Monophonic speech noise reduction method, system, device and readable storage medium

Abstract

The present invention belongs to the field of speech noise reduction, and discloses a monophonic speech noise reduction method, system, device and readable storage medium, including obtaining a monophonic speech to be denoised; constructing an initial noise reduction model based on an LSTM neural network; obtaining a preset number of enhanced training samples, using a preset number of enhanced training samples to train the initial noise reduction model to obtain a noise reduction model; using the noise reduction model to reduce the noise of the monophonic speech to be reduced to obtain human voice audio. The noise reduction process is not limited to the limitation of binaural channels, and can achieve noise reduction processing for any monophonic speech. Based on the characteristics that the noise categories of the entire time series in the same speech are roughly the same, the model is trained based on the LSTM neural network, which facilitates learning the noise law of the entire time series, thereby achieving a better noise reduction effect. At the same time, based on the complexity of noise influencing factors, the initial noise reduction model is trained again by enhancing the training samples to further improve the noise reduction effect of the noise reduction model.

Description

Technical Field

[0001] The present invention belongs to the field of speech noise reduction, and relates to a monophonic speech noise reduction method, system, device and readable storage medium. Background Art

[0002] A few years ago, we often heard various noises when we were on the phone, which greatly affected the call quality. But now, with the popularity of smartphones, we can clearly feel the reduction of noise during calls. This is because most smartphones now use Qualcomm chips, and most of these chips are equipped with Qualcomm's proprietary CVC technology. CVC technology is a call noise reduction technology. Its working principle is to obtain sound through the built-in dual microphones of the mobile phone during a call. Among them, the main microphone is at the speaker's mouth and can receive a louder speaker's voice; the secondary microphone is far away from the speaker's mouth and receives a smaller speaker's voice, while the two microphones can receive almost the same amount of environmental noise. By combining the sound signals collected by the main and secondary microphones, we can use a certain algorithm to distinguish which sounds are the speakers we want, thereby achieving noise reduction calls.

[0003] However, the technology still has the following defects. First, the technology cannot process mono audio, and requires the mobile phone to have dual microphones, which is not effective for single-microphone mobile phones. In addition, it has certain requirements for the speaker's posture during the call, requiring the speaker's voice source to be very close to the main microphone. If the speaker is far away from the microphone or wearing a headset with a single microphone, it cannot be used. Summary of the invention

[0004] The purpose of the present invention is to overcome the shortcomings of the above-mentioned prior art that the existing call noise reduction technology cannot process monophonic audio, and to provide a monophonic voice noise reduction method, system, device and readable storage medium.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] In a first aspect of the present invention, a method for reducing noise of a single-channel speech comprises the following steps:

[0007] Obtain the monophonic speech to be denoised; construct an initial denoising model based on the LSTM neural network; obtain a preset number of enhanced training samples, and use the preset number of enhanced training samples to train the initial denoising model to obtain a denoising model; denoise the monophonic speech to be denoised by using the denoising model to obtain human voice audio.

[0008] The monophonic speech noise reduction method of the present invention is further improved in that:

[0009] The construction of the initial noise reduction model based on the LSTM neural network includes: obtaining a number of human voice audios and a number of noise audios and randomly combining them to obtain a number of mixed audios, each of which includes a human voice audio and at least one noise audio; performing frame windowing and Fourier transform on the mixed audio to obtain a number of mixed audio frame spectra; dividing the number of mixed audio frame spectra into a training set and a test set, establishing an LSTM neural network model for binary classification, training the LSTM neural network model through the training set, testing the trained LSTM neural network model through the test set, and obtaining the initial noise reduction model after passing the test.

[0010] The training of the LSTM neural network model through the training set includes: inputting the mixed audio frame spectrum in the training set into the LSTM neural network model, obtaining the human voice spectrum and the noise spectrum and performing inverse Fourier transform to obtain predicted human voice audio and noise audio; according to the error between the predicted human voice audio and the actual human voice audio, iteratively updating each parameter in the LSTM neural network model until the number of training times reaches a preset value or the error between the predicted human voice audio and the actual human voice audio no longer decreases.

[0011] The method of obtaining a preset number of enhanced training samples and using the preset number of enhanced training samples to train the initial noise reduction model includes: obtaining a number of monophonic voices that fail to meet the noise reduction requirements of the initial noise reduction model as a preset number of enhanced training samples; obtaining a number of test samples formed by combining human voice audio and a number of noise audios; and using an unsupervised learning method to train the initial noise reduction model using the preset number of enhanced training samples until the noise reduction effect of the trained initial noise reduction model on the test samples is within a preset error range and the noise reduction effect of the trained initial noise reduction model on the enhanced training samples is greater than a preset threshold of the noise reduction effect of the initial noise reduction model on the enhanced training samples.

[0012] When the initial denoising model is trained by using a preset number of enhanced training samples, the parameters of several hidden layers in the initial denoising model are fixed.

[0013] Before training the initial denoising model with a preset number of enhanced training samples, a number of random nonlinear layers are added between the hidden layer and the classification layer of the initial denoising model.

[0014] The method of reducing the noise of the monophonic speech to be reduced by using a noise reduction model to obtain human voice audio includes: performing Fourier transform on the monophonic speech to be reduced to obtain a call speech spectrum; inputting the call speech spectrum into the noise reduction model to obtain a human voice spectrum and a noise spectrum; and performing inverse Fourier transform on the human voice spectrum to obtain human voice audio.

[0015] A second aspect of the present invention is a monophonic speech noise reduction system, comprising:

[0016] An acquisition module is used to acquire the monophonic speech to be denoised; a model building module is used to build an initial denoising model based on an LSTM neural network; an enhanced training module is used to acquire a preset number of enhanced training samples, and use the preset number of enhanced training samples to train the initial denoising model to obtain a denoising model; and a denoising module is used to denoise the monophonic speech to be denoised through the denoising model to obtain human voice audio.

[0017] In a third aspect of the present invention, a terminal device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned monophonic speech noise reduction method when executing the computer program.

[0018] A fourth aspect of the present invention is a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned monophonic speech noise reduction method are implemented.

[0019] Compared with the prior art, the present invention has the following beneficial effects:

[0020] The monophonic speech denoising method of the present invention, after obtaining the monophonic speech to be denoised, constructs an initial denoising model, obtains a preset number of enhanced training samples, uses the preset number of enhanced training samples to train the initial denoising model, obtains the denoising model, and then directly denoises the monophonic speech to be denoised through the denoising model to obtain human voice audio, the denoising process is not limited to the limitation of binaural channels, and can realize the denoising processing of any monophonic speech, wherein the denoising model is obtained by training the initial denoising model with enhanced training samples, and the initial denoising model is constructed by training the LSTM neural network with mixed audio, based on the characteristics that the noise categories of the entire time series in the same voice are roughly the same, by using the LSTM neural network as the basis for model training, it is easy to learn the noise law of the entire time series, and then achieve a better denoising effect. At the same time, based on the complexity of noise influencing factors, it is difficult for training data to cover all possible noise environments, which leads to the initial denoising model not being robust to some noises, and the initial denoising model is retrained by enhancing the training samples, and the noise types that have not been learned are relearned, and the denoising effect of the denoising model is further improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a flowchart of a method for single-channel speech noise reduction according to an embodiment of the present invention;

[0022] Figure 2 This is a flowchart of the initial denoising model training process according to an embodiment of the present invention;

[0023] Figure 3 This is a flowchart of the denoising model training process according to an embodiment of the present invention;

[0024] Figure 4 This is a structural block diagram of a monophonic speech noise reduction system according to an embodiment of the present invention;

[0025] Figure 5 This is a block diagram of the terminal device structure of an embodiment of the present invention. DETAILED DESCRIPTION

[0026] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0027] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0028] The present invention is further described in detail below in conjunction with the accompanying drawings:

[0029] See also Figure 1 In one embodiment of the present invention, a monophonic voice noise reduction method is provided to achieve effective noise reduction of monophonic voice, which can be well used for single-microphone mobile phones and single-microphone headset calls. Specifically, the monophonic voice noise reduction method includes the following steps:

[0030] S1: Obtain the monophonic speech to be denoised.

[0031] Specifically, when using a single-microphone mobile phone or a single-microphone headset for a call, a mono voice is obtained in real time through the microphone, and the mono voice here can also be a mono voice downloaded from the Internet. When performing noise reduction processing on the call voice, the noise reduction can be performed before the mono voice is sent, or when the mono voice is sent to the corresponding call device, or at the same time.

[0032] S2: Build an initial denoising model based on LSTM neural network.

[0033] Specifically, in order to obtain a noise reduction model for monophonic speech noise reduction, first, an initial noise reduction model needs to be obtained. In addition, since the call process occurs in a time series, and in many cases, the noise category of the entire time series in the same call is roughly the same (such as the sound of wind throughout), in this embodiment, an LSTM long short-term memory network is used to construct an initial model, which is convenient for learning the noise law of the entire time series, and then an initial noise reduction model is obtained by training the LSTM neural network model through mixed audio. For details, see Figure 2 , including the following steps:

[0034] S201: A large number of pure human voice audios and pure noise audios are collected, and then the human voice audios and noise audios are randomly combined to obtain a number of mixed audios. In order to achieve the noise reduction requirement, each mixed audio includes a human voice audio and at least one noise audio, and may also include multiple human voice audios.

[0035] S202: The mixed audio is subjected to framing windowing and Fourier transform to obtain several mixed audio frame spectra. Specifically, the mixed audio is subjected to framing windowing. In this embodiment, the mixed audio is framed according to the framing requirements of 25ms frame length and 10ms frame shift. After the mixed audio is framed, each frame of the mixed audio needs to be windowed. The window function of the windowing process generally has a low-pass characteristic. The purpose of the windowing function is to reduce leakage in the frequency domain. The commonly used window functions in speech signal analysis include rectangular windows, Hamming windows and Hanning windows. Different window functions can be selected according to different situations. Then, through Fourier transform, the mixed audio after framing windowing is converted into time domain and frequency domain, and the time domain features are mapped into frequency spectrum features of a certain dimension, such as mapping into several basic sine waves, thereby obtaining several mixed audio frame spectra.

[0036] S203: Divide the spectra of several mixed audio frames into a training set and a test set, establish an LSTM neural network model for binary classification, train the LSTM neural network model through the training set, test the trained LSTM neural network model through the test set, and obtain an initial noise reduction model after passing the test.

[0037] Specifically, a number of mixed audio frame spectra are randomly divided into a training set and a test set. The ratio of the mixed audio frame spectra in the training set and the test set can be custom set. During iterative training, the mixed audio frame spectra in the training set and the test set can also be exchanged.

[0038] Then, an LSTM neural network model for binary classification is established. The existing LSTM neural network model generally includes an input layer, several hidden layers and an output layer. The mixed audio frame spectrum is used as the input layer, and several hidden layers are set. The separated human voice spectrum and noise spectrum are used as the binary classification results of the output layer. Then, the constructed LSTM neural network model is trained with the training set, and the trained LSTM neural network model is tested with the test set. When the pass rate of the test result meets the preset pass rate threshold, the test is qualified and the initial noise reduction model is obtained. Among them, the pass rate of the test result refers to the ratio of the mixed audio frame spectrum of the clear human voice audio obtained after each mixed audio frame spectrum in the test set passes through the trained LSTM neural network model to all mixed audio frame spectra.

[0039] Among them, the specific method of training the LSTM neural network model through the training set is:

[0040] The mixed audio frame spectrum in the training set is input into the constructed LSTM neural network model, and the predicted human voice spectrum and noise spectrum are output, and then the human voice audio and noise audio are obtained through inverse Fourier transform. Then, according to the error between the predicted human voice audio and the actual human voice audio, this error is back-propagated and the various parameters in the LSTM neural network model, including the forget gate, input gate and output gate weights, are iteratively updated through the stochastic gradient descent algorithm until the number of training times reaches the preset value or the error between the predicted human voice audio and the actual human voice audio no longer decreases.

[0041] Preferably, when training the LSTM neural network model through the training set, the learning rate of the LSTM neural network model is updated through the annealing algorithm. Among them, the learning rate is a hyperparameter that guides how to use the gradient of the loss function to adjust the model weight in the gradient descent method. If the learning rate is too large, the loss function may directly cross the global optimal point. If the learning rate is too small, the loss function changes very slowly, which will greatly increase the convergence complexity of the network and it is easy to be trapped in the local minimum or saddle point. By using the annealing algorithm, the learning rate can be changed over time, which can solve this problem well.

[0042] Then, the trained LSTM neural network model is tested with the test set, and the mixed audio frame spectrum in the test set is input into the trained LSTM neural network model. When the denoising pass rate of the mixed audio frame spectrum in the test set is within the preset range, the trained LSTM neural network model can be considered to have passed the test and the initial denoising model is obtained.

[0043] S3: Obtain a preset number of enhanced training samples, and use the preset number of enhanced training samples to train an initial denoising model to obtain a denoising model.

[0044] Specifically, based on the initial noise reduction model, the noise reduction requirements of monophonic speech in some scenarios can be achieved. However, the inventors found in actual work that due to the complexity of various factors such as noise type, hardware, microphone recording distance, etc., it is difficult for training data to cover all possible noise environments, which leads to the initial noise reduction model not being robust to some noises. When the initial noise reduction model is performing noise reduction, the intelligibility and clarity of the denoised human voice audio sometimes still cannot meet the requirements. At this time, it is necessary to perform noise data pattern learning on the initial noise reduction model according to specific noise data, that is, enhanced training of the initial noise reduction model, the purpose of which is to re-learn the types of noise that have not been learned, so as to achieve pattern recognition of this type of noise and achieve the purpose of noise reduction. Based on this, in this embodiment, the monophonic speech that fails to be denoised by the initial noise reduction model is used as an enhanced training sample, and the final noise reduction model is obtained by training the initial noise reduction model with the enhanced training sample.

[0045] Specifically, in this embodiment, a method for obtaining enhanced training samples is provided. Based on the premise of noise continuity, generally when a monophonic speech is denoised by the initial noise reduction model, if there is a similar waveform in the obtained human voice audio that appears more than a preset number of times, and the energy of the similar waveform is greater than a preset energy threshold, it is considered that the current monophonic speech is unqualified for noise reduction, and these unqualified monophonic speech are used as enhanced training samples, and different users generate different enhanced training samples.

[0046] For details, see Figure 3 ,The specific method of using enhanced training samples to train the initial denoising model is:

[0047] A number of monophonic speech samples that fail to meet the noise reduction requirements of an initial noise reduction model are obtained as a preset number of enhanced training samples; a number of test samples formed by combining human voice audio and a number of noise audio samples are obtained; an initial noise reduction model is trained using a preset number of enhanced training samples in an unsupervised learning manner, until the noise reduction effect of the trained initial noise reduction model on the test samples is within a preset error range, and the noise reduction effect of the trained initial noise reduction model on the enhanced training samples is greater than a preset threshold of the noise reduction effect of the initial noise reduction model on the enhanced training samples, thereby obtaining a noise reduction model.

[0048] If the initial denoising model after enhanced training causes a significant loss in the denoising effect of the test sample, or has no obvious effect on the denoising of the enhanced training sample, the initial denoising model will not be updated, and the enhanced training will be considered to have failed, and the enhanced training sample will be obtained again for training.

[0049] When the initial denoising model is trained by a preset number of enhanced training samples, the parameters of several hidden layers in the initial denoising model are fixed. Since the training is an unsupervised learning process, it is necessary to fix some hidden layers in the initial denoising model, and no parameter adjustment is made during the enhanced training process to prevent overfitting of the model training. Regarding the selection of the fixed parameter hidden layers, generally all hidden layers except the classification layer are selected.

[0050] Preferably, when the number of visual enhancement training samples exceeds a certain number, several random nonlinear layers are added between the hidden layer and the classification layer of the initial denoising model. The purpose of adding the additional random nonlinear layers is to ensure that the model can perform higher modeling of this noise pattern in order to obtain better training results.

[0051] S4: The monophonic speech to be denoised is denoised by using a denoising model to obtain human voice audio.

[0052] Specifically, in this embodiment, the monophonic speech to be denoised is subjected to Fourier transform to obtain a call speech spectrum, and the spectrum is extracted as a feature; the call speech spectrum is input into a preset noise reduction model to obtain a human voice spectrum and a noise spectrum; the human voice spectrum is subjected to inverse Fourier transform to obtain human voice audio.

[0053] In summary, the monophonic speech denoising method of the present invention, after obtaining the monophonic speech to be denoised, constructs an initial denoising model, and obtains a preset number of enhanced training samples, uses a preset number of enhanced training samples to train the initial denoising model, obtains a denoising model, and directly denoises the monophonic speech to be denoised through the denoising model to obtain human voice audio, wherein the denoising model is obtained by training the initial denoising model with enhanced training samples, and the initial denoising model is constructed by training the LSTM neural network model with mixed audio, based on the characteristics that the noise categories of the entire time series in the same voice are roughly the same, and the model is trained based on the LSTM neural network model, so as to facilitate learning the noise law of the entire time series, thereby achieving a better denoising effect. At the same time, based on the complexity of noise influencing factors, it is difficult for training data to cover all possible noise environments, which leads to the initial denoising model not being robust to some noises, and the initial denoising model is retrained by enhancing the training samples, and the noise types that have not been learned are relearned, further improving the denoising effect of the denoising model.

[0054] The biggest advantage of this mono voice noise reduction method is that it is "mono", which has the advantages of wide application range and good compatibility. Applied to mobile phone calls, this method does not require the mobile phone to have dual microphones, and is applicable to all mobile phones with microphones. Therefore, after the method is applied, mobile phone manufacturers no longer need to equip mobile phones with dual microphones, which can reduce the cost and weight of mobile phones, and save space for a microphone to make the mobile phone thinner. Secondly, this method has no requirements on the distance between the speaker and the microphone. It can be used not only in scenarios where the speaker is far away from the microphone, but also in scenarios where the speaker uses a single-microphone headset to make calls. Finally, this method can also perform noise reduction on mono voice obtained by any means, such as voice downloaded from the Internet.

[0055] The following are device embodiments of the present invention, which can be used to perform method embodiments of the present invention. For details not disclosed in the device embodiments, please refer to the method embodiments of the present invention.

[0056] See also Figure 4 In another embodiment of the present invention, a monophonic speech noise reduction system is provided, which can be used to implement the above-mentioned monophonic speech noise reduction method. Specifically, the monophonic speech noise reduction system includes an acquisition module, a model building module, an enhanced training module and a noise reduction module.

[0057] Among them, the acquisition module is used to obtain the monophonic speech to be denoised; the model construction module is used to build an initial denoising model based on the LSTM neural network; the enhanced training module is used to obtain a preset number of enhanced training samples, and use the preset number of enhanced training samples to train the initial denoising model to obtain the denoising model; the denoising module is used to denoise the monophonic speech to be denoised through the denoising model to obtain human voice audio.

[0058] Preferably, the model building module of the monophonic speech noise reduction system includes a mixed audio acquisition module, a mixed audio processing module and a training module.

[0059] Among them, the mixed audio acquisition module is used to obtain a number of human voice audios and a number of noise audios and randomly combine them to obtain a number of mixed audios, each of which includes a human voice audio and at least one noise audio; the mixed audio processing module is used to perform frame windowing and Fourier transform on the mixed audio to obtain a number of mixed audio frame spectra; the training module is used to divide the number of mixed audio frame spectra into a training set and a test set, establish an LSTM neural network model for binary classification, train the LSTM neural network model through the training set, test the trained LSTM neural network model through the test set, and obtain the initial noise reduction model after passing the test.

[0060] Preferably, the training module of the monophonic speech noise reduction system includes a prediction module and a parameter updating module.

[0061] Among them, the prediction module is used to input the mixed audio frame spectrum in the training set into the LSTM neural network model, obtain the human voice spectrum and the noise spectrum and perform inverse Fourier transform to obtain the predicted human voice audio and noise audio; the parameter update module is used to iteratively update the parameters in the LSTM neural network model according to the error between the predicted human voice audio and the actual human voice audio, until the number of training times reaches the preset value or the error between the predicted human voice audio and the actual human voice audio no longer decreases.

[0062] Preferably, the enhanced training module of the monophonic speech noise reduction system includes an enhanced training sample acquisition module, a test sample acquisition module and an enhanced training management module.

[0063] Among them, the enhanced training sample acquisition module is used to obtain a number of monophonic speech that fail the noise reduction of the initial noise reduction model as a preset number of enhanced training samples; the test sample acquisition module is used to obtain a number of test samples formed by combining human voice audio and a number of noise audios; the enhanced training management module is used to train the initial noise reduction model through a preset number of enhanced training samples in an unsupervised learning manner, until the noise reduction effect of the trained initial noise reduction model on the test samples and the noise reduction effect of the initial noise reduction model on the test samples are within a preset error, and the noise reduction effect of the trained initial noise reduction model on the enhanced training samples is greater than the preset threshold of the noise reduction effect of the initial noise reduction model on the enhanced training samples.

[0064] Preferably, the enhanced training module of the monophonic speech denoising system further comprises a parameter fixing module for fixing the parameters of several hidden layers in the initial denoising model when the initial denoising model is trained by a preset number of enhanced training samples.

[0065] Preferably, the enhanced training module of the monophonic speech denoising system further includes a random nonlinear layer adding module for adding a number of random nonlinear layers between the hidden layer and the classification layer of the initial denoising model before training the initial denoising model with a preset number of enhanced training samples.

[0066] See also Figure 5In another embodiment of the present invention, a terminal device is provided, including: at the hardware level, the terminal device includes: a processor and a memory, and optionally also includes an internal bus and a network interface. Among them, the memory may include a memory, such as a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage, etc. Of course, the terminal device may also include hardware required for other services. The processor, the network interface, and the memory are interconnected through an internal bus, and the internal bus can be an industrial standard architecture bus, a peripheral component interconnection standard bus, an extended industrial standard structure bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. The memory is used to store programs. Specifically, the program may include a program code, and the program code includes computer operation instructions. The memory may include a memory and a non-volatile memory, and provides instructions and data to the processor.

[0067] The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it, forming the above-mentioned terminal device at the logical level. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, and is specifically suitable for loading and executing one or more instructions to implement the corresponding method flow or corresponding function; the processor described in the embodiment of the present invention can be used for the operation of the monophonic speech denoising method, including obtaining the monophonic speech to be denoised; constructing an initial denoising model based on an LSTM neural network; obtaining a preset number of enhanced training samples, using a preset number of enhanced training samples to train the initial denoising model, and obtaining a denoising model; denoising the monophonic speech to be denoised by the denoising model to obtain human voice audio.

[0068] In another embodiment, the present invention further provides a computer-readable storage medium (Memory), which is a memory device in a terminal device for storing programs and data. It is understandable that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and the extended storage medium supported by the terminal device. The computer-readable storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and these instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory.

[0069] One or more instructions stored in a computer-readable storage medium can be loaded and executed by a processor to implement the corresponding steps of the monophonic speech denoising method in the above-mentioned embodiment; one or more instructions in the computer-readable storage medium are loaded by the processor and the following steps are executed: obtaining the monophonic speech to be denoised; constructing an initial denoising model based on an LSTM neural network; obtaining a preset number of enhanced training samples, and using the preset number of enhanced training samples to train the initial denoising model to obtain a denoising model; denoising the monophonic speech to be denoised by the denoising model to obtain human voice audio.

[0070] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0071] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0072] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0073] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0074] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A monophonic speech noise reduction method, characterized in that: The following steps are involved: Get the monophonic voice to be denoised; Construct an initial denoising model based on LSTM neural network; Obtain a preset number of enhanced training samples, and use the preset number of enhanced training samples to train an initial denoising model to obtain a denoising model; The monophonic speech to be denoised is denoised by a denoising model to obtain human voice audio; The obtaining of a preset number of enhanced training samples and using the preset number of enhanced training samples to train the initial denoising model comprises: Acquire a number of monophonic voices that fail to meet the denoising criteria of the initial denoising model as a preset number of enhanced training samples; Obtain a number of test samples formed by combining human voice audio and a number of noise audio; An unsupervised learning approach is adopted to train the initial denoising model with a preset number of enhanced training samples, until the denoising effect of the trained initial denoising model on the test samples is within a preset error range, and the denoising effect of the trained initial denoising model on the enhanced training samples is greater than a preset threshold of the denoising effect of the initial denoising model on the enhanced training samples.

2. The monophonic speech noise reduction method according to claim 1, characterized in that: The construction of the initial denoising model based on the LSTM neural network includes: Acquire a number of human voice audios and a number of noise audios and randomly combine them to obtain a number of mixed audios, each of which includes a human voice audio and at least one noise audio; The mixed audio is subjected to frame division, windowing and Fourier transformation to obtain several mixed audio frame spectra; Several mixed audio frame spectra are divided into training sets and test sets, and an LSTM neural network model for binary classification is established. The LSTM neural network model is trained with the training set, and the trained LSTM neural network model is tested with the test set. When the pass rate of the test results meets the preset pass rate threshold, the test is qualified and the initial noise reduction model is obtained.

3. The monophonic speech noise reduction method according to claim 2, characterized in that: The training of the LSTM neural network model by the training set includes: The mixed audio frame spectrum in the training set is input into the LSTM neural network model to obtain the human voice spectrum and noise spectrum and perform inverse Fourier transform to obtain the predicted human voice audio and noise audio; According to the error between the predicted human voice audio and the actual human voice audio, each parameter in the LSTM neural network model is iteratively updated until the number of training times reaches a preset value or the error between the predicted human voice audio and the actual human voice audio no longer decreases.

4. The monophonic speech noise reduction method according to claim 1, characterized in that: When the initial denoising model is trained by using a preset number of enhanced training samples, the parameters of several hidden layers in the initial denoising model are fixed.

5. The monophonic speech noise reduction method according to claim 1, characterized in that: Before training the initial denoising model with a preset number of enhanced training samples, a number of random nonlinear layers are added between the hidden layer and the classification layer of the initial denoising model.

6. The monophonic speech noise reduction method according to claim 1, characterized in that: The method of reducing the noise of the monophonic speech to be reduced by using the noise reduction model to obtain the human voice audio comprises: Perform Fourier transform on the monophonic speech to be denoised to obtain a call speech spectrum; Input the call speech spectrum into the noise reduction model to obtain the human voice spectrum and the noise spectrum; Perform inverse Fourier transform on the human voice spectrum to obtain the human voice audio.

7. A monophonic speech noise reduction system, characterized in that: include: An acquisition module is used to acquire the monophonic speech to be denoised; Model building module, used to build an initial denoising model based on LSTM neural network; An enhanced training module is used to obtain a preset number of enhanced training samples, and use the preset number of enhanced training samples to train an initial denoising model to obtain a denoising model; as well as A noise reduction module, used to reduce the noise of the monophonic speech to be reduced by using a noise reduction model to obtain human voice audio; The enhanced training module includes an enhanced training sample acquisition module, a test sample acquisition module and an enhanced training management module; wherein the enhanced training sample acquisition module is used to acquire a number of monophonic voices that fail to meet the noise reduction requirements of the initial noise reduction model as a preset number of enhanced training samples; the test sample acquisition module is used to acquire a number of test samples formed by combining human voice audio and a number of noise audios; the enhanced training management module is used to train the initial noise reduction model using a preset number of enhanced training samples in an unsupervised learning manner, until the noise reduction effect of the trained initial noise reduction model on the test samples is within a preset error, and the noise reduction effect of the trained initial noise reduction model on the enhanced training samples is greater than a preset threshold of the noise reduction effect of the initial noise reduction model on the enhanced training samples.

8. A terminal device, characterized in that: The method comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the monophonic speech noise reduction method according to any one of claims 1 to 6 when executing the computer program.

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

Images