Speech wide dynamic range compression method, device, equipment and storage medium
Through multi-channel voice signal processing and wide dynamic range compression technology, the problem of easy perception of background noise under the single-channel method is solved, personalized hearing compensation and speech signal clarity are achieved, helping hearing impaired patients to better identify normal What people say.
Patent Information
- Application Number
- CN202311834658.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-12-27
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2043-12-27
AI Technical Summary
The single-channel wide dynamic range compression method will make the originally smaller background noise more easily perceived after hearing compensation, affecting the speech recognition effect of hearing-impaired patients.
Using multi-channel voice signal processing method, the voice information to be processed is decomposed into a preset number of channel signals, and a wide dynamic range compression process is performed on each channel. The frequency band is divided by the bark scale, and the average channel energy and sound pressure level are calculated. , use a wide dynamic range compression curve for gain calculation and smooth filtering to reduce noise impact.
It realizes personalized hearing compensation, reduces the impact of background noise on hearing-impaired patients, improves the clarity and comfort of speech signals, and allows hearing-impaired patients to hear sounds more clearly that normal people hear.
Smart Images

Figure CN117789735B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of speech signal processing technology, and in particular to a speech wide dynamic range compression method, device, equipment and storage medium. Background Art
[0002] Hearing aids are mainly for patients with sensorineural hearing loss. The hearing impairment of hearing-impaired patients is mainly manifested in increased hearing threshold and narrowed auditory dynamic range. The increased hearing threshold, that is, the minimum sound intensity that can be perceived is higher than that of normal people, and the degree of increase in the hearing threshold at different frequencies is also different, usually the high-frequency part increases more; however, the pain threshold of hearing-impaired patients generally increases very little, or can be considered unchanged, which makes the patient's auditory dynamic range narrower. The main manifestation of the narrowing of the auditory dynamic range of hearing-impaired patients is that the patients cannot hear subtle sounds, or because they cannot hear the high-frequency components of the sound, they can hear others talking but find it difficult to understand what they are saying.
[0003] In order to address the problems of increased hearing thresholds and narrowed auditory dynamic range in hearing-impaired patients, it is necessary to compensate for the loudness of sounds and "map" the sounds within the dynamic range of normal people's hearing to the dynamic range of hearing in hearing-impaired patients. In particular, the high-frequency components that have an important impact on speech recognition should be adequately compensated.
[0004] At present, the single-channel wide dynamic range compression (WDRC) algorithm is commonly used to achieve loudness compensation. When using the single-channel WDRC algorithm to compensate for sound loudness, a fast compression method is generally used, that is, when the sound is strong, the gain drops quickly, and when the sound is weak, the gain rises quickly. However, the use of the single-channel WDRC method will make some originally small background noises easier to perceive after hearing compensation. Summary of the invention
[0005] In order to help solve the problem that a single-channel wide dynamic range compression method makes some originally small background noises more perceptible after hearing compensation, the present application provides a speech wide dynamic range compression method, apparatus, device and storage medium.
[0006] In a first aspect, the present application provides a speech wide dynamic range compression method, which adopts the following technical solution: the method is applied to a speech wide dynamic range compression system, the speech wide dynamic range compression system includes a hearing aid, and the method includes:
[0007] Acquire voice information to be processed, wherein the voice information to be processed includes sound information within a normal hearing dynamic range acquired from the hearing aid;
[0008] Converting the speech information to be processed into a speech signal in the frequency domain and decomposing the signal into a preset number of channel signals;
[0009] Performing wide dynamic range compression processing on each of the channel signals to obtain a channel compression signal;
[0010] Each of the channel compression signals is synthesized into a full-channel frequency domain signal, and converted into a time domain speech signal before output.
[0011] Through the above technical solution, since the hearing loss of hearing-impaired patients varies at different frequencies, personalized hearing compensation can be achieved by decomposing a voice signal into a multi-channel processing method, and independent compression and amplification on different channels can better fit the patient's hearing curve and meet the hearing needs of different frequencies. In addition, the wide dynamic range compression processing method can compress and amplify the sound, convert the sound within the normal hearing dynamic range to the hearing range of the hearing-impaired, and ensure the auditory comfort and clarity of the sound as much as possible, thereby reducing the situation where the background noise is more easily perceived by the hearing-impaired patients after hearing compensation, and then the hearing-impaired patients can minimize the impact of background noise when using hearing aids to hear the sounds heard by normal people.
[0012] In a specific implementation scheme, converting the voice information to be processed into a voice signal in the frequency domain and decomposing it into a preset number of channel signals includes:
[0013] Converting the speech information to be processed into a frequency domain representation and obtaining a speech frequency domain signal;
[0014] Evenly dividing the speech frequency domain signal into a plurality of speech sub-band signals according to the frequency domain bandwidth;
[0015] According to the bark scale, the speech subband signal is divided into a preset number of channel signals.
[0016] Through the above technical solution, considering that the human ear's perception of the frequency of sound is not linearly related to the actual frequency, but approximately logarithmically related, the mapping of linear frequency to logarithmic frequency is used, and the bark scale is used to divide the frequency band into corresponding channels to obtain a multi-channel signal. Processing the sound signal of each channel signal in the multi-channel signal can help reduce background noise.
[0017] In a specific implementation manner, performing wide dynamic range compression processing on each of the channel signals to obtain a channel compression signal comprises:
[0018] Calculating the channel average energy of each of the channel signals;
[0019] Calculate the channel input sound pressure level of each channel signal according to the average energy of each channel;
[0020] Calculate the channel output sound pressure level of each of the channel signals according to the channel input sound pressure level;
[0021] Calculating a channel gain of each of the channel signals according to the channel input sound pressure level and the channel output sound pressure level;
[0022] The channel gain is applied to the channel signal to generate the channel compression signal.
[0023] Through the above technical solution, a channel compression signal can be obtained by compressing the signal in each channel by using wide dynamic range compression processing. Wide dynamic range compression hearing compensation is a core algorithm of speech processing technology in hearing aids. The purpose of wide dynamic range compression processing is to compress and amplify the sound, convert the sound within the normal hearing dynamic range to the hearing range of the hearing impaired, and ensure the auditory comfort and clarity of the sound as much as possible, so that the hearing-impaired patients can hear the sound that normal people can hear.
[0024] In a specific implementation manner, calculating the channel output sound pressure level of each channel signal according to the channel input sound pressure level comprises:
[0025] Obtaining a wide dynamic range compression curve for describing the corresponding relationship between input sound pressure level and output sound pressure level;
[0026] Finding an input region corresponding to the wide dynamic range compression curve according to the channel input sound pressure level;
[0027] The channel output sound pressure level is calculated based on the corresponding relationship between the input sound pressure level and the output sound pressure level of the input area.
[0028] Through the above technical solution, the wide dynamic range compression curve is a curve diagram used to describe the corresponding relationship between the input sound pressure level and the output sound pressure level. The actual hearing conditions of different hearing-impaired patients will be different. In order to take into account the actual hearing conditions of hearing-impaired patients and the influence of environmental noise, a 5-segment curve for suppressing low noise is adopted. Input signals below the lower sound pressure level can be regarded as background noise and suppressed, thereby reducing the impact of noise on the hearing-impaired patients hearing external sounds; among them, the input sound pressure level represents the hearing sound pressure level range of normal people, and the output sound pressure level represents the sound pressure level range of hearing-impaired patients.
[0029] In a specific implementation manner, after calculating the channel gain of each channel signal according to the channel input sound pressure level and the channel output sound pressure level, the method further includes:
[0030] Performing smoothing filtering on the calculated channel gain to obtain a channel smoothing gain;
[0031] The applying the channel gain to the channel signal and generating the channel compression signal comprises:
[0032] The channel smoothing gain is applied to the channel signal to generate the channel compression signal.
[0033] Through the above technical solution, if the result of the gain calculation fluctuates greatly, it will affect the sound quality. Therefore, the calculated channel gain is smoothed and filtered to reduce the fluctuation of the gain calculation result and adjust the start time and release time to reduce the fluctuation of the calculation result, thereby improving the output sound quality of the processed voice signal.
[0034] In a specific implementation scheme, the channel signal includes a plurality of speech subband signals, and the calculating of the channel average energy of each of the channel signals includes:
[0035] Calculating the energy of a plurality of the speech subband signals;
[0036] The channel average energy of the channel signal of each channel is calculated.
[0037] In a specific implementation scheme, the channel input sound pressure level includes:
[0038]
[0039] In a second aspect, the present application provides a speech wide dynamic range compression device, which adopts the following technical solution: the device is applied to a speech wide dynamic range compression system, the speech wide dynamic range compression system includes a hearing aid, and the device includes:
[0040] A voice information acquisition module, used to acquire voice information to be processed, wherein the voice information to be processed includes sound information within a normal hearing dynamic range acquired from the hearing aid;
[0041] A voice information decomposition module, used for converting the voice information to be processed into a voice signal in the frequency domain, and decomposing it into a preset number of channel signals;
[0042] A compression processing module, used for performing wide dynamic range compression processing on each of the channel signals to obtain a channel compression signal;
[0043] The speech information synthesis module is used to synthesize each of the channel compression signals into a full-channel frequency domain signal, and then convert it into a time domain speech signal for output.
[0044] In a third aspect, the present application provides a computer device, which adopts the following technical solution: it includes a memory and a processor, and the memory stores a computer program that can be loaded by the processor and execute any of the above-mentioned speech wide dynamic range compression methods.
[0045] In a fourth aspect, the present application provides a computer-readable storage medium, which adopts the following technical solution: storing a computer program that can be loaded by a processor and execute any of the above-mentioned speech wide dynamic range compression methods.
[0046] In summary, this application has the following beneficial technical effects:
[0047] Since hearing-impaired patients have different degrees of hearing loss at different frequencies, personalized hearing compensation can be achieved by decomposing a voice signal into multi-channel processing methods. Independent compression and amplification on different channels can better fit the patient's hearing curve and meet the hearing needs of different frequencies. In addition, the wide dynamic range compression processing method can compress and amplify the sound, convert the sound within the normal hearing dynamic range to the hearing range of the hearing-impaired, and ensure the auditory comfort and clarity of the sound as much as possible, thereby reducing the situation where background noise is more easily perceived by hearing-impaired patients after hearing compensation, and thus enabling hearing-impaired patients to minimize the impact of background noise when using hearing aids to hear the sounds heard by normal people. BRIEF DESCRIPTION OF THE DRAWINGS
[0048] Figure 1 It is a graph used to illustrate the dynamic range of hearing for normal people and hearing-impaired people;
[0049] Figure 2 is a flow chart of a method for compressing speech with a wide dynamic range in an embodiment of the present application;
[0050] Figure 3 This is a framework diagram of the multi-channel wide dynamic range compression processing method in the embodiment of the present application
[0051] Figure 4 It is a schematic diagram of three-stage and four-stage wide dynamic range compression curves;
[0052] Figure 5 It is a five-stage wide dynamic range compression curve diagram in the embodiment of the present application;
[0053] Figure 6 is a flow chart of a hearing compensation process for wide dynamic range compression processing in an embodiment of the present application;
[0054] Figure 7 is a schematic diagram of a speech wide dynamic range compression device in an embodiment of the present application;
[0055] Figure 8It is a schematic diagram used to embody a computer device in an embodiment of the present application.
[0056] Attached figure numerals: 701, voice information acquisition module; 702, voice information decomposition module; 703, compression processing module; 704, voice information synthesis module. DETAILED DESCRIPTION
[0057] The following is combined with Figure 1-8 This application is described in further detail.
[0058] The embodiment of the present application discloses a speech wide dynamic range compression method, which is applied to a speech wide dynamic range compression system. By adopting this method, users can convert sounds within the auditory dynamic range of normal people into the auditory dynamic range that can be heard by hearing-impaired patients. The hearing impairment of hearing-impaired patients is mainly manifested in an increase in hearing threshold and a narrowing of the auditory dynamic range. The increase in hearing threshold, that is, the minimum perceivable sound intensity is higher than that of normal people, and the degree of increase in hearing threshold at different frequencies is also different, usually the high-frequency part increases more; while the pain threshold of hearing-impaired patients generally increases very little, or can be considered unchanged, which narrows the patient's auditory dynamic range. Figure 1 , Figure 1 In the above table, THR refers to threshold, and UCL refers to uncomfortable level. The subscript n represents the audible threshold range of sound pressure level for normal people, and the subscript u represents the audible threshold range of sound pressure level for hearing-impaired patients. Figure 1 It can be seen that compared with people with normal hearing, the dynamic range of hearing of people with hearing loss is greatly reduced, and the degree of reduction is different in different frequency ranges. In order to enable hearing-impaired patients to hear sounds that normal people can hear, and at the same time meet the requirements of dynamic acceptance of sounds, it is necessary to perform dynamic hearing compensation for hearing-impaired patients according to different frequency ranges and sound intensities, and to compress or amplify the input sound in a targeted manner, so as to convert the sound within the dynamic range of normal people's hearing into the dynamic range of hearing of hearing-impaired patients. At present, a single-channel wide dynamic range compression (WDRC) algorithm is commonly used to achieve loudness compensation. However, the use of a single-channel WDRC method will make some originally smaller background noises easier to perceive after hearing compensation. Therefore, the present application proposes a speech wide dynamic range compression method.
[0059] Reference Figure 2 , the method comprises the following steps:
[0060] S10, obtaining voice information to be processed, where the voice information to be processed includes a section of sound information within a normal hearing dynamic range obtained from a hearing aid.
[0061] Specifically, the system obtains a piece of voice information to be processed. The voice information to be processed is a piece of sound information within the dynamic hearing range of a normal person obtained from a hearing aid. After the user obtains the sound information with the help of a hearing aid, it is transmitted to the processing system to process the sound information and convert it into sound that can be heard by the hearing-impaired patient.
[0062] S20, converting the speech information to be processed into a speech signal in the frequency domain, and decomposing it into a preset number of channel signals.
[0063] Specifically, the obtained voice information to be processed is first converted into a voice signal in the frequency domain, and then decomposed into multiple channel signals for processing. The number of channels can be set according to the actual processing situation of the user. Multi-channel processing technology is the basis for personalized wide dynamic range compression hearing compensation for hearing impairment. Since hearing-impaired patients have different degrees of hearing loss at different frequencies, the use of multi-channel processing methods can achieve accurate personalized hearing compensation.
[0064] S30, performing wide dynamic range compression processing on each channel signal to obtain a channel compression signal.
[0065] Specifically, after decomposing the speech information to be processed, multiple channel signals can be obtained, and then wide dynamic range compression hearing compensation can be performed on the signals on each channel to obtain channel compression signals. Wide dynamic range compression hearing compensation is a core algorithm of speech processing technology in hearing aids. The purpose of the algorithm is to compress and amplify the sound, convert the sound within the dynamic range of normal hearing to the hearing range of the hearing-impaired, and ensure the auditory comfort and clarity of the sound as much as possible, so that the hearing-impaired patients can hear the sound within the dynamic range of normal hearing.
[0066] S40, synthesizing the compressed signal of each channel into a full-channel frequency domain signal, converting it into a time domain speech signal and then outputting it.
[0067] Specifically, after decomposing the speech information into multiple channels for hearing compensation processing, the processed multi-channel signals need to be synthesized into full-channel frequency domain signals and then converted into time domain speech signals before the output can be heard by hearing-impaired patients.
[0068] Reference Figure 3 , Figure 3 This is a framework diagram of a multi-channel wide dynamic range compression processing method. The input is the speech information to be processed obtained with the help of a hearing aid, which is decomposed into channel signals from channel 1 to channel N through a filter bank. Then, wide dynamic range compression hearing compensation is performed on each channel signal. Figure 3 WDRC, or wide dynamic range compression, is processed and output by hearing compensation, and then combined into a voice signal output using a filter.
[0069] In the present application, since hearing-impaired patients have different degrees of hearing loss at different frequencies, personalized hearing compensation can be achieved by decomposing a voice signal into a multi-channel processing method. Independent compression and amplification on different channels can better fit the patient's hearing curve and meet the hearing needs of different frequencies. In addition, the wide dynamic range compression processing method can compress and amplify the sound, convert the sound within the normal hearing dynamic range to the hearing range of the hearing-impaired, and ensure the auditory comfort and clarity of the sound as much as possible, thereby reducing the situation where the background noise is more easily perceived by the hearing-impaired patients after hearing compensation, and thus enabling the hearing-impaired patients to minimize the impact of background noise when using hearing aids to hear the sounds heard by normal people.
[0070] In one embodiment, converting the speech information to be processed into a speech signal in the frequency domain and decomposing it into a preset number of channel signals may be specifically performed as follows:
[0071] First, the acquired speech information to be processed is converted into a frequency domain representation, and the speech frequency domain signal is obtained before subsequent processing. After the speech information to be processed is converted into a frequency domain signal, the speech frequency domain signal is decomposed and processed, and evenly divided into a number of speech sub-band signals according to the frequency domain bandwidth, wherein the signal decomposition processing operation can adopt a filter bank technology, and the filter bank adopted can be an analysis filter bank, that is, a WOLA filter bank, and the design of the analysis filter bank should make the overlap between the decomposed frequency bands as small as possible, so as to provide sufficient degrees of freedom for the subsequent wide dynamic range compression hearing compensation. Afterwards, the obtained several speech sub-band signals are divided into a preset number of channel signals, and the specific way of dividing the speech sub-band signals can be to divide the speech sub-band signals into corresponding channels using the bark scale. Considering that the human ear's perception of the high and low frequency of sound is not linearly related to the high and low of the actual frequency, but is approximately logarithmically related, therefore, if the speech sub-band signals are directly divided in an equal width combination, it is not practical, and the result obtained by the compression processing may be biased. Therefore, the user can use the mapping of linear frequency to logarithmic frequency according to the actual situation, and use the bark scale to divide the sub-band signals into corresponding channels. According to the bark scale, there are 24 critical frequency bands in the normal human hearing frequency range of 22Hz to 22kHz. If the sampling frequency of fs = 16khz is used, there are 20 standard bark critical frequency bands in the frequency range of 0-8khz according to the bark scale table. Since the useful information of speech information is concentrated in 0-8khz, in the embodiment of the present application, the signal of 0-8khz is mainly processed. If the critical frequency band is represented by h(f), the corresponding relationship between the critical frequency band and the linear scale frequency f can be expressed as:
[0072] h(f)=13*arctan(0.76*f)+0.35*arctan((f / 7.5)^2)
[0073] Among them, f represents the linear scale frequency and h(f) represents the critical frequency band.
[0074] Specifically, the speech subband signal can be divided by using the bark scale according to the number of channels. For the band division method using a uniform filter group with k=64 bands, if the sampling rate fs is 16khz, the corresponding band interval, that is, the frequency domain resolution is fs / K=250hz. For the frequency range of 0-8k, except for the first and last bands, whose bandwidths are 125hz, the bandwidths of the remaining bands are the same as the band intervals. The upper boundary frequency of each band is used to determine the bark channel in which each band is located, so that each band is mapped to the bark channel to achieve bark channel division.
[0075] In the present application, considering that the human ear's perception of the high and low frequencies of sound is not linearly related to the high and low frequencies of the actual frequencies, but approximately logarithmically related, the mapping of linear frequency to logarithmic frequency is used, and the bark scale is used to divide the frequency band into corresponding channels to obtain a multi-channel signal. Processing the sound signal of each channel signal in the multi-channel signal can help reduce background noise.
[0076] In one embodiment, performing wide dynamic range compression processing on each channel signal to obtain a channel compression signal may be specifically performed as follows:
[0077] First, the average energy of each channel signal is calculated. The average energy can be calculated by first calculating the energy of each speech subband signal in each channel signal. The energy of each speech subband signal can be calculated by:
[0078] P 子带 =|X(n)| 2
[0079] Wherein, X(n) represents a speech subband signal, and n represents a time frame index.
[0080] Afterwards, the average energy of each channel signal is calculated based on the energy of the speech subband signal. The average energy of each channel signal can be calculated as follows:
[0081]
[0082] Where P(n) represents the average energy of the channel signal, n represents the time frame index, and α and β represent constants.
[0083] If the input voice fluctuates too quickly, the voice energy may fluctuate quickly, and the fast energy fluctuation may cause the calculated gain to fluctuate quickly, resulting in excessive modulation of the voice, and then causing intermodulation distortion of the voice. The average energy calculation method of the channel signal actually uses a first-order filter. The constants α and β are used to independently control and adjust the start time and release time, so that the input energy can be smoothed to avoid modulation distortion caused by fast energy fluctuations.
[0084] Channel energy is smoothed by configurable attack and release times, which are used to control the dynamics of the algorithm; for example, by using slower attack and / or release times, the rate of gain change can be slowed down; this smoothing also ensures that the algorithm does not adjust gain too quickly, which would cause signal distortion.
[0085] After that, the channel input sound pressure level can be calculated by taking the logarithm of the calculated average energy of the channel signal. The channel input sound pressure level can be calculated as follows:
[0086]
[0087] Among them, P(n) is the average energy of the channel; P(0) is a constant, which represents the reference sound pressure level; SPL in The calculated channel input sound pressure level; the reference sound pressure level P(0) = 2*10^(-5)Pa.
[0088] Afterwards, the channel output sound pressure level is calculated based on the calculated channel input sound pressure level. The specific calculation can be obtained by the corresponding relationship between the input sound pressure level and the output sound pressure level. The corresponding relationship between the input sound pressure level and the output sound pressure level can be expressed by a wide dynamic range compression curve or other formulas describing the corresponding relationship between the input and output sound pressure levels, etc., which are not limited here. Afterwards, the channel gain of the channel signal is calculated. The channel gain can be calculated as follows:
[0089] g dB (n) = SPL out -SPL in
[0090] Among them, g dB (n) represents the calculated logarithmic domain channel gain, SPL out Indicates the channel output sound pressure level, SPL in Indicates the channel input sound pressure level.
[0091] Since the gain is calculated using the sound pressure level, the gain result is in the logarithmic domain. The calculated gain needs to be converted to the linear domain. The conversion method can be:
[0092]
[0093] Where gain(n) represents the final calculated channel gain, g dB (n) represents the calculated channel gain in logarithmic domain.
[0094] Finally, the calculated channel gain is applied to the input channel signal to generate a channel compression signal.
[0095] It should be noted that before calculating the average energy of the channel signal, peak envelope detection can be used to find the peak of the signal and estimate the energy of the input signal to facilitate subsequent rapid detection of the signal and rapid calculation of the average energy of the signal.
[0096] In the present application, a channel compression signal can be obtained by compressing the signal in each channel using wide dynamic range compression processing. Wide dynamic range compression hearing compensation is a core algorithm of speech processing technology in hearing aids. The purpose of wide dynamic range compression processing is to compress and amplify the sound, convert the sound within the normal hearing dynamic range to the hearing range of the hearing impaired, and ensure the auditory comfort and clarity of the sound as much as possible, so that the hearing-impaired patients can hear the sound that normal people can hear.
[0097] In one embodiment, calculating the channel output sound pressure level of each channel signal according to the channel input sound pressure level may be specifically performed as follows:
[0098] The wide dynamic range compression curve is used to amplify and compress the dynamic range of each channel separately. First, a wide dynamic range compression curve is obtained. The wide dynamic range compression curve describes the corresponding relationship between the input sound pressure level and the output sound pressure level. The wide dynamic range compression curve is different according to different situations. For example, refer to Figure 4 , Figure 4 (a) is a four-stage curve that takes into account the optimum sound pressure level, that is, the wide dynamic range compression curve that takes into account the most comfortable listening range; Figure 4 (b) is a corresponding curve that is linear in the entire hearing range, that is, a curve that does not consider the optimal sound pressure level, and is a three-segment curve. However, the three-segment or four-segment curve is not ideal for suppressing low noise. In order to take into account the patient's actual hearing condition and the influence of environmental noise, for input signals below a lower sound pressure level, which can also be understood as low noise, a five-segment curve is required to suppress low noise. In the embodiment of the present application, a five-segment curve is used for calculation, for example, referring to Figure 5, the horizontal axis is the input sound pressure level, that is, the normal hearing sound pressure level range, and the vertical axis is the output sound pressure level, that is, the sound pressure level range of the hearing-impaired patient; according to the corresponding relationship between the normal hearing sound pressure level range and the sound pressure level range of the hearing-impaired patient, the corresponding inflection point and threshold on the curve can be determined, and the corresponding compression ratio can also be determined. According to the threshold inflection point and the compression ratio, the normal person's dynamic hearing range DRn (Dynamic Range) can be completely mapped to the patient's dynamic hearing range DRu. Figure 5 In the figure, ith0~ith5 represent the threshold parameters of the input sound pressure level, oth1~oth5 represent the parameters of the output sound pressure level, and k0~k5 represent the slope of the curve; the values of ith0~ith5 and oth1~oth5 are determined by fitting the hearing-impaired patients according to their actual personal conditions after they wear hearing aids.
[0099] In the five-segment wide dynamic range compression curve, the channel output sound pressure level is calculated differently in different curve segments. The method for calculating the channel output sound pressure level based on the channel input sound pressure level is as follows:
[0100] When SOL in ≤ith2, the first curve is usually the expansion zone curve, the expansion ratio means that when the input sound pressure level drops by 1dB below the expansion threshold ith2, the output level will drop by the amount, and the expansion ratio k1 is greater than 1; for example, an expansion ratio of 1:k1 means that a signal below the expansion threshold 1dB will be attenuated by k1dB relative to the signal at the expansion threshold after wide dynamic range compression; since the noise level is usually low, the introduction of the expansion zone curve can avoid unnecessary amplification of the noise level, and the expansion ratio Output sound pressure level SPL out =k1*(SPL in -ith1)+oth1, where oth1=0. Although it is below the normal hearing range, in order to maintain the continuity of the sound and suppress the noise at the same time, the input is expanded and amplified by k1 segments. In the actual processing process, the sound with a sound pressure level less than oth1 cannot be heard and can be simply processed as 0.
[0101] When ith2<SPL in ≤ith3, the second curve is considered to be the low linear region amplification curve, with a slope of k2 and an upper boundary inflection point of (ith3, oth3). The slope is calculated as follows: The output sound pressure level is calculated as SPL out =k2*(SPL in -ith2)+oth2.
[0102] When ith3<SPL in≤ith4, the third curve is usually a compression curve, compression ratio The corresponding slope k3 = 1 / CR, the output sound pressure level SPL out =k3*(SPL in -ith3)+oth3.
[0103] When ith4<SPL in ≤ith5, the fourth curve is considered to be a high linear region amplification curve, and the corresponding slope is Output sound pressure level SPL out =k4*(SPL in -ith4)+oth4.
[0104] When SPL in >ith5, the fifth curve is usually the limiting area, that is, when it is at the hearing discomfort threshold, in order to protect the patient's hearing, the peak clipping algorithm is started, the curve slope k5 = 0, and the output sound pressure level SPL out -oth5.
[0105] After that, the logarithmic domain channel gain is calculated according to the segmented output sound pressure level. First, determine the region in which the input sound pressure level is located, then calculate the output sound pressure level in the region, and then calculate the channel gain in the logarithmic domain. The calculation method can be specifically as follows:
[0106] When SPL in ≤ith2, using the extension area curve:
[0107] g dB (n) = SPL out -SPL in =k1*(SPL in -th1)+oth1-SPL in
[0108] When ith2<SPL in ≤ith3, start low linear region curve:
[0109] g dB (n) = SPL out -SPL in =k2*(SPL in -ith2)+oth2-SPL in
[0110] When ith3<SPL in ≤ith4, start compression curve:
[0111] g dB (n) = k3*(SPL in -ith3)+oth3-SPL in
[0112] When ith4<SPL in ≤ith5, the high linear region amplification curve is used to map the sound pressure level hearing value of a normal person to the hearing value of the patient:
[0113] g dB (n) = SPL out -SPL in =k4*(SPL in -ith4)+oth4-SPL in
[0114] When SPL in >ith5, using a limit curve with a slope of 0 to protect the hearing of hearing-impaired patients:
[0115] Q dB (n) = SPL out -SPL in =oth5-SPL in
[0116] In the present application, the wide dynamic range compression curve is a curve diagram used to describe the corresponding relationship between the input sound pressure level and the output sound pressure level. The actual hearing conditions of different hearing-impaired patients will be different. In order to take into account the actual hearing conditions of hearing-impaired patients and the influence of environmental noise, a 5-segment curve for suppressing low noise is adopted. Input signals below a lower sound pressure level can be considered as background noise and suppressed, thereby reducing the impact of noise on the hearing-impaired patients hearing external sounds; wherein, the input sound pressure level represents the normal hearing sound pressure level range, and the output sound pressure level represents the sound pressure level range of hearing-impaired patients.
[0117] In one embodiment, after the channel gain of each channel signal is calculated according to the channel input sound pressure level and the channel output sound pressure level, the following steps may be performed:
[0118] The calculated channel gain is smoothed and filtered to obtain the channel smoothing gain. The channel smoothing gain can be calculated as follows:
[0119]
[0120] in, is the calculated channel smoothing gain, where gb1 and gb2 are constants.
[0121] Afterwards, the channel smoothing gain is applied to the channel signal to generate a channel compression signal, which can be expressed as:
[0122]
[0123] Among them, x(n) represents the input channel signal, is the channel smoothing gain, and y(n) represents the calculated channel compression signal.
[0124] It should be noted that after the wide dynamic range compression process is completed, that is, after the channel signal is amplified by the channel smoothing gain, if the sound is still not loud enough, the signal can be subjected to the channel gain function again, and the signal can be amplified again and then output so that the patient can hear the sound. The calculation method can be:
[0125] X out (n) = y(n) * gain(n)
[0126] Among them, X out (n) represents the signal after further amplification, y(n) represents the channel compression signal, and gain(n) represents the channel gain.
[0127] In the present application scheme, if the result of the gain calculation fluctuates greatly, it will affect the sound quality. Therefore, the calculated channel gain is smoothed and filtered to reduce the fluctuation of the gain calculation result and adjust the start time and release time to reduce the fluctuation of the calculation result, thereby improving the output sound quality of the processed voice signal.
[0128] Reference Figure 6 , which is a complete flow chart for realizing hearing compensation by multi-channel wide dynamic range compression. First, the peak value of the signal can be found through peak envelope detection, and then the average energy and input and output sound pressure levels of the channel signal can be calculated. Then, the channel gain of each channel signal is calculated, and the channel gain is smoothed and filtered. Then, the channel smoothing gain is applied to the channel signal, and the channel compression signal is output. The channel compression signal is the signal obtained after wide dynamic range compression processing.
[0129] Figure 2 FIG. 1 is a flow chart of a method for compressing speech with a wide dynamic range in one embodiment. It should be understood that although Figure 2 The steps in the flowchart are shown in sequence as indicated by the arrows, but the steps are not necessarily executed in the order indicated by the arrows; unless otherwise specified in this document, there is no strict order restriction for the execution of the steps, and the steps may be executed in other orders; and Figure 2 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.
[0130] Based on the above method, an embodiment of the present application also discloses a speech wide dynamic range compression device.
[0131] Reference Figure 7 , the device includes the following modules:
[0132] The voice information acquisition module 701 is used to acquire the voice information to be processed, where the voice information to be processed includes a sound information within a normal hearing dynamic range acquired from the hearing aid;
[0133] The voice information decomposition module 702 is used to convert the voice information to be processed into a voice signal in the frequency domain and decompose it into a preset number of channel signals;
[0134] The compression processing module 703 is used to perform wide dynamic range compression processing on each channel signal to obtain a channel compression signal;
[0135] The speech information synthesis module 704 is used to synthesize the compressed signal of each channel into a full-channel frequency domain signal, and convert it into a time domain speech signal for output.
[0136] In one embodiment, the speech information decomposition module 702 is specifically used to convert the speech information to be processed into a frequency domain representation and obtain a speech frequency domain signal; divide the speech frequency domain signal evenly into a number of speech sub-band signals according to the frequency domain bandwidth; and divide the speech sub-band signal into a preset number of channel signals according to the bark scale.
[0137] In one embodiment, the compression processing module 703 is specifically used to calculate the channel average energy of each channel signal; calculate the channel input sound pressure level of each channel signal according to each channel average energy; calculate the channel output sound pressure level of each channel signal according to the channel input sound pressure level; calculate the channel gain of each channel signal according to the channel input sound pressure level and the channel output sound pressure level; apply the channel gain to the channel signal, and generate a channel compression signal.
[0138] In one embodiment, the compression processing module 703 is further used to obtain a wide dynamic range compression curve for describing the corresponding relationship between the input sound pressure level and the output sound pressure level; search for an input area corresponding to the wide dynamic range compression curve according to the channel input sound pressure level; and calculate the channel output sound pressure level according to the corresponding relationship between the input sound pressure level and the output sound pressure level of the input area.
[0139] In one embodiment, the compression processing module 703 is also used to perform smoothing filtering on the calculated channel gain and obtain a channel smoothing gain; applying the channel gain to the channel signal and generating a channel compression signal includes: applying the channel smoothing gain to the channel signal and generating a channel compression signal.
[0140] In one embodiment, the compression processing module 703 is further used to calculate the energy of a plurality of speech sub-band signals; and calculate the channel average energy of the channel signal of each channel.
[0141] In one embodiment, the channel input sound pressure level in the compression processing module 703 includes:
[0142]
[0143] The speech wide dynamic range compression device provided in the embodiment of the present application can be applied to the speech wide dynamic range compression method provided in the above embodiment. For relevant details, refer to the above method embodiment. The implementation principle and technical effect are similar and will not be repeated here.
[0144] It should be noted that: when the speech wide dynamic range compression device provided in the embodiment of the present application performs speech wide dynamic range compression, only the division of the above-mentioned functional modules / functional units is used as an example. In actual applications, the above-mentioned functions can be assigned to different functional modules / functional units as needed, that is, the internal structure of the speech wide dynamic range compression device is divided into different functional modules / functional units to complete all or part of the functions described above. In addition, the implementation method of the speech wide dynamic range compression method provided in the above method embodiment and the implementation method of the speech wide dynamic range compression device provided in this embodiment belong to the same concept. The specific implementation process of the speech wide dynamic range compression device provided in this embodiment is detailed in the above method embodiment, and will not be repeated here.
[0145] The embodiment of the present application also discloses a computer device.
[0146] Specifically, if Figure 8As shown, the computer device can be a computer device such as a desktop computer, a laptop computer, a PDA, and a cloud server. The computer device may include, but is not limited to, a processor and a memory. The processor and the memory may be connected via a bus or otherwise. The processor may be a central processing unit (CPU). The processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, graphics processors (GPU), embedded neural network processors (NPU) or other dedicated deep learning coprocessors, discrete gates or transistor logic devices, discrete hardware components and other chips, or a combination of the above-mentioned various chips.
[0147] As a non-transient computer-readable storage medium, the memory can be used to store non-transient software programs, non-transient computer executable programs and modules, such as program instructions / modules corresponding to the methods in the above-mentioned embodiments of the present application. The processor executes various functional applications and data processing of the processor by running the non-transient software programs, instructions and modules stored in the memory, that is, the method in the above-mentioned method implementation is realized. The memory may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created by the processor, etc. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network and a combination thereof.
[0148] The embodiment of the present application also discloses a computer-readable storage medium.
[0149] Specifically, a computer-readable storage medium is used to store a computer program, and when the computer program is executed by a processor, the method in the above-mentioned method implementation is implemented. Those skilled in the art will understand that the implementation of all or part of the process in the above-mentioned implementation method of the present application can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and the program, when executed, may include the process of the implementation of the above-mentioned methods. Among them, the storage medium may be a disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory (Flash Memory), a hard disk (Hard Disk Drive, abbreviated as: HDD) or a solid-state drive (SSD), etc.; the storage medium may also include a combination of the above-mentioned types of memory.
[0150] This specific embodiment is merely an explanation of the present invention and is not a limitation of the present invention. After reading this specification, those skilled in the art may make non-creative modifications to the present embodiment as needed. However, as long as they are within the scope of the claims of the present invention, they are protected by the patent law.
Claims
1. A speech wide dynamic range compression method, characterized in that: The method is applied to a speech wide dynamic range compression system, the speech wide dynamic range compression system includes a hearing aid, and the method includes: Acquire voice information to be processed, wherein the voice information to be processed includes sound information within a normal hearing dynamic range acquired from the hearing aid; Converting the speech information to be processed into a speech signal in the frequency domain and decomposing the signal into a preset number of channel signals; Performing wide dynamic range compression processing on each of the channel signals to obtain a channel compression signal; The compressed signals of each channel are synthesized into a full-channel frequency domain signal, and converted into a time domain speech signal and then output; The step of performing wide dynamic range compression processing on each of the channel signals to obtain a channel compression signal comprises: Calculating the channel average energy of each of the channel signals; Calculate the channel input sound pressure level of each channel signal according to the average energy of each channel; Calculate the channel output sound pressure level of each of the channel signals according to the channel input sound pressure level; Calculating a channel gain of each of the channel signals according to the channel input sound pressure level and the channel output sound pressure level; Applying the channel gain to the channel signal to generate the channel compression signal; Calculating the channel output sound pressure level of each channel signal according to the channel input sound pressure level comprises: Obtaining a wide dynamic range compression curve for describing the corresponding relationship between input sound pressure level and output sound pressure level; Finding an input region corresponding to the wide dynamic range compression curve according to the channel input sound pressure level; Calculating the output sound pressure level of the channel according to the corresponding relationship between the input sound pressure level and the output sound pressure level of the input area; A five-segment curve is used to suppress low noise.
2. The method according to claim 1, characterized in that: The converting the to-be-processed voice information into a voice signal in the frequency domain and decomposing it into a preset number of channel signals comprises: Converting the speech information to be processed into a frequency domain representation and obtaining a speech frequency domain signal; Evenly dividing the speech frequency domain signal into a plurality of speech sub-band signals according to the frequency domain bandwidth; According to the bark scale, the speech subband signal is divided into a preset number of channel signals.
3. The method according to claim 1, characterized in that: After calculating the channel gain of each channel signal according to the channel input sound pressure level and the channel output sound pressure level, the method further includes: Performing smoothing filtering on the calculated channel gain to obtain a channel smoothing gain; The applying the channel gain to the channel signal and generating the channel compression signal comprises: The channel smoothing gain is applied to the channel signal to generate the channel compression signal.
4. The method according to claim 1, characterized in that: The channel signal includes a plurality of speech subband signals, and the calculating of the channel average energy of each of the channel signals includes: Calculating the energy of a plurality of the speech subband signals; The channel average energy of the channel signal of each channel is calculated.
5. The method according to claim 1, characterized in that: The channel input sound pressure level includes: Wherein, P(n) is the average energy of the channel; P(0) is a constant, which represents the reference sound pressure level; SPL in The input sound pressure level for the channel is calculated.
6. A speech wide dynamic range compression device, applied to the speech wide dynamic range compression method according to any one of claims 1 to 5, characterized in that: The device is applied to a speech wide dynamic range compression system, the speech wide dynamic range compression system includes a hearing aid, and the device includes: A voice information acquisition module (701) is used to acquire voice information to be processed, wherein the voice information to be processed includes sound information within a normal hearing dynamic range acquired from the hearing aid; A voice information decomposition module (702), used for converting the voice information to be processed into a voice signal in the frequency domain, and decomposing it into a preset number of channel signals; A compression processing module (703), used for performing wide dynamic range compression processing on each of the channel signals to obtain a channel compression signal; The speech information synthesis module (704) is used to synthesize each of the channel compression signals into a full-channel frequency domain signal, and convert it into a time domain speech signal for output.
7. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executes any one of the methods of claims 1 to 5.
8. A computer-readable storage medium, characterized in that: A computer program is stored which can be loaded by a processor and execute the method according to any one of claims 1 to 5.
