second fixed threshold, the average value of the frequency domain correlation sequence in the frequency domain information of the two-way signal is continued to be calculated. The specific steps are as follows: Step D1: if it does not exceed the second fixed threshold , then directly set the probability of the first wind noise occurring is 0, as shown in formula (8), (8); Step D2: if the second fixed threshold is exceeded, the frequency correlation in the frequency domain information of the two channels of signals is continued to be calculated. The specific steps are as follows: Step D21, according to the spectrum sequence and Calculate the frequency domain correlation sequence of two signals , as shown in formula (9), (9) Among them, * is conjugated; Step D22, frequency domain correlation sequence Normalize and get the normalized frequency domain correlation sequence , as shown in formula (10), (10); Step D23, intercepting a portion of the normalized frequency domain correlation sequence And calculate the average value of the frequency domain correlation series , as shown in formula (11), (11); in, The number of frequency points intercepted to calculate the average value of the frequency domain correlation sequence.
[0010] In the aforementioned low-complexity wind noise detection method for digital hearing aids based on joint criteria, step E is to average the frequency domain correlation sequence. The wind noise existence probability is obtained by limiting and cutting, and then the wind noise existence probability is processed by a first-order infinite impulse response filter to output the second wind noise occurrence probability. , the specific steps are as follows, Step E1, average the frequency domain correlation sequence Perform shearing, as shown in formula (12), (12) in, is the lower limit threshold of the average value of the preset frequency domain correlation sequence; Step E2, average the frequency domain correlation sequence The limiting is performed as shown in formula (13): (13) in, is the upper threshold of the average value of the preset frequency domain correlation sequence; Step E3: Process the wind noise existence probability through a first-order infinite impulse response filter and output the second wind noise occurrence probability , the probability of the second wind noise occurring is It is used to characterize the possibility of wind noise in the current environment. The transfer function of the first-order infinite impulse response filter is shown in formula (14): (14) in, and are constants, For the previous frame .
[0011] A low-complexity wind noise detection system for a digital hearing aid based on a joint criterion, comprising a signal framing and DC removal module, a signal calculation module, a first threshold judgment module, a second threshold judgment module and a wind noise output module, wherein the signal framing and DC removal module is used to frame a first microphone continuous original signal collected by a first microphone and a second microphone continuous original signal collected by a second microphone, and then remove the DC offset component respectively, to obtain a first time domain framed signal after DC removal and a second time domain framed signal after DC removal; The signal calculation module is used to calculate the long-term steady-state energy of the first time domain frame signal after DC removal, and then calculate FFT on the first time domain frame signal after DC removal and the second time domain frame signal after DC removal to obtain frequency domain information of the two signals; The first threshold judgment module is used to judge whether the long-term steady-state energy exceeds the first fixed threshold. If it does not exceed the first fixed threshold, the probability of wind noise occurrence is directly set to 0, and the frequency domain correlation sequence in the frequency domain information of the two-way signal is no longer calculated. If it exceeds the first fixed threshold, the frequency point logarithmic energy difference in the frequency domain information of the two-way signal continues to be calculated; The second threshold judgment module is used to judge whether the frequency point logarithmic energy difference exceeds the second fixed threshold. If it does not exceed the second fixed threshold, the probability of wind noise occurrence is directly set to 0, and the frequency domain correlation sequence in the frequency domain information of the two-way signal is no longer calculated. If it exceeds the second fixed threshold, the average value of the frequency domain correlation sequence in the frequency domain information of the two-way signal continues to be calculated; The wind noise output module is used to limit and cut the average value of the frequency domain correlation sequence to obtain the probability of wind noise existence, and then process the wind noise existence probability through a first-order infinite impulse response filter to output the wind noise occurrence probability.
[0012] The beneficial effect of the present invention is as follows: a low-complexity wind noise detection method and system for a digital hearing aid based on a joint criterion of the present invention, firstly, the continuous original signal of the first microphone collected by the first microphone and the continuous original signal of the second microphone collected by the second microphone are framed, and the DC offset component is removed respectively to obtain the first time domain framed signal after DC removal and the second time domain framed signal after DC removal, then the long-term steady-state energy of the first time domain framed signal after DC removal is calculated, and then the FFT of the first time domain framed signal after DC removal and the second time domain framed signal after DC removal is calculated to obtain the frequency domain information of the two signals, and then whether the long-term steady-state energy exceeds the first fixed threshold value is judged, if it does not exceed the first fixed threshold value, the probability of occurrence of the first wind noise is directly set to 0, and the frequency domain correlation sequence in the frequency domain information of the two signals is no longer calculated, if it exceeds the first fixed threshold value, the frequency point logarithmic energy difference in the frequency domain information of the two signals is continued to be calculated, and then whether the frequency point logarithmic energy difference exceeds the second fixed threshold value is judged, if it does not exceed the second fixed threshold value , then the first wind noise occurrence probability is directly set to 0, and the frequency domain correlation sequence in the frequency domain information of the two-way signal is no longer calculated. If it exceeds the second fixed threshold, the frequency domain correlation sequence average value in the frequency domain information of the two-way signal is continued to be calculated, and finally the frequency domain correlation sequence average value is limited and clipped to obtain the wind noise existence probability, and then the wind noise existence probability is processed by a first-order infinite impulse response filter to output the second wind noise occurrence probability; the present invention realizes the function of detecting wind noise by calculating dual-microphone related information in different outdoor listening environments, and by calculating multiple statistics as a criterion for detecting wind noise, it can detect different degrees of wind noise in different environments, and the calculation process of controlling the dual-microphone information using the signal collected by a single microphone greatly reduces the calculation amount and complexity of the wind noise detection algorithm compared with the traditional dual-microphone wind noise detection algorithm, which is not only more suitable for the design of digital hearing aids, but also has good application prospects, and improves the robustness and efficiency of wind noise detection. BRIEF DESCRIPTION OF THE DRAWINGS
[0013] Figure 1 It is a flow chart of a low-complexity wind noise detection method for a digital hearing aid based on a joint criterion of the present invention; Figure 2 It is a comparison diagram of the original input signal and the output wind noise existence probability of wind noise detection in an embodiment of the present invention. DETAILED DESCRIPTION
[0014] The present invention will be further described below in conjunction with the accompanying drawings.
[0015] like Figure 1 As shown, a low-complexity wind noise detection method and system for a digital hearing aid based on a joint criterion of the present invention comprises the following steps: Step A, collecting the first microphone continuous original signal The continuous original signal of the second microphone collected by the second microphone After framing, the DC offset components are removed to obtain the first time domain frame signal after DC removal. And the second time domain frame signal after DC removal , where a high-pass filter is used to remove the DC offset component after framing.
[0016] Step B, calculate the first time domain frame signal after DC removal The long-term steady-state energy, and then the first time domain frame signal after DC And the second time domain frame signal after DC removal Calculate FFT to obtain the frequency domain information of the two signals. The specific steps are as follows: Step B1, calculate the first time domain frame signal after DC removal The long-term steady-state energy is as follows: Step B11, calculate the time domain signal energy of the mth frame As shown in formula (1), (1) in, is the number of samples per frame, is the square of the sampling value of the nth sample signal; Step B12: Count the average energy of the historical N_Frame frame time domain signal at the mth frame As shown in formula (2), (2) in, is the time domain signal energy of the nth frame, The energy sum of the time domain signals from the m-N_Frame+1th frame to the mth frame, a total of N_Frame frames; Step B13, update long-term steady-state energy , as shown in formula (3), (3) in, and is the smoothing coefficient for updating the long-term steady-state energy; Step B2: the first time domain frame signal after DC And the second time domain frame signal after DC removal Calculate FFT to obtain the frequency domain information of the two signals, where the frequency domain information of the two signals includes a spectrum sequence and a power spectrum. The spectrum sequence is and , the power spectrum is and , and the power spectrum and The calculation process of is shown in formula (4) and formula (5). (4) (5).
[0017] Step C, determine the long-term steady-state energy Whether it exceeds the first fixed threshold, if it does not exceed the first fixed threshold, the probability of the first wind noise occurring is directly set to 0, and the frequency domain correlation sequence in the frequency domain information of the two-way signal is no longer calculated; if it exceeds the first fixed threshold, the frequency point logarithmic energy difference in the frequency domain information of the two-way signal is continued to be calculated. The specific steps are as follows: Step C1, if the long-term steady-state energy Does not exceed the first fixed threshold , then directly set the probability of the first wind noise occurring is 0, as shown in formula (6), (6) Step C2, if the long-term steady-state energy Exceeding the first fixed threshold , then continue to calculate the frequency point logarithmic energy difference in the frequency domain information of the two signals , where the frequency point logarithmic energy difference in the frequency domain information of the two signals is calculated by converting the power spectrum and After converting to logarithm, the difference is calculated, as shown in formula (7): (7) in, is the first frequency point number participating in the power spectrum summation, wherein the first frequency point number represents the number of low-frequency cutoff frequency points, is the number of second frequency points participating in the power spectrum summation, where the second frequency point number represents the number of high-frequency cutoff frequency points.
[0018] Step D: Determine the frequency point logarithmic energy difference Whether it exceeds the second fixed threshold, if it does not exceed the second fixed threshold, the probability of the first wind noise occurring is directly set to 0, and the frequency domain correlation sequence in the frequency domain information of the two-way signal is no longer calculated; if it exceeds the second fixed threshold, the average value of the frequency domain correlation sequence in the frequency domain information of the two-way signal is continued to be calculated. The specific steps are as follows: Step D1: if it does not exceed the second fixed threshold , then directly set the probability of the first wind noise occurring is 0, as shown in formula (8), (8); Step D2: if the second fixed threshold is exceeded, the frequency correlation in the frequency domain information of the two signals is continued to be calculated. The specific steps are as follows: Step D21, according to the spectrum sequence and Calculate the frequency domain correlation sequence of two signals , as shown in formula (9), (9) Among them, * is conjugated; Step D22, frequency domain correlation sequence Normalize and get the normalized frequency domain correlation sequence , as shown in formula (10), (10); Step D23, intercepting a portion of the normalized frequency domain correlation sequence And calculate the average value of the frequency domain correlation series , as shown in formula (11), (11); in, The number of frequency points intercepted to calculate the average value of the frequency domain correlation sequence.
[0019] Step E: average the frequency domain correlation sequence The wind noise existence probability is obtained by limiting and cutting, and then the wind noise existence probability is processed by a first-order infinite impulse response filter to output the second wind noise occurrence probability. , the specific steps are as follows, Step E1, average the frequency domain correlation sequence Perform shearing, as shown in formula (12), (12) in, is the lower limit threshold of the average value of the preset frequency domain correlation sequence; Step E2, average the frequency domain correlation sequence The limiting is performed as shown in formula (13): (13) in, is the upper threshold of the average value of the preset frequency domain correlation sequence; Step E3: Process the wind noise existence probability through a first-order infinite impulse response filter and output the second wind noise occurrence probability , the probability of the second wind noise occurring is It is used to characterize the possibility of wind noise in the current environment. The transfer function of the first-order infinite impulse response filter is shown in formula (14): (14) in, and are constants, For the previous frame .
[0020] A low-complexity wind noise detection system for a digital hearing aid based on a joint criterion, comprising a signal framing and DC removal module, a signal calculation module, a first threshold judgment module, a second threshold judgment module and a wind noise output module, wherein the signal framing and DC removal module is used to frame a first microphone continuous original signal collected by a first microphone and a second microphone continuous original signal collected by a second microphone, and then remove the DC offset component respectively, to obtain a first time domain framed signal after DC removal and a second time domain framed signal after DC removal; The signal calculation module is used to calculate the long-term steady-state energy of the first time domain frame signal after DC removal, and then calculate FFT on the first time domain frame signal after DC removal and the second time domain frame signal after DC removal to obtain frequency domain information of the two signals; The first threshold judgment module is used to judge whether the long-term steady-state energy exceeds the first fixed threshold. If it does not exceed the first fixed threshold, the probability of wind noise occurrence is directly set to 0, and the frequency domain correlation sequence in the frequency domain information of the two-way signal is no longer calculated. If it exceeds the first fixed threshold, the frequency point logarithmic energy difference in the frequency domain information of the two-way signal continues to be calculated; The second threshold judgment module is used to judge whether the frequency point logarithmic energy difference exceeds the second fixed threshold. If it does not exceed the second fixed threshold, the probability of wind noise occurrence is directly set to 0, and the frequency domain correlation sequence in the frequency domain information of the two-way signal is no longer calculated. If it exceeds the second fixed threshold, the average value of the frequency domain correlation sequence in the frequency domain information of the two-way signal continues to be calculated; The wind noise output module is used to limit and cut the average value of the frequency domain correlation sequence to obtain the probability of wind noise existence, and then process the wind noise existence probability through a first-order infinite impulse response filter to output the wind noise occurrence probability.
[0021] In order to further illustrate the use effect of the present invention, Figure 2 The figure shows a comparison diagram of the original input signal and the output wind noise existence probability of wind noise detection in an embodiment of the present invention. It can be seen from the figure that the upper channel is the input signal collected by one of the microphones, and the lower channel is the output wind noise existence probability; when there is no voice or wind noise at the beginning, the wind noise existence probability remains at 0. When wind noise of different intensities appears, the wind noise existence probability curve can follow and reflect whether wind noise exists in the current environmental state and the intensity of the detected wind noise, thereby achieving the purpose of detecting wind noise.
[0022] In summary, the present invention provides a low-complexity wind noise detection method and system for a digital hearing aid based on a joint criterion. First, the continuous original signal of the first microphone collected by the first microphone and the continuous original signal of the second microphone collected by the second microphone are framed, and the DC offset component is removed respectively to obtain the first time domain framed signal after DC removal and the second time domain framed signal after DC removal. Then, the long-term steady-state energy of the first time domain framed signal after DC removal is calculated, and then the FFT of the first time domain framed signal after DC removal and the second time domain framed signal after DC removal is calculated to obtain the frequency domain information of the two signals. Then, it is determined whether the long-term steady-state energy exceeds a first fixed threshold. If it does not exceed the first fixed threshold, the probability of occurrence of the first wind noise is directly set to 0, and the frequency domain correlation sequence in the frequency domain information of the two signals is no longer calculated. If it exceeds the first fixed threshold, the frequency point logarithmic energy difference in the frequency domain information of the two signals is continued to be calculated, and then it is determined whether the frequency point logarithmic energy difference exceeds the second fixed threshold. If it does not exceed the second fixed threshold, Then, the first wind noise occurrence probability is set to 0, and the frequency domain correlation sequence in the frequency domain information of the two-way signal is no longer calculated. If it exceeds the second fixed threshold, the average value of the frequency domain correlation sequence in the frequency domain information of the two-way signal is continued to be calculated. Finally, the average value of the frequency domain correlation sequence is limited and clipped to obtain the probability of wind noise existence. The probability of wind noise existence is then processed by a first-order infinite impulse response filter and then output as the second wind noise occurrence probability. The present invention realizes the function of detecting wind noise by calculating dual-microphone related information in different outdoor listening environments, and by calculating multiple statistics as a criterion for detecting wind noise, different degrees of wind noise can be detected in different environments. Compared with the traditional dual-microphone wind noise detection algorithm, the calculation process of controlling the dual-microphone information using the signal collected by a single microphone greatly reduces the calculation amount and complexity of the wind noise detection algorithm. It is not only more suitable for the design of digital hearing aids, but also has good application prospects, and improves the robustness and efficiency of wind noise detection.
[0023] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.
Claims
1. A low-complexity wind noise detection method for digital hearing aids based on joint criteria. Features: The following steps are included: Step A, collecting the first microphone continuous original signal The continuous original signal of the second microphone collected by the second microphone After framing, the DC offset components are removed to obtain the first time domain frame signal after DC removal. And the second time domain frame signal after DC removal ; Step B, calculate the first time domain frame signal after DC removal The long-term steady-state energy, and then the first time domain frame signal after DC And the second time domain frame signal after DC removal Calculate FFT to obtain the frequency domain information of the two signals; Step C, judging whether the long-term steady-state energy exceeds a first fixed threshold value, if not, directly setting the probability of occurrence of the first wind noise to 0, and no longer calculating the frequency domain correlation sequence in the frequency domain information of the two-way signal, if exceeding the first fixed threshold value, continuing to calculate the frequency point logarithmic energy difference in the frequency domain information of the two-way signal; Step D, determining whether the frequency point logarithmic energy difference exceeds a second fixed threshold value, if not exceeding the second fixed threshold value, directly setting the probability of occurrence of the first wind noise to 0, and no longer calculating the frequency domain correlation sequence in the frequency domain information of the two-way signal; if exceeding the second fixed threshold value, continuing to calculate the average value of the frequency domain correlation sequence in the frequency domain information of the two-way signal; Step E: limiting and clipping the average value of the frequency domain correlation sequence to obtain the probability of wind noise existence, and then processing the probability of wind noise existence through a first-order infinite impulse response filter to output the second wind noise occurrence probability.
2. A low-complexity wind noise detection method for digital hearing aids based on joint criteria according to claim 1, Features: Step A, collecting the first microphone continuous original signal The continuous original signal of the second microphone collected by the second microphone After framing, the DC offset components are removed to obtain the first time domain frame signal after DC removal. And the second time domain frame signal after DC removal , where a high-pass filter is used to remove the DC offset component after framing.
3. The low-complexity wind noise detection method for digital hearing aids based on joint criteria according to claim 2, Features: Step B, calculate the first time domain frame signal after DC removal The long-term steady-state energy, and then the first time domain frame signal after DC And the second time domain frame signal after DC removal Calculate FFT to obtain the frequency domain information of the two signals. The specific steps are as follows: Step B1, calculate the first time domain frame signal after DC removal The long-term steady-state energy is as follows: Step B11, calculate the time domain signal energy of the mth frame As shown in formula (1), (1) in, is the number of samples per frame, is the square of the sampling value of the nth sample signal; Step B12: Count the average energy of the historical N_Frame frame time domain signal at the mth frame As shown in formula (2), (2) in, is the time domain signal energy of the nth frame, The energy sum of the time domain signals from the m-N_Frame+1th frame to the mth frame, a total of N_Frame frames; Step B13, update long-term steady-state energy , as shown in formula (3), (3) in, and is the smoothing coefficient for updating the long-term steady-state energy; Step B2: the first time domain frame signal after DC And the second time domain frame signal after DC removal Calculate FFT to obtain the frequency domain information of the two signals, where the frequency domain information of the two signals includes a spectrum sequence and a power spectrum. The spectrum sequence is and , the power spectrum is and , and the power spectrum and The calculation process of is shown in formula (4) and formula (5). (4) (5)。 4. The low-complexity wind noise detection method for digital hearing aids based on joint criteria according to claim 3, Features: Step C, determine the long-term steady-state energy Whether it exceeds the first fixed threshold, if it does not exceed the first fixed threshold, the probability of the first wind noise occurring is directly set to 0, and the frequency domain correlation sequence in the frequency domain information of the two-way signal is no longer calculated; if it exceeds the first fixed threshold, the frequency point logarithmic energy difference in the frequency domain information of the two-way signal is continued to be calculated. The specific steps are as follows: Step C1, if the long-term steady-state energy Does not exceed the first fixed threshold , then directly set the probability of the first wind noise occurring is 0, as shown in formula (6), (6) Step C2, if the long-term steady-state energy Exceeding the first fixed threshold , then continue to calculate the frequency point logarithmic energy difference in the frequency domain information of the two signals , where the frequency point logarithmic energy difference in the frequency domain information of the two signals is calculated by converting the power spectrum and After converting to logarithm, the difference is calculated, as shown in formula (7): (7) in, is the first frequency point number participating in the power spectrum summation, wherein the first frequency point number represents the number of low-frequency cutoff frequency points, is the number of second frequency points participating in the power spectrum summation, where the second frequency point number represents the number of high-frequency cutoff frequency points.
5. The low-complexity wind noise detection method for digital hearing aids based on joint criteria according to claim 4, Features: Step D: Determine the frequency point logarithmic energy difference Whether it exceeds the second fixed threshold, if it does not exceed the second fixed threshold, the probability of the first wind noise occurring is directly set to 0, and the frequency domain correlation sequence in the frequency domain information of the two-way signal is no longer calculated; if it exceeds the second fixed threshold, the average value of the frequency domain correlation sequence in the frequency domain information of the two-way signal is continued to be calculated. The specific steps are as follows: Step D1: if it does not exceed the second fixed threshold , then directly set the probability of the first wind noise occurring is 0, as shown in formula (8), (8); Step D2: if the second fixed threshold is exceeded, the frequency correlation in the frequency domain information of the two channels of signals is continued to be calculated. The specific steps are as follows: Step D21, according to the spectrum sequence and Calculate the frequency domain correlation sequence of two signals , as shown in formula (9), (9) Among them, * is conjugated; Step D22, frequency domain correlation sequence Normalize and get the normalized frequency domain correlation sequence , as shown in formula (10), (10); Step D23, intercepting a portion of the normalized frequency domain correlation sequence And calculate the average value of the frequency domain correlation series , as shown in formula (11), (11); in, The number of frequency points intercepted to calculate the average value of the frequency domain correlation sequence.
6. The low-complexity wind noise detection method for digital hearing aids based on joint criteria according to claim 5, Features: Step E: average the frequency domain correlation sequence The wind noise existence probability is obtained by limiting and cutting, and then the wind noise existence probability is processed by a first-order infinite impulse response filter to output the second wind noise occurrence probability. , the specific steps are as follows, Step E1, average the frequency domain correlation sequence Perform shearing, as shown in formula (12), (12) in, is the lower limit threshold of the average value of the preset frequency domain correlation sequence; Step E2, average the frequency domain correlation sequence The limiting is performed as shown in formula (13): (13) in, is the upper threshold of the average value of the preset frequency domain correlation sequence; Step E3: Process the wind noise existence probability through a first-order infinite impulse response filter and output the second wind noise occurrence probability , the probability of the second wind noise occurring is It is used to characterize the possibility of wind noise in the current environment. The transfer function of the first-order infinite impulse response filter is shown in formula (14): (14) in, and are constants, For the previous frame .
7. A digital hearing aid low-complexity wind noise detection system based on joint criteria, wherein the detection process of the digital hearing aid low-complexity wind noise detection system is based on the digital hearing aid low-complexity wind noise detection method according to any one of claims 1 to 6, Features: It includes a signal framing and DC removal module, a signal calculation module, a first threshold judgment module, a second threshold judgment module and a wind noise output module, wherein the signal framing and DC removal module is used to remove the DC offset component of the first microphone continuous original signal collected by the first microphone and the second microphone continuous original signal collected by the second microphone after framing, so as to obtain a first time domain framed signal after DC removal and a second time domain framed signal after DC removal; The signal calculation module is used to calculate the long-term steady-state energy of the first time domain frame signal after DC removal, and then calculate FFT on the first time domain frame signal after DC removal and the second time domain frame signal after DC removal to obtain frequency domain information of the two signals; The first threshold judgment module is used to judge whether the long-term steady-state energy exceeds the first fixed threshold. If it does not exceed the first fixed threshold, the probability of wind noise occurrence is directly set to 0, and the frequency domain correlation sequence in the frequency domain information of the two-way signal is no longer calculated. If it exceeds the first fixed threshold, the frequency point logarithmic energy difference in the frequency domain information of the two-way signal continues to be calculated; The second threshold judgment module is used to judge whether the frequency point logarithmic energy difference exceeds the second fixed threshold. If it does not exceed the second fixed threshold, the probability of wind noise occurrence is directly set to 0, and the frequency domain correlation sequence in the frequency domain information of the two-way signal is no longer calculated. If it exceeds the second fixed threshold, the average value of the frequency domain correlation sequence in the frequency domain information of the two-way signal continues to be calculated; The wind noise output module is used to limit and cut the average value of the frequency domain correlation sequence to obtain the probability of wind noise existence, and then process the wind noise existence probability through a first-order infinite impulse response filter to output the wind noise occurrence probability.
Citation Information
Patent Citations
Real-time wind buffet noise detection
CN106024018A
Time-domain self-adaptive speech detection method based on dynamic noise estimation
CN106098076A
Wind noise detection method and wind noise detection device
CN112037806A
Low-complexity dual-microphone directional pickup method for digital hearing aid
CN114501283A
Wind noise detection and suppression
US8781137B1
