Acoustic signal separation method, system and equipment based on logarithmic processing mechanism and medium

Abstract

The invention provides an acoustic signal separation method, system and device based on a logarithmic processing mechanism and a medium, and the method comprises the steps: receiving a mixed noisy signal, carrying out the preprocessing, and extracting parameters; the parameters comprise amplitude spectrums of a target signal, noise and a noisy signal and short signal power estimation; based on the extracted parameters, constructing a logarithmic probability density distribution model, and obtaining logarithmic probability density distribution of the target signal, the noise and the noisy signal through the model; based on the obtained logarithmic probability density distribution, generating a T-F mask by using an MMSE estimation theory; and processing the noisy signal according to the T-F mask to obtain spectrum amplitude estimation of the target signal. According to the method, the target sound signal can be estimated from the noisy signal, and the logarithmic spectrum distribution of the sound signal is obtained by adopting an unsupervised mode, so that the method is insensitive to noise types, good in generalization and low in calculation complexity, and the noise suppression effect can be realized in numerous low-calculation-power hardware platforms.

Description

Technical Field

[0001] The present invention relates to the technical field of acoustic signal processing, and in particular to an acoustic signal separation method, system, device and medium based on a logarithmic processing mechanism. Background Art

[0002] In the application scenarios of acoustic detection equipment, environmental noise and background interference are inevitable problems, which leads to a large amount of noise signals mixed in the sound source signals received by the acoustic detection equipment, affecting the normal operation of the detection equipment and its detection effect. In order to improve the detection ability of the acoustic detection equipment for the target sound signal, it is necessary to use signal enhancement technology to pre-process the noisy signal. Its core purpose is to effectively suppress the background noise while minimizing the distortion of the target signal.

[0003] In the related art, the more effective signal enhancement method is mainly based on the time-frequency (TF) mask principle, that is, by acting on each frequency band of the noisy signal, the TF component dominated by the target signal is selectively retained, and the component dominated by the noise is shielded, thereby achieving effective extraction of the target signal.

[0004] For TF mask, there is a method to estimate TF mask based on the statistical characteristics of target and noise, assuming the statistical distribution of target and noise as a specific statistical model in the frequency domain, and combining different optimization criteria to obtain the TF mask for recovering the target signal from the noisy signal. In addition to this acoustic signal separation method based on frequency domain model, there is also a method based on logarithmic spectrum domain model, such as: a method based on hidden Markov model (HMM), which uses HMM to train sample data of target signal and noise to complete the modeling of their logarithmic spectrum distribution, and derives TF mask with the help of cross product to estimate target signal information. A method based on Gaussian mixture model, which pre-trains the target signal and noise by using different numbers of mixed Gaussian components, and constructs a logarithmic probability density distribution to recover the target sound signal.

[0005] Since the perception of different frequencies of sound signals is not linear and presents the characteristics of logarithmic distribution, the logarithmic spectrum domain model is more suitable for processing sound signals than the frequency domain model. However, due to the nonlinear characteristics of the logarithmic spectrum domain, it is a difficult problem to accurately derive the estimation of sound signals from a mathematical point of view. Therefore, most signal separation methods still choose to directly process the sound source signal in the frequency domain. Obviously, the methods based on the frequency domain model will have some limitations in signal separation capabilities.

[0006] In order to overcome the problem of difficulty in accurately deriving the target acoustic signal expression due to logarithmic nonlinearity, related technologies train acoustic data through autoregressive models and hidden Markov models to directly obtain the logarithmic distribution of target signals and noise. However, noise signals vary greatly and are of many types, and the training samples are limited, which makes the noise-based training method have the problem of insufficient noise generalization. Summary of the invention

[0007] The present invention provides a sound signal separation method, system, device and medium based on a logarithmic processing mechanism, which solves the problem of insufficient generalization caused by reliance on noise training in a signal separation method based on a logarithmic spectral domain.

[0008] In order to achieve the above objectives, this application adopts the following technical solutions:

[0009] In a first aspect, a method for separating acoustic signals based on a logarithmic processing mechanism is provided, comprising:

[0010] receiving a mixed noisy signal, performing preprocessing and extracting parameters; the parameters include amplitude spectra of the target signal, noise and noisy signal, and a short signal power estimate;

[0011] Based on the extracted parameters, a logarithmic probability density distribution model is constructed, and the logarithmic probability density distribution of the target signal, noise and noisy signal is obtained through the model;

[0012] Based on the logarithmic probability density distribution of the target signal, the noise and the noisy signal, a TF mask is generated using MMSE estimation theory;

[0013] The noisy signal is processed according to the TF mask to obtain a spectrum amplitude estimation of the target signal.

[0014] In a first possible implementation manner of the first aspect, the receiving a mixed noisy signal, performing preprocessing, and extracting parameters includes:

[0015] Performing short-time Fourier transform on the mixed noisy signal to obtain amplitude spectra of the target signal, noise and noisy signal;

[0016] The power spectrum of the noisy signal is obtained according to the amplitude spectrum of the noisy signal, and the power estimation of the transient signal is calculated based on the power spectrum of the noisy signal.

[0017] In a second possible implementation manner of the first aspect, constructing a logarithmic probability density distribution model based on the extracted parameters, and obtaining the logarithmic probability density distribution of the target signal, the noise, and the noisy signal through the model includes:

[0018] Performing Fourier transform and logarithmic processing on the mixed noisy signal to obtain a representation of the logarithmic amplitude spectrum of the target signal, the noise, and the noisy signal;

[0019] Using Laplace distribution and Rayleigh distribution to approximate the probability density distribution of the amplitude spectrum of the target signal and the noise respectively;

[0020] Converting the probability density distribution of the amplitude spectrum of the target signal and the noise into a corresponding logarithmic probability density distribution;

[0021] Based on the mixed maximum model relationship between the target signal, the noise and the noisy signal in the logarithmic domain, the logarithmic probability density distribution of the noisy signal is obtained.

[0022] Based on the second possible implementation manner of the first aspect, in a third possible implementation manner of the first aspect, converting the probability density distribution of the amplitude spectrum of the target signal and the noise into a corresponding logarithmic probability density distribution includes:

[0023] Calculate the logarithmic cumulative distribution function of the target signal and noise amplitude spectrum coefficients respectively;

[0024] The corresponding logarithmic cumulative distribution functions are derived respectively to obtain the logarithmic probability density distribution of the target signal and the noise.

[0025] Based on the third possible implementation manner of the first aspect, in a fourth possible implementation manner of the first aspect, the logarithmic amplitude spectra of the target signal, the noise, and the noisy signal are expressed as:

[0026] S log =[S log,1 ,...,S log,k ,...,S log,K ] T

[0027] N log =[N log,1 ,...,N log,k ,...,N log,K ] T

[0028] Y log =[Y log,1 ,...,Y log,k ,...,Y log,K ] T ,

[0029] Among them, S log Represents the amplitude log spectrum vector of the target signal, N log Represents the magnitude log spectrum vector of the noise signal, Y logrepresents the amplitude log spectrum vector of the noisy signal, K represents the total number of vector elements, and k represents the kth element in the vector;

[0030] The probability density distribution of the amplitude spectrum of the target signal and the noise is approximated by Laplace distribution and Rayleigh distribution respectively, including:

[0031]

[0032]

[0033] Among them, vectors SS and NN are the amplitude spectrum vectors of the target signal and noise, and can be expressed as SS = [SS1...,SS k ...,SS K ] T and NN=[NN1...,NN k ...,NN K ] T ,λ s and λ n Power estimates of short bursts representing target signals and noise;

[0034] The logarithmic probability density distribution of the target signal and noise is:

[0035]

[0036]

[0037] The logarithmic probability density distribution of the noisy signal is:

[0038]

[0039] Based on the fourth possible implementation manner of the first aspect, in a fifth possible implementation manner of the first aspect, the TF mask is:

[0040]

[0041] The spectral amplitude of the target signal is estimated as:

[0042]

[0043] In a second aspect, a sound signal separation system based on a logarithmic processing mechanism is provided, comprising:

[0044] A preprocessing module, used to receive a mixed noisy signal, perform preprocessing and extract parameters; the parameters include the amplitude spectrum of the target signal, noise and noisy signal and the power estimation of the short signal;

[0045] A logarithmic probability density distribution model module, used to construct a logarithmic probability density distribution model based on the extracted parameters, and obtain the logarithmic probability density distribution of the target signal, noise and noisy signal through the model;

[0046] A TF mask module, used for generating a TF mask based on the logarithmic probability density distribution of the target signal, the noise and the noisy signal using the MMSE estimation theory;

[0047] A signal separation module is used to process the noisy signal according to the TF mask to obtain a spectrum amplitude estimation of the target signal.

[0048] In a first possible implementation manner of the second aspect, the logarithmic probability density distribution model module is specifically used to:

[0049] Performing Fourier transform and logarithmic processing on the mixed noisy signal to obtain a representation of the logarithmic amplitude spectrum of the target signal, the noise, and the noisy signal;

[0050] Using Laplace distribution and Rayleigh distribution to approximate the probability density distribution of the amplitude spectrum of the target signal and the noise respectively;

[0051] Converting the probability density distribution of the amplitude spectrum of the target signal and the noise into a corresponding logarithmic probability density distribution;

[0052] Based on the mixed maximum model relationship between the target signal, the noise and the noisy signal in the logarithmic domain, the logarithmic probability density distribution of the noisy signal is obtained.

[0053] In a third aspect, an electronic device is provided, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the computer program is executed by the processor, the steps of the acoustic signal separation method based on the logarithmic processing mechanism as described in the first aspect are implemented.

[0054] In a fourth aspect, a readable storage medium is provided, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the acoustic signal separation method based on the logarithmic processing mechanism as described in the first aspect are implemented.

[0055] This application adopts a method based on an unsupervised strategy to obtain the logarithmic function of noise and target signal. This method converts the distribution of target signal and noise signal in frequency domain to logarithmic domain through a transformation function, and then uses the logarithmic relationship between target signal, noise and noisy signal to separate the target signal using the minimum mean square error estimation theory. Compared with the signal separation method based on data training, this unsupervised method makes the method less susceptible to the influence of noise type, has better generalization ability for noise, and is better adaptable to different environments in specific use.

[0056] At the same time, since the present application does not rely on data training, it does not have large model parameters, thereby reducing the computational complexity during implementation, so that it can run on a hardware platform with lower computing power.

[0057] In summary, the acoustic signal separation method proposed in this application can estimate the target acoustic signal from the noisy signal. Since this application adopts an unsupervised method to obtain the logarithmic spectrum distribution of the sound signal, it is insensitive to the type of noise, has good generalization, and has low computational complexity, and can achieve noise suppression effects in many low-computing hardware platforms. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 A schematic flow chart of a method for separating acoustic signals based on a logarithmic processing mechanism provided in an embodiment of the present application;

[0059] Figure 2 A schematic flow chart of another acoustic signal separation method based on a logarithmic processing mechanism provided in an embodiment of the present application;

[0060] Figure 3 A structural block diagram of an acoustic signal separation system based on a logarithmic processing mechanism provided in an embodiment of the present application;

[0061] Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0062] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined purpose, the technical solutions in the embodiments of the present application are clearly described. Obviously, the described embodiments are 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 belong to the scope of protection of the present application.

[0063] The terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the terms used in this way are interchangeable where appropriate, so that the embodiments of the present application can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first", "second", etc. are generally of one type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" in the specification and claims represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0064] The description of the method flow in the specification of the present application and the steps of the flowchart in the drawings of the present specification do not have to be strictly executed according to the step numbers, and the method steps can be executed in a different order. Moreover, some steps can be omitted, multiple steps can be combined into one step, and / or one step can be decomposed into multiple steps.

[0065] The following is a detailed description of the acoustic signal separation method, system, device and medium based on the logarithmic processing mechanism provided in the embodiments of the present application in combination with the accompanying drawings and preferred embodiments.

[0066] First, the application scenario of the acoustic signal separation method based on the logarithmic processing mechanism in the embodiment of the present application is described in detail.

[0067] As for TF mask, it originates from the human ear's perception mechanism of sound signals, which is also commonly known as the cocktail party effect. Inspired by the principles of human perception, computational auditory scene analysis (CASA) technology was proposed to generate TF mask, which aims to estimate and separate the sound source of interest from multiple sound sources by utilizing the potential relationship of the inherent acoustic cues of the audio signal.

[0068] Related technologies include methods based on inherent acoustic clues, such as modulation spectrum-based methods that use pitch continuity and amplitude modulation mechanisms to group voiced spectral components in mixed signals to achieve voiced sound separation. Cluster distance-based methods use the ratio of inter-cluster distance to intra-cluster distance to separate speech spectral components of different speakers.

[0069] In addition to the above methods based on inherent acoustic cues, there are also methods that estimate TF masks based on the statistical characteristics of the target and noise. They usually assume that the statistical distribution of the target and noise in the frequency domain is a specific statistical model, and combine different optimization criteria to obtain the TF mask that recovers the target signal from the noisy signal.

[0070] In addition to the above-mentioned sound signal separation methods based on frequency domain models, some methods consider that logarithmic spectrum domain models are more suitable for directly processing sound signals than frequency domain models, namely methods based on logarithmic spectrum domain models, such as methods based on hidden Markov models (HMM) and methods based on Gaussian mixture models.

[0071] However, in the logarithmic spectrum domain, due to the nonlinear characteristics of logarithms, it is a difficult problem to accurately derive the estimation of sound signals from a mathematical perspective. In order to overcome the problem of difficulty in accurately deriving the target sound signal expression caused by logarithmic nonlinearity, many studies have trained sound data through autoregressive models and hidden Markov models to directly obtain the logarithmic distribution of target signals and noise. This method avoids some difficult calculation problems to a certain extent, but noise signals vary greatly and are of many types. The training samples are limited, which makes the noise-based training method have the problem of insufficient noise generalization; that is, the model relies on the specific features or patterns of the training data, and the diversity unique to the field of noise signals leads to the inability to handle new data inputs with changes.

[0072] Based on this, the present application proposes a new technical solution, which derives a new method for obtaining the logarithmic probability density function of the acoustic signal and uses it in the signal separation scheme to estimate the time-frequency mask. Compared with the existing time-frequency mask estimation method based on the logarithmic spectrum domain, the proposed method does not use a training strategy to describe the logarithmic probability distribution of the signal, so it has better generalization to noise and lower complexity.

[0073] This application first adopts a simplified unsupervised strategy to obtain the logarithmic function of noise and target signal, then establishes the logarithmic model relationship between target signal, noise and noisy signal, and finally uses the minimum mean square error (MMSE) estimation strategy under the logarithmic spectrum to separate the target signal. This application estimates TF masking based on the logarithmic probability density function of the acoustic signal, that is, the logarithmic probability density distribution of the signal is directly obtained by unsupervised means, and the TF masking estimate is generated and used for signal separation; since this application does not rely on data training, its computational complexity is low and it is not easily affected by the type of noise. It can run on hardware platforms with lower computing power and has good generalization capabilities for environmental noise. Specifically:

[0074] See also Figure 1-2 , the embodiment of the present application provides a method for separating acoustic signals based on a logarithmic processing mechanism, such as Figure 1-2 As shown, the recommended method of the embodiment of the present application includes the following steps:

[0075] Step S1, receiving a mixed noisy signal, performing preprocessing and extracting parameters; the parameters include the amplitude spectra of the target signal, noise and noisy signal, and a transient signal power estimation.

[0076] The received mixed noisy signal is a signal in the time domain; the preprocessing includes windowing, framing, short-time Fourier transform (STFT), etc., and the relevant parameters are further extracted through the preprocessing.

[0077] Specifically: Assume that the mixed noisy signal is:

[0078] y=s+d

[0079] Among them, y represents the vector of noisy signals, s represents the vector of target signals, and d represents the vector of noise signals.

[0080] Based on this, specifically, step S1 includes:

[0081] Step S101, performing short-time Fourier transform on the mixed noisy signal to obtain amplitude spectra of the target signal, noise and noisy signal.

[0082] Step S102: obtaining a power spectrum of the noisy signal according to the amplitude spectrum of the noisy signal, and calculating a power estimate of the transient signal based on the power spectrum of the noisy signal.

[0083] Specifically, for the short signal power estimation of the target signal, the short energy of the target signal can be estimated by obtaining the minimum value of the power spectrum of the noisy signal in two sub-windows, that is:

[0084]

[0085] Among them, Y c,k is the minimum value of the noisy power spectrum within a certain range after the current moment, Y ac,k is the minimum value of the noisy power spectrum within a certain range before the current moment.

[0086] Power estimation of short signal for noise signal It can be obtained using the classic minimum tracking algorithm, that is, to find the minimum value of the noisy power spectrum within a certain range before the current moment.

[0087] Step S2, constructing a logarithmic probability density distribution model based on the extracted parameters, and obtaining the logarithmic probability density distribution of the target signal, noise and noisy signal through the model.

[0088] The principle of the logarithmic probability density distribution model is: first, the probability density function of the amplitude spectrum is given, and then the probability density function of the amplitude spectrum is converted into the probability density function of the logarithmic spectrum with the help of mathematical statistics and probability theory; wherein, the relationship between the logarithmic amplitude spectrum and the amplitude spectrum is: logarithmic amplitude spectrum = log (amplitude spectrum).

[0089] Specifically, step S2 includes:

[0090] Step S201, performing Fourier transform and logarithmic processing on the mixed noisy signal to obtain a representation of the logarithmic amplitude spectrum of the target signal, noise and noisy signal.

[0091] After Fourier transform and logarithmic processing of the mixed noisy signal, the logarithmic amplitude spectrum of the target signal, noise and noisy signal can be obtained, which is expressed as follows:

[0092] S log =[S log,1 ,...,S log,k ,...,S log,K ] T

[0093] N log =[N log,1 ,...,N log,k ,...,N log,K ] T

[0094] Y log =[Y log,1 ,...,Y log,k ,...,Y log,K ] T ,

[0095] Among them, S log Represents the amplitude log spectrum vector of the target signal, N log Represents the magnitude log spectrum vector of the noise signal, Y log represents the amplitude logarithmic spectrum vector of the noisy signal, K represents the total number of vector elements, and k represents the kth element in the vector.

[0096] Step S202: Approximate the probability density distribution of the amplitude spectrum of the target signal and the noise using Laplace distribution and Rayleigh distribution respectively.

[0097] When it is assumed that the different spectral components of the target signal and the noise are uncorrelated, they can be processed independently. Considering that the distribution of the target signal is more inclined to super-Gaussian distribution, the amplitude distribution of the target signal and the noise are represented by Laplace and Rayleigh distribution respectively:

[0098]

[0099]

[0100] Among them, vectors SS and NN are the amplitude spectrum vectors of the target signal and noise, and can be expressed as SS = [SS1...,SS k...,SS K ] T and NN=[NN1...,NN k ...,NN K ] T ,λ s and λ n The power estimation of the short signal representing the target signal and the noise is calculated and obtained in step S102.

[0101] Step S203: convert the probability density distribution of the amplitude spectrum of the target signal and the noise into a corresponding logarithmic probability density distribution.

[0102] Considering that the logarithmic distribution characteristics are similar to the perceptual characteristics of the human ear basilar membrane to frequency, the audio signal is more suitable for processing in the logarithmic domain. To this end, the present application converts the probability density distribution of the target signal and the noise amplitude spectrum into their logarithmic probability density distribution. Further, step S202 includes:

[0103] Step S203a, respectively calculating the logarithmic cumulative distribution functions of the target signal and the noise amplitude spectrum coefficients.

[0104] For the noise component and the target component, the logarithmic cumulative distribution function of their amplitude spectrum coefficients can be obtained first, as shown below:

[0105]

[0106]

[0107] Step S203b, respectively deriving the corresponding logarithmic cumulative distribution functions to obtain the logarithmic probability density distribution of the target signal and the noise.

[0108] After calculating the logarithmic cumulative distribution function of the target signal and the noise, the logarithmic probability density distribution of the target signal and the noise signal can be obtained by respectively taking the derivative of their cumulative distribution functions, that is, the logarithmic probability density distribution of the target signal and the noise can be obtained as:

[0109]

[0110]

[0111] Step S204, based on the mixed maximum model relationship between the target signal, the noise and the noisy signal in the logarithmic domain, obtain the logarithmic probability density distribution of the noisy signal.

[0112] In order to achieve the subsequent separation of the target sound signal, it is also necessary to obtain the logarithmic probability density distribution of the noisy signal.

[0113] By using the mixed maximum model relationship between the target signal, noise and noisy signal in the logarithmic domain: log =max(S log ,N log ), we can get the logarithmic probability density distribution of the noisy signal as:

[0114]

[0115] Step S3, based on the logarithmic probability density distribution of the target signal, noise and noisy signal, a TF mask is generated using MMSE estimation theory.

[0116] Using the minimum mean square error (MMSE) estimation, the minimum mean square error estimate of the log spectrum of the target signal can be obtained as:

[0117]

[0118] in, is the logarithmic probability distribution of the noisy signal under the condition of known target signal log spectrum, which can be obtained through its cumulative distribution function, that is:

[0119]

[0120] Among them, U(·) is a step function. Based on the above formula, the logarithmic distribution function of the noisy signal can be obtained as follows:

[0121]

[0122] Here, δ(·) is the Dirac function.

[0123] Based on this, the minimum mean square error estimation of the logarithmic spectrum of the target signal can be further expressed as:

[0124]

[0125] Among them, C is the Euler constant, and its value is 0.577256649; E i (-z) is the exponential integral function, and its expression is

[0126] Therefore, using the minimum mean square error value of the logarithmic spectrum of the target signal, the TF mask (time-frequency mask) of the target signal can be obtained.

[0127]

[0128] Step S4: Process the noisy signal according to the TF mask to obtain a spectrum amplitude estimation of the target signal.

[0129] Using the obtained TF mask of the target signal, the spectral amplitude estimate of the target signal can be calculated as,

[0130]

[0131] Furthermore, the spectrum amplitude estimation of the target signal can be obtained:

[0132]

[0133] Finally, the inverse short-time Fourier transform (ISTFT) is used to perform overlap-addition to achieve signal separation.

[0134] This application adopts a method based on an unsupervised strategy to obtain the logarithmic function of noise and target signal. This method converts the distribution of target signal and noise signal in frequency domain to logarithmic domain through a transformation function, and then uses the logarithmic relationship between target signal, noise and noisy signal to separate the target signal using the minimum mean square error estimation theory. Compared with the signal separation method based on data training, this unsupervised method makes the method less susceptible to the influence of noise type, has better generalization ability for noise, and is better adaptable to different environments in specific use.

[0135] At the same time, since the present application does not rely on data training, it does not have large model parameters, thereby reducing the computational complexity during implementation, so that it can run on a hardware platform with lower computing power.

[0136] In summary, the acoustic signal separation method proposed in this application can estimate the target acoustic signal from the noisy signal. Since this application adopts an unsupervised method to obtain the logarithmic spectrum distribution of the sound signal, it is insensitive to the type of noise, has good generalization, and has low computational complexity, and can achieve noise suppression effects in many low-computing hardware platforms.

[0137] See also Figure 3 , corresponding to the above-mentioned embodiment of the acoustic signal separation method based on the logarithmic processing mechanism, the embodiment of the present application provides an acoustic signal separation system based on the logarithmic processing mechanism, the system comprising:

[0138] The preprocessing module 1001 is used to receive the mixed noisy signal, perform preprocessing and extract parameters; the parameters include the amplitude spectrum of the target signal, noise and noisy signal and the short signal power estimation;

[0139] A logarithmic probability density distribution model module 1002 is used to construct a logarithmic probability density distribution model based on the extracted parameters, and obtain the logarithmic probability density distribution of the target signal, noise and noisy signal through the model;

[0140] A TF mask module 1003 is used to generate a TF mask based on the logarithmic probability density distribution of the target signal, the noise and the noisy signal using the MMSE estimation theory;

[0141] The signal separation module 1004 is used to process the noisy signal according to the TF mask to obtain a spectrum amplitude estimation of the target signal.

[0142] Furthermore, the logarithmic probability density distribution model module is specifically used for:

[0143] Performing Fourier transform and logarithmic processing on the mixed noisy signal to obtain a representation of the logarithmic amplitude spectrum of the target signal, the noise, and the noisy signal;

[0144] Using Laplace distribution and Rayleigh distribution to approximate the probability density distribution of the amplitude spectrum of the target signal and the noise respectively;

[0145] Converting the probability density distribution of the amplitude spectrum of the target signal and the noise into a corresponding logarithmic probability density distribution;

[0146] Based on the mixed maximum model relationship between the target signal, the noise and the noisy signal in the logarithmic domain, the logarithmic probability density distribution of the noisy signal is obtained.

[0147] The above-mentioned sound signal separation system based on logarithmic processing mechanism implements the steps and various processes of the above-mentioned sound signal separation method embodiment based on logarithmic processing mechanism, and can achieve the same technical effect. To avoid repetition, they are not repeated here.

[0148] See also Figure 4 Corresponding to the above-mentioned embodiment of the acoustic signal separation method based on the logarithmic processing mechanism, an embodiment of the present application provides an electronic device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, the steps of the above-mentioned embodiment of the acoustic signal separation method based on the logarithmic processing mechanism and the various processes of the embodiment are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0149] The memory 1009 can be used to store software programs and various data. The memory 1009 may mainly include a first storage area for storing programs or instructions and a second storage area for storing data, wherein the first storage area may store an operating system, an application program or instructions required for at least one function (such as a sound playback function, an image playback function, etc.), etc. In addition, the memory 1009 may include a volatile memory or a non-volatile memory, or the memory 1009 may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), a static random access memory (SRAM), a dynamic random access memory (DRAM), a synchronous dynamic random access memory (SDRAM), a double data rate synchronous dynamic random access memory (DDRSDRAM), an enhanced synchronous dynamic random access memory (ESDRAM), a synchronous link dynamic random access memory (SLDRAM) and a direct memory bus random access memory (DRRAM). The memory 1009 in the embodiment of the present application includes but is not limited to these and any other suitable types of memory.

[0150] The processor 1010 may include one or more processing units; optionally, the processor 1010 integrates an application processor and a modem processor, wherein the application processor mainly processes operations related to an operating system, a user interface, and application programs, and the modem processor mainly processes wireless communication signals, such as a baseband processor. It is understandable that the modem processor may not be integrated into the processor 1010.

[0151] Corresponding to the above-mentioned embodiment of the sound signal separation method based on the logarithmic processing mechanism, the embodiment of the present application also provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by the processor, the steps and various processes of the above-mentioned embodiment of the sound signal separation method based on the logarithmic processing mechanism are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0152] The processor is the processor in the electronic device described in the above embodiment of the present application. The readable storage medium includes a computer readable storage medium, such as a computer read-only memory ROM, a random access memory RAM, a magnetic disk or an optical disk.

[0153] It should be noted that, in this article, the terms "comprise", "include" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes 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 one..." do not exclude the presence of other identical elements in the process, method, article or device including the element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved, for example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0154] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM / RAM, a disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0155] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.

Claims

1. A method for separating acoustic signals based on a logarithmic processing mechanism, characterized in that: include: receiving a mixed noisy signal, performing preprocessing and extracting parameters; the parameters include amplitude spectra of the target signal, noise and noisy signal, and a short signal power estimate; Based on the extracted parameters, a logarithmic probability density distribution model is constructed, and the logarithmic probability density distribution of the target signal, noise and noisy signal is obtained through the model; Based on the logarithmic probability density distribution of the target signal, the noise and the noisy signal, a TF mask is generated using MMSE estimation theory; The noisy signal is processed according to the TF mask to obtain a spectrum amplitude estimation of the target signal.

2. The acoustic signal separation method based on logarithmic processing mechanism according to claim 1 is characterized in that: The receiving of the mixed noisy signal, preprocessing and extracting parameters includes: Performing short-time Fourier transform on the mixed noisy signal to obtain amplitude spectra of the target signal, noise and noisy signal; The power spectrum of the noisy signal is obtained according to the amplitude spectrum of the noisy signal, and the power estimation of the transient signal is calculated based on the power spectrum of the noisy signal.

3. The acoustic signal separation method based on logarithmic processing mechanism according to claim 1 is characterized in that: The method of constructing a logarithmic probability density distribution model based on the extracted parameters, and obtaining the logarithmic probability density distribution of the target signal, noise and noisy signal through the model, includes: Performing Fourier transform and logarithmic processing on the mixed noisy signal to obtain a representation of the logarithmic amplitude spectrum of the target signal, the noise, and the noisy signal; Using Laplace distribution and Rayleigh distribution to approximate the probability density distribution of the amplitude spectrum of the target signal and the noise respectively; Converting the probability density distribution of the amplitude spectrum of the target signal and the noise into a corresponding logarithmic probability density distribution; Based on the mixed maximum model relationship between the target signal, the noise and the noisy signal in the logarithmic domain, the logarithmic probability density distribution of the noisy signal is obtained.

4. The acoustic signal separation method based on logarithmic processing mechanism according to claim 3 is characterized in that: The step of converting the probability density distribution of the amplitude spectrum of the target signal and the noise into a corresponding logarithmic probability density distribution comprises: Calculate the logarithmic cumulative distribution function of the target signal and noise amplitude spectrum coefficients respectively; The corresponding logarithmic cumulative distribution functions are derived respectively to obtain the logarithmic probability density distribution of the target signal and the noise.

5. The acoustic signal separation method based on logarithmic processing mechanism according to claim 4 is characterized in that: The logarithmic amplitude spectrum of the target signal, noise and noisy signal is expressed as: S log =[S log,1 ,...,S log,k ,...,S log,K ] T N log =[N log,1 ,...,N log,k ,...,N log,K ] T AND log =[And log,1 ,...,AND log,k ,...,AND log,K ] T , Among them, S log Represents the amplitude log spectrum vector of the target signal, N log Represents the magnitude log spectrum vector of the noise signal, Y log represents the amplitude log spectrum vector of the noisy signal, K represents the total number of vector elements, and k represents the kth element in the vector; The probability density distribution of the amplitude spectrum of the target signal and the noise is approximated by Laplace distribution and Rayleigh distribution respectively, including: Among them, vectors SS and NN are the amplitude spectrum vectors of the target signal and noise, and can be expressed as SS = [SS1...,SS k ...,SS K ] T and NN=[NN1...,NN k ...,NN K ] T ,λ s and λ n Power estimates of short bursts representing target signals and noise; The logarithmic probability density distribution of the target signal and noise is: The logarithmic probability density distribution of the noisy signal is:

6. The acoustic signal separation method based on logarithmic processing mechanism according to claim 5, characterized in that: The TF mask is: The spectral amplitude of the target signal is estimated as:

7. An acoustic signal separation system based on a logarithmic processing mechanism, characterized in that: include: A preprocessing module, used to receive a mixed noisy signal, perform preprocessing and extract parameters; the parameters include the amplitude spectrum of the target signal, noise and noisy signal and the power estimation of the short signal; A logarithmic probability density distribution model module, used to construct a logarithmic probability density distribution model based on the extracted parameters, and obtain the logarithmic probability density distribution of the target signal, noise and noisy signal through the model; A TF mask module, used for generating a TF mask based on the logarithmic probability density distribution of the target signal, the noise and the noisy signal using the MMSE estimation theory; A signal separation module is used to process the noisy signal according to the TF mask to obtain a spectrum amplitude estimation of the target signal.

8. The acoustic signal separation system based on logarithmic processing mechanism according to claim 7, characterized in that: The logarithmic probability density distribution model module is specifically used for: Using Laplace distribution and Rayleigh distribution to approximate the probability density distribution of the amplitude spectrum of the target signal and the noise respectively; Converting the probability density distribution of the amplitude spectrum of the target signal and the noise into a corresponding logarithmic probability density distribution; Based on the mixed maximum model relationship between the target signal, the noise and the noisy signal in the logarithmic domain, the logarithmic probability density distribution of the noisy signal is obtained.

9. An electronic device, characterized in that: The electronic device comprises: a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the computer program is executed by the processor, the steps of the acoustic signal separation method based on the logarithmic processing mechanism as claimed in any one of claims 1 to 6 are implemented.

10. A readable storage medium, characterized in that: The readable storage medium stores a program or instruction, and when the program or instruction is executed by a processor, the steps of the acoustic signal separation method based on a logarithmic processing mechanism as claimed in any one of claims 1 to 6 are implemented.

Images