Voice processing method and device

The particle filtering method extracts speech signals in colored noise scenarios, solving the problem of indistinguishability between noise and speech, realizing the accuracy and robustness of speech endpoint detection, and reducing delay.

CN118942480AActive Publication Date: 2024-11-12SMARTER SILICON (SHANGHAI) TECH CO LTD
View PDF 7 Cites 0 Cited by

Patent Information

Application Number
CN202411216270.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-30
Publication Date
2024-11-12
Estimated Expiration
2044-08-30

AI Technical Summary

Technical Problem

In colored noise scenarios, noise and speech are difficult to distinguish, especially under low signal-to-noise ratio conditions, traditional noise estimation methods are prone to delay, affecting the accuracy and robustness of speech endpoint detection.

Method used

The original speech signal is preprocessed by particle filtering method, and an estimated particle set is generated by characterizing the first function of pure speech signal and the second function of characterizing the original speech signal, and the particle weight is updated according to the probability density function to obtain the pure speech signal. Then, voice endpoint detection is performed.

Benefits of technology

It effectively overcomes the problem of insignificant colored noise characteristics, improves the accuracy and predictability of speech estimation, reduces the delay in speech endpoint detection, and improves the robustness and accuracy of detection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118942480A_ABST
    Figure CN118942480A_ABST
Patent Text Reader

Abstract

The invention discloses a voice processing method and device, and the method comprises the steps: carrying out the preprocessing of an original voice signal, and obtaining the preprocessing data; filtering the preprocessed data based on a first function representing a pure voice signal and a second function representing an original voice signal to obtain the pure voice signal, the filtering processing comprising obtaining an estimation particle set based on the first function and determining the weight of each particle in the estimation particle set based on the second function, the pure voice signal is related to the weight of each particle.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical field of natural language processing, and more specifically, to a speech processing method and device. Background Art

[0002] The purpose of speech endpoint detection is to correctly distinguish speech segments from non-speech segments in the detection signal. At present, speech endpoint detection is widely used in speech enhancement, speech recognition, speech coding and other fields. For speech endpoint detection systems, accuracy and robustness are very important indicators. The current mainstream solution is to use noise estimation to extract speech features and determine whether it is a speech frame through traditional threshold judgment or model matching. However, in colored noise scenarios, noise estimation is difficult, especially at low signal-to-noise ratios, when noise and speech are difficult to distinguish, and noise estimation is prone to delays. Summary of the invention

[0003] In view of this, this application provides the following technical solutions:

[0004] The first aspect of the present application provides a speech processing method, the method comprising:

[0005] Preprocessing the original speech signal to obtain preprocessed data;

[0006] The preprocessed data is filtered based on a first function representing a pure speech signal and a second function representing an original speech signal to obtain a pure speech signal, wherein the filtering includes obtaining an estimated particle set based on the first function and determining a weight of each particle in the estimated particle set based on the second function, and the pure speech signal is related to the weight of each particle.

[0007] In an optional implementation, the method further includes:

[0008] Constructing a speech model including a first function characterizing a pure speech signal and a second function characterizing an original speech signal;

[0009] The pre-processed data is filtered using the speech model to obtain a pure speech signal.

[0010] In an optional implementation, filtering the preprocessed data based on the first function characterizing the pure speech signal and the second function characterizing the original speech signal to obtain the pure speech signal includes:

[0011] Generate an initial particle set according to the initial prior probability, wherein the weight of each particle in the initial particle set is the same;

[0012] Obtain an estimated particle set based on the initial particle set and a transfer function corresponding to the first function;

[0013] Update the weight of each particle in the estimated particle set based on the estimated particle set and the probability density function corresponding to the second function;

[0014] A pure speech signal is determined based on the estimated particle set and the updated weight of each particle therein.

[0015] In an optional implementation, updating the weight of each particle in the estimated particle set based on the estimated particle set and the probability density function corresponding to the second function includes:

[0016] Determining the mean and variance of all particles in the estimated particle set;

[0017] The weight of each particle in the estimated particle set is updated based on the mean, the variance and the probability density function.

[0018] In an optional implementation, after updating the weight of each particle in the estimated particle set based on the estimated particle set, the method further includes:

[0019] determining a weight degradation indicator based on the updated weight value;

[0020] When the weight degradation index exceeds a set threshold, resampling of the particle set is performed.

[0021] In an optional implementation, after filtering the preprocessed data based on the first function characterizing the pure speech signal and the second function characterizing the original speech signal to obtain the pure speech signal, the method further includes:

[0022] Determine the probability of speech existence based on the pure speech signal and the original speech signal;

[0023] Speech endpoint detection is performed based on the speech existence probability to determine the speech period in the original speech signal.

[0024] In an optional implementation, determining the speech existence probability based on the pure speech signal and the original speech signal includes:

[0025] Determine a priori signal-to-noise ratio and a posteriori signal-to-noise ratio based on the pure speech signal and the original speech signal respectively;

[0026] Determining a likelihood ratio of the original speech signal based on the a priori signal-to-noise ratio and the a posteriori signal-to-noise ratio;

[0027] An activation function is used to perform nonlinear mapping processing on the likelihood ratio to obtain the probability of speech existence.

[0028] In an optional implementation, the activation function includes a threshold parameter, and the threshold parameter is updated in a histogram statistics adaptive updating manner.

[0029] In an optional implementation, the pre-processed data includes a plurality of speech frames, and the performing speech endpoint detection based on the speech existence probability includes:

[0030] The speech state of the current speech frame is determined based on the speech state of the previous speech frame, the speech existence probability of the current speech frame and the set threshold value, and the speech state includes waiting for speech to start, speech in progress and speech end.

[0031] A second aspect of the present application provides a speech processing device, the device comprising:

[0032] A preprocessing module, used for preprocessing the original speech signal to obtain preprocessed data;

[0033] A filtering processing module is used to filter the preprocessed data based on a first function representing the pure speech signal and a second function representing the original speech signal to obtain a pure speech signal, wherein the filtering processing includes obtaining an estimated particle set based on the first function and determining the weight of each particle in the estimated particle set based on the second function, and the pure speech signal is related to the weight of each particle. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0035] Figure 1 A flowchart of a speech processing method disclosed in an embodiment of the present application;

[0036] Figure 2 A flowchart of estimating a pure speech signal disclosed in an embodiment of the present application;

[0037] Figure 3 A schematic diagram of a process for estimating a pure speech signal disclosed in an embodiment of the present application;

[0038] Figure 4 A flowchart of determining a speech period based on a pure speech signal disclosed in an embodiment of the present application;

[0039] Figure 5 A schematic diagram of a process for adaptively updating a threshold using a histogram disclosed in an embodiment of the present application;

[0040] Figure 6 A schematic diagram of the VAD decision process disclosed in the embodiment of the present application;

[0041] Figure 7 A schematic diagram of the overall implementation flow of a signal processing solution disclosed in an embodiment of the present application;

[0042] Figure 8 A schematic diagram of the structure of a speech processing device disclosed in an embodiment of the present application;

[0043] Fig. 9 A schematic diagram of the structure of an electronic device disclosed in an embodiment of the present application. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0045] The embodiments of the present application can be applied to electronic devices. The present application does not limit the product form of the electronic device, which can include but is not limited to smart phones, tablet computers, wearable devices, personal computers (PCs), netbooks, etc., and can be selected according to application requirements.

[0046] Figure 1 This is a flow chart of a speech processing method disclosed in an embodiment of the present application. Figure 1 As shown, the speech processing method may include:

[0047] Step 101: pre-process the original speech signal to obtain pre-processed data.

[0048] The original voice signal may be a voice signal collected by a voice collection device without any processing, and the original voice information generally includes a pure voice signal and a noise signal. The technical solution disclosed in the embodiment of the present application is to obtain a pure voice signal from the original voice signal, and the specific implementation is described in detail in the embodiment below.

[0049] The preprocessing includes at least framing the original speech signal. In other implementations, the preprocessing includes windowing or other processing in addition to the framing processing, and the windowing is to avoid the spectrum leakage problem when the speech signal is processed from the time domain to the frequency domain. The preprocessing enables the subsequent related processing and analysis to be carried out based on the speech frame, providing available basic data for the subsequent filtering process.

[0050] Step 102: Filter the preprocessed data based on a first function representing a pure speech signal and a second function representing an original speech signal to obtain a pure speech signal, wherein the filtering includes obtaining an estimated particle set based on the first function and determining the weight of each particle in the estimated particle set based on the second function, and the pure speech signal is related to the weight of each particle.

[0051] In the implementation, a first function representing a pure speech signal and a second function representing an original speech signal may be constructed first, and then the first function and the second function may be derived based on the correlation analysis of the first function and the second function so that the first function and the second function are computable; finally, the preprocessed data may be filtered based on the analysis and derivation results of the first function and the second function to obtain a pure speech signal. The pure speech signal obtained here may be an estimated value of the pure speech signal.

[0052] The filtering process may be, but is not limited to, a particle filtering process. Those skilled in the art will know that in a colored noise scenario, the noise features are variable and not obvious, and have non-Gaussian features, so it is difficult to estimate the noise; and particle filtering is a nonlinear non-Gaussian Bayesian filtering method based on Monte Carlo, which can well solve the problem of nonlinear non-Gaussian systems. Therefore, the particle filtering method can well realize the estimation of pure speech signals in colored noise scenarios, and the accuracy of speech estimation results is also relatively high.

[0053] During the filtering process, an estimated particle set can be obtained based on a first function representing the pure speech signal, and then the weight of each particle in the estimated particle set can be determined based on a second function representing the original speech signal. The weight can represent the possibility that the corresponding particle is a pure speech signal. Therefore, the pure speech signal is related to the weight of each particle.

[0054] The speech processing method described in the embodiment of the present application performs filtering processing on the preprocessed data based on the first function representing the pure speech signal and the second function representing the original speech signal, thereby estimating the pure speech signal of the preprocessed data and obtaining an estimated pure speech signal. Since the speech estimation is performed by filtering, the problem that the colored noise feature is not obvious and difficult to estimate can be overcome. At the same time, the speech estimation has a certain predictability, so the delay existing in the traditional noise estimation can be significantly improved.

[0055] Based on the contents of the above embodiments, in other implementations, the speech processing method may also include: constructing a speech model including a first function characterizing a pure speech signal and a second function characterizing an original speech signal, and then using the speech model to filter the preprocessed data to obtain a pure speech signal, that is, the process of filtering the preprocessed data based on the first function and the second function to obtain a pure speech signal in the aforementioned embodiment.

[0056] In one implementation, the constructed speech model may be a TVAR (Time-Varying Autoregressive) model. The TVAR model includes a first function and a second function, wherein the first function represents a pure speech signal sk, and the second function represents an original speech signal (noise signal) y k ,s k and k The expression is as follows:

[0057]

[0058] y k =s k +v k

[0059] Among them, α k =(α 1,k , α 2,k , ..., α p,k ) T is the coefficient vector of the kth frame p dimension; u k is the process noise, v k is the observation noise, all of which are Gaussian white noise, and their variances are and The obtained k Conditional probability density:

[0060]

[0061] Observation value y k The likelihood function is expressed as a mean of s k , variance v k The normal distribution of:

[0062]

[0063] In order to better describe the time-varying characteristics of the noise sequence, the noise logarithmic deviation is taken, And its changing law conforms to the Gaussian model:

[0064]

[0065] For α in the TVAR model k The coefficient can also be expressed by a Gaussian model to describe its time-varying characteristics:

[0066]

[0067] I p is the identity matrix of order p. 2 ) means the mean is u and the variance is σ2 Gaussian distribution.

[0068] Therefore, the speech model can be expressed as:

[0069]

[0070] The unknown parameters in the model are Select the initial value according to the Gaussian distribution, and the parameter that changes quickly over time Can be assumed to be a constant, if possible

[0071] Based on the above, the transfer function (also called transfer probability density) of the state variable (pure speech signal) is:

[0072]

[0073] The probability density function of the system observation variable (original speech signal) is:

[0074]

[0075] It is understandable that the TVAR model initially constructed only two functions that characterize data relationships. There is no way to directly calculate these two functions. The above transfer function and probability density function express the way the model changes, making the TVAR model computable and usable.

[0076] The above content can be understood as the preparation content for filtering processing. Subsequently, the pre-processed data can be filtered based on the transfer function and the probability density function to obtain a pure speech signal.

[0077] Specifically, Figure 2 The process of estimating pure speech signal is shown. Figure 2 The filtering process of the preprocessed data based on the first function representing the pure speech signal and the second function representing the original speech signal to obtain the pure speech signal may include:

[0078] Step 201: Generate an initial particle set according to the initial prior probability, wherein the weight of each particle in the initial particle set is the same.

[0079] Step 201 is the initialization of the particle set, including setting the number of particles N, and the weight of each particle is 1 / N. The initial prior probability can be assumed to be a Gaussian distribution.

[0080] Step 202: Obtain an estimated particle set based on the initial particle set and a transfer function corresponding to the first function.

[0081] After obtaining the initial particle set, it is necessary to update the state estimate at time k based on the transfer function of the first function, that is, the transfer function of the state variable, to obtain the estimated particle set Calculate the mean and variance

[0082] Step 203: updating the weight of each particle in the estimated particle set based on the estimated particle set and the probability density function corresponding to the second function.

[0083] The calculation formula for estimating the weight of each particle in the particle set is:

[0084]

[0085] in is an important density function. To simplify the calculation, we can take Then, the particle weight update formula can be simplified as:

[0086]

[0087] Furthermore, the weights can be normalized:

[0088] Based on the above, the implementation of updating the weights of each particle in the estimated particle set based on the estimated particle set and the probability density function corresponding to the second function may include: determining the mean and variance of all particles in the estimated particle set; and updating the weights of each particle in the estimated particle set based on the mean, the variance and the probability density function.

[0089] Step 204: Determine a pure speech signal based on the estimated particle set and the updated weight of each particle therein.

[0090] Specifically, the estimated value of the target state of the k-th frame speech signal is calculated:

[0091]

[0092] That is, the determined pure speech signal may be an estimated value of each frame of speech signal.

[0093] In an embodiment of the present application, a particle filter is used for speech estimation. The particle filter has a certain robustness for Gaussian models and non-Gaussian models, as well as linear systems and nonlinear systems. The accuracy can be close to the optimal estimation, and it can better estimate the pure speech signal, making the subsequent feature calculation more accurate. At the same time, its prediction of the speech signal can reduce the processing delay in the speech endpoint detection application.

[0094] Based on the disclosed content of the above embodiment, after updating the weights of each particle in the estimated particle set based on the estimated particle set, it can also include: determining a weight degradation index based on the updated weight; and resampling the particle set when the weight degradation index exceeds a set threshold.

[0095] Figure 3 The following is a schematic diagram of a process for estimating a pure speech signal disclosed in an embodiment of the present application. Figure 3 As shown in the figure, in the process of estimating the pure speech signal, based on the existing particle set, the state estimation at time k is first updated to obtain the estimated particle set corresponding to time k, and then the weight is updated and normalized, that is, the weight of each particle in the estimated particle set is updated based on the probability density function of the second function, and the updated weight is normalized, and finally the pure speech signal is estimated based on the normalized weight (weight value) of each particle in the estimated particle set; after each weight update in the process, the weight degradation index is calculated based on the updated weight value, and it is determined whether the weight degradation index exceeds the set threshold. If so, it means that the weight has degraded to a certain extent and the accuracy is not high, and the particle set needs to be resampled to ensure the accuracy of the estimated pure speech signal.

[0096] In other implementations, after obtaining the pure speech signal, further processing may be performed based on the estimated pure speech signal to determine the speech period in the original speech signal. In an embodiment, the speech period includes at least human voice, and the non-speech period has only noise or no sound. Figure 4 This is a flow chart of determining a speech period based on a pure speech signal disclosed in an embodiment of the present application. Figure 4 As shown, the process of determining the speech period may include:

[0097] Step 401: Determine the probability of speech existence based on the pure speech signal and the original speech signal.

[0098] In implementation, the prior signal-to-noise ratio and the posterior signal-to-noise ratio can be determined based on the pure speech signal and the original speech signal respectively, and the likelihood ratio of the original speech signal can be determined based on the prior signal-to-noise ratio and the posterior signal-to-noise ratio. Then, an activation function is used to perform nonlinear mapping processing on the likelihood ratio to obtain the probability of speech existence.

[0099] Specifically, the priori signal-to-noise ratio and the posterior signal-to-noise ratio are first calculated to obtain the likelihood ratio feature. In implementation, the speech signal can be converted from the time domain to the frequency domain, and then the posterior signal-to-noise ratio σ(k) can be calculated using the converted pure speech signal S(k) and the original speech signal Y(k):

[0100]

[0101] Calculate the prior signal-to-noise ratio ρ(k):

[0102] ρ(k)=α·H(k-1) 2 ·σ(k-1)+(1-α)·max{(σ(k)-1),0}

[0103] Where k represents the kth frame, α is the smoothing coefficient, which can be 0.99. H(k-1) is the Wiener filter coefficient of the previous frame:

[0104]

[0105] Calculate the likelihood ratio Δ(k):

[0106]

[0107] After obtaining the likelihood ratio, the likelihood ratio can be activated to obtain the probability of speech existence. The likelihood ratio is activated using a nonlinear function, such as a sigmoid function, a tanh(x) function, etc. For example, the hyperbolic tangent function tanh(x), the value range of tanh(x) is (-1,1), and through calculation, the value range of the final indicator function is (0,1), that is, when the signal is speech, the value of ind approaches 1, and when the signal is non-speech, the value of ind approaches 0.

[0108] Calculate the indicator function formula:

[0109] ind=0.5·(tanh(w·(Δ(k)-T))+1)

[0110] Where w is a transition parameter that controls the smoothness of the mapping from 0 to 1, and T is the threshold parameter.

[0111] Update the speech existence probability P(k). The process of updating the speech existence probability can be understood as a smoothing process. The calculation formula of P(k) is:

[0112] P(k)=(1-α p )·P(k-1)+α p ·ind

[0113] where α p is the smoothing coefficient and can be taken as 0.1.

[0114] In this implementation, the obtained eigenvalues ​​are controlled within a certain range through the nonlinear mapping of the activation function, and nonlinear factors are introduced to make the results applicable to more nonlinear problems.

[0115] In other implementations, the activation function includes a threshold parameter, and the threshold parameter can be updated in a manner of histogram statistics adaptive update. Figure 5This is a schematic diagram of the process of using histogram adaptive updating threshold value disclosed in the embodiment of the present application, which can be found in Figure 5 Understand the specific implementation of updating the threshold parameters. Using histogram statistics, the threshold parameters in the activation function are updated adaptively, so that the output speech existence probability of the activation function is more accurate, and the result can better match more scenes. Since the use of histogram adaptive updating of thresholds is a technical content that already exists in the field, this application will not introduce it in detail.

[0116] Step 402: Perform speech endpoint detection based on the speech existence probability to determine the speech period in the original speech signal.

[0117] After determining the probability of speech existence, speech endpoint detection can be performed based on the probability of speech existence, that is, VAD (Voice activity detection) judgment. Specifically, the speech state of the current speech frame can be determined based on the speech state of the previous speech frame, the speech probability of the current speech frame and the set threshold value, and the speech state includes waiting for speech to start, speech in progress and speech end.

[0118] When performing VAD judgment, a single threshold method, a double threshold method, or an adaptive threshold method may be used, and this application does not impose any fixed restrictions thereon. The following will take the double threshold method as an example to illustrate the specific implementation of VAD judgment.

[0119] Figure 6 This is a schematic diagram of the VAD decision process disclosed in the embodiment of this application. Figure 6 As shown in the figure, after obtaining the speech existence probability P, the corresponding judgment is made in combination with the status of the current frame. Status has three states: when status is 0, it indicates the speech waiting to start state; when status is 1, it indicates the speech in progress state; when status is 2, it indicates the speech ending state. During the implementation process, the status is continuously updated, and the speech endpoint detection is realized in combination with the speech existence probability of each frame of the speech signal.

[0120] When status is 0, first determine whether P is greater than threshold 2. If so, it is considered that speech has started. If not, further determine whether P is greater than threshold 1. If it is greater than threshold 1, it is also considered that speech has started. If it is not greater than threshold 1, it is considered that the current state is still the speech waiting to start state, that is, status is 0. Threshold 2 is greater than threshold 1. Threshold 1 and threshold 2 can be set based on experience and continuously adjusted during the application process.

[0121] When status is 1, first determine whether P is greater than the average of threshold 1 and threshold 2. If so, it is considered that the speech is continuing and status is still 1; otherwise, accumulate the non-speech duration and continue to determine whether the non-speech duration is less than the maximum pause (preset value). If so, status is 1. If both are true, it is considered that the speech is over and status is determined to be 2.

[0122] When status is 2, it is determined that the voice has ended. When status is 0, it is waiting for the next voice to start, and the non-voice duration is reset to zero.

[0123] In summary, the overall implementation of a signal processing solution is as follows Figure 7 As shown, it can be combined Figure 7 Understand the contents of the above embodiments. The speech processing method described in the embodiment of the present application estimates the pure speech signal through particle filtering, can obtain more accurate likelihood ratio parameters, and then use the activation function for nonlinear mapping to obtain more robust speech existence probability characteristics, and finally perform threshold judgment to obtain the speech judgment result. This implementation has the advantages of high result accuracy and low latency.

[0124] For the aforementioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the order of the actions described, because according to the present application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0125] The method is described in detail in the embodiments disclosed in the above-mentioned application. The method of the application can be implemented by various forms of devices. Therefore, the application also discloses a device, and a specific embodiment is given below for detailed description.

[0126] Figure 8 Schematic diagram of the structure of a speech processing device disclosed in an embodiment of the present application. Figure 8 As shown, the speech processing device 80 may include:

[0127] The preprocessing module 801 is used to preprocess the original speech signal to obtain preprocessed data.

[0128] The filtering processing module 802 is used to filter the preprocessed data based on a first function representing the pure speech signal and a second function representing the original speech signal to obtain a pure speech signal. The filtering processing includes obtaining an estimated particle set based on the first function, and determining the weight of each particle in the estimated particle set based on the second function. The pure speech signal is related to the weight of each particle.

[0129] The speech processing device described in the embodiment of the present application performs filtering processing on the preprocessed data based on the first function representing the pure speech signal and the second function representing the original speech signal, thereby estimating the pure speech signal of the preprocessed data and obtaining the estimated pure speech signal. Since the speech estimation is performed by filtering, the problem that the colored noise feature is not obvious and difficult to estimate can be overcome. At the same time, the speech estimation has a certain predictability, so the delay existing in the traditional noise estimation can be significantly improved.

[0130] In one implementation, the device may also include: a model construction module, used to construct a speech model including a first function representing a pure speech signal and a second function representing an original speech signal, and then filter the preprocessed data based on the constructed speech model to obtain a pure speech signal.

[0131] In one implementation, the filtering processing module includes: a particle set initialization module, which is used to generate an initial particle set according to an initial prior probability, and the weights of each particle in the initial particle set are the same; a particle set estimation module, which is used to obtain an estimated particle set based on the initial particle set and the transfer function corresponding to the first function; a weight updating module, which is used to update the weights of each particle in the estimated particle set based on the estimated particle set and the probability density function corresponding to the second function; and a signal determination module, which is used to determine a pure speech signal based on the estimated particle set and the updated weights of each particle therein.

[0132] In one implementation, the weight updating module may be specifically configured to: determine the mean and variance of all particles in the estimated particle set; and update the weight of each particle in the estimated particle set based on the mean, the variance and the probability density function.

[0133] In one implementation, the filtering processing module may further include: a resampling module, configured to determine a weight degradation index based on the updated weight; and to resample the particle set when the weight degradation index exceeds a set threshold.

[0134] In one implementation, the speech processing device may further include: an existence probability determination module, used to determine the speech existence probability based on the pure speech signal and the original speech signal; an endpoint detection module, used to perform speech endpoint detection based on the speech existence probability to determine the speech period in the original speech signal.

[0135] In one implementation, the existence probability determination module can be specifically used to: determine the prior signal-to-noise ratio and the posterior signal-to-noise ratio based on the pure speech signal and the original speech signal, respectively; determine the likelihood ratio of the original speech signal based on the prior signal-to-noise ratio and the posterior signal-to-noise ratio; and use an activation function to perform nonlinear mapping processing on the likelihood ratio to obtain the probability of speech existence.

[0136] In one implementation, the activation function includes a threshold parameter, and the threshold parameter is updated by adaptively updating the histogram statistics.

[0137] In one implementation, the preprocessed data includes multiple speech frames, and the endpoint detection module can be specifically used to determine the speech state of the current speech frame based on the speech state of the previous speech frame, the probability of the speech existence of the current speech frame and a set threshold value, and the speech state includes waiting for the speech to start, speaking and ending.

[0138] The specific implementation of the above-mentioned speech processing device and the modules it contains, as well as other possible implementations, can be found in the corresponding parts of the method embodiment, which will not be repeated here.

[0139] Any of the speech processing devices described in the above embodiments includes a processor and a memory. The preprocessing module, filtering processing module, model building module, particle set initialization module, particle set estimation module, weight update module, signal determination module, resampling module, etc. in the above embodiments are all stored in the memory as program modules, and the processor executes the above program modules stored in the memory to realize corresponding functions.

[0140] The processor includes a kernel, which retrieves the corresponding program module from the memory. One or more kernels can be set, and the processing of the access data can be realized by adjusting the kernel parameters.

[0141] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0142] In an exemplary embodiment, a computer-readable storage medium is also provided, which can be directly loaded into the internal memory of a computer and contains software code. After being loaded and executed by a computer, the computer program can implement the steps shown in any embodiment of the above-mentioned speech processing method.

[0143] In an exemplary embodiment, a computer program product is also provided, which can be directly loaded into the internal memory of a computer and contains software codes. After being loaded and executed by a computer, the computer program can implement the steps shown in any embodiment of the speech processing method described above.

[0144] Furthermore, an embodiment of the present application provides an electronic device. Fig. 9 This is a schematic diagram of the structure of an electronic device disclosed in an embodiment of the present application. Fig. 9As shown, the electronic device 90 includes at least one processor 901, and at least one memory 902 and a bus 903 connected to the processor; wherein the processor and the memory communicate with each other via the bus; the processor is used to call the program instructions in the memory to execute the above-mentioned voice processing method.

[0145] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0146] It should also be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0147] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0148] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A speech processing method, the method include: Preprocessing the original speech signal to obtain preprocessed data; The preprocessed data is filtered based on a first function representing a pure speech signal and a second function representing an original speech signal to obtain a pure speech signal, wherein the filtering includes obtaining an estimated particle set based on the first function and determining a weight of each particle in the estimated particle set based on the second function, and the pure speech signal is related to the weight of each particle.

2. The speech processing method according to claim 1, further comprising: include: Constructing a speech model including a first function characterizing a pure speech signal and a second function characterizing an original speech signal; The pre-processed data is filtered using the speech model to obtain a pure speech signal.

3. The speech processing method according to claim 1, wherein the pre-processed data is filtered based on a first function representing a pure speech signal and a second function representing an original speech signal to obtain a pure speech signal. include: Generate an initial particle set according to the initial prior probability, wherein the weight of each particle in the initial particle set is the same; Obtain an estimated particle set based on the initial particle set and a transfer function corresponding to the first function; Update the weight of each particle in the estimated particle set based on the estimated particle set and the probability density function corresponding to the second function; A pure speech signal is determined based on the estimated particle set and the updated weight of each particle therein.

4. The speech processing method according to claim 3, wherein the weight of each particle in the estimated particle set is updated based on the probability density function corresponding to the estimated particle set and the second function, include: Determining the mean and variance of all particles in the estimated particle set; The weight of each particle in the estimated particle set is updated based on the mean, the variance and the probability density function.

5. The speech processing method according to claim 3, after updating the weight of each particle in the estimated particle set based on the estimated particle set, further include: determining a weight degradation indicator based on the updated weight value; When the weight degradation index exceeds a set threshold, resampling of the particle set is performed.

6. The speech processing method according to claim 1, after filtering the pre-processed data based on the first function representing the pure speech signal and the second function representing the original speech signal to obtain the pure speech signal, further include: Determine the probability of speech existence based on the pure speech signal and the original speech signal; Speech endpoint detection is performed based on the speech existence probability to determine the speech period in the original speech signal.

7. The speech processing method according to claim 6, wherein the speech existence probability is determined based on the pure speech signal and the original speech signal. include: Determine a priori signal-to-noise ratio and a posteriori signal-to-noise ratio based on the pure speech signal and the original speech signal respectively; Determining a likelihood ratio of the original speech signal based on the a priori signal-to-noise ratio and the a posteriori signal-to-noise ratio; An activation function is used to perform nonlinear mapping processing on the likelihood ratio to obtain the probability of speech existence.

8. According to the speech processing method of claim 7, the activation function includes a threshold parameter, and the threshold parameter is updated by adaptively updating the histogram statistics.

9. The speech processing method according to claim 6, wherein the pre-processed data comprises a plurality of speech frames, and the speech endpoint detection is performed based on the speech existence probability. include: The speech state of the current speech frame is determined based on the speech state of the previous speech frame, the speech existence probability of the current speech frame and the set threshold value, and the speech state includes waiting for speech to start, speech in progress and speech end.

10. A speech processing device, the device include: A preprocessing module, used for preprocessing the original speech signal to obtain preprocessed data; A filtering processing module is used to filter the preprocessed data based on a first function representing the pure speech signal and a second function representing the original speech signal to obtain a pure speech signal, wherein the filtering processing includes obtaining an estimated particle set based on the first function and determining the weight of each particle in the estimated particle set based on the second function, and the pure speech signal is related to the weight of each particle.

Citation Information

Patent Citations

  • Method and device for voice noise reduction, computing device and computer readable storage medium

    CN110164467A

  • Method and device for filtering residual noise after machine learning speech enhancement

    CN112289337A

  • Signal processing method and device of microphone array

    CN116612776A

  • Jitter buffer control method and system for streaming media transmission, medium and equipment

    CN116805938A

  • Noise suppressing device, computer program, and speech recognition system

    JP2007041499A