Noise reduction method, device, storage medium and electronic device on wearable device

Abstract

The present invention discloses a noise reduction method, device, storage medium and electronic device on a wearable device. The noise reduction method on the wearable device includes: a dual-microphone array, a neural network noise reduction and post-processing solution in a strong noise environment, which makes up for the shortcomings of the traditional signal processing algorithm; in a living environment, comprehensively determine whether the orientation information and transient energy meet the update conditions of the adaptive filter, and then obtain the wearer's clean voice signal by combining the neural network noise reduction algorithm, which largely solves the troubles caused by mixed noise in complex life scenes, and further solves the technical problem of low noise reduction accuracy of wearable devices in the prior art.

Description

Technical Field

[0001] The present invention relates to the field of computers, and in particular to a noise reduction method, device, storage medium and electronic device on a wearable device. Background Art

[0002] With the development of the Internet of Things, more and more smart wearable devices can help or assist people in their daily lives. The requirements for smart wearable devices are getting higher and higher, and they need to have higher noise reduction capabilities in different scenarios.

[0003] In related technologies, the scattered noise is strong in strong noise scenes, the signal-to-noise ratio is very low, and the processing effect of the adaptive filtering algorithm is poor; in life scenes, there may be human voice interference and various steady-state noises, among which human voice noise has strong coherence, and the neural network noise reduction algorithm is not effective. If the steady-state noise is strong, the traditional signal processing algorithm cannot completely process it.

[0004] If the usage scenario of the device switches back and forth, only one of the algorithms cannot achieve a good noise reduction effect. That is, in the prior art, the noise reduction accuracy of wearable devices is low.

[0005] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0006] Embodiments of the present invention provide a noise reduction method, apparatus, storage medium and electronic device on a wearable device, so as to at least solve the technical problem of low noise reduction accuracy of the wearable device in the prior art.

[0007] According to one aspect of an embodiment of the present invention, a noise reduction method on a wearable device is provided, comprising: obtaining a mixed voice signal received by a dual-microphone array, and performing a short-time Fourier transform on the mixed voice signal to obtain a time-frequency domain of first channel data and a time-frequency domain of second channel data; passing the time-frequency domain of the first channel data and the time-frequency domain of the second channel data through an adaptive filter to obtain a near-end estimation signal; passing the near-end estimation signal through a noise estimation algorithm to obtain noise energy; when the noise energy value is greater than or equal to a first threshold, passing the near-end estimation signal through a neural network noise reduction and a post-processing algorithm to obtain a clean time-frequency domain voice signal of the wearer, wherein the wearer wears the wearable device; performing an inverse short-time Fourier transform on the clean time-frequency domain voice signal to obtain a time-domain signal of the wearer's voice.

[0008] Optionally, the method also includes: when the noise energy value is less than a first threshold, obtaining angle information of each frame by azimuth estimation of the near-end estimation signal; estimating the first speech energy corresponding to the first channel data and the second speech energy corresponding to the second channel data; sending the first speech energy and the second channel speech energy to the adaptive filter, and updating the adaptive filter when the angle information or the first speech energy and the second speech energy meet preset conditions; denoising the speech signal filtered by the adaptive filter through the neural network to obtain a clean speech signal of the wearer; and performing short-time Fourier inverse transform on the clean time-frequency domain speech signal to obtain the wearer's speech time domain signal.

[0009] Optionally, when the angle information or the first speech energy and the second speech energy meet a preset condition, the adaptive filter is updated, including: when the angle information is greater than preset angle information, or the first speech energy is less than the second speech energy, or the difference between the first speech energy and the second speech energy is less than a preset speech energy value, the adaptive filter is updated.

[0010] Optionally, the method further includes: if the angle information or the first speech energy and the second speech energy do not satisfy a preset condition, not updating the adaptive filter.

[0011] According to another aspect of an embodiment of the present invention, a noise reduction device on a wearable device is also provided, including: an acquisition unit, used to acquire a mixed voice signal received by a dual microphone array, and perform short-time Fourier transform on the mixed voice signal to obtain a time-frequency domain of first channel data and a time-frequency domain of second channel data; a first filtering unit, used to pass the time-frequency domain of the first channel data and the time-frequency domain of the second channel data through an adaptive filter to obtain a near-end estimation signal; a first estimation unit, used to pass the near-end estimation signal through a noise estimation algorithm to obtain noise energy; a first noise reduction unit, used to, when the noise energy value is greater than or equal to a first threshold, pass the near-end estimation signal through a neural network noise reduction and a post-processing algorithm to obtain a clean time-frequency domain voice signal of the wearer, wherein the wearer wears the wearable device; a first transformation unit, used to perform short-time Fourier inverse transform on the clean time-frequency domain voice signal to obtain the wearer's voice time domain signal.

[0012] Optionally, the device also includes: a second estimation unit, used to obtain the angle information of each frame by azimuth estimation of the near-end estimation signal when the noise energy value is less than the first threshold; a third estimation unit, used to estimate the first speech energy corresponding to the first channel data, and the second speech energy corresponding to the second channel data; an updating unit, used to send the first speech energy and the second channel speech energy into an adaptive filter, and update the adaptive filter when the angle information or the first speech energy and the second speech energy meet a preset condition; a second noise reduction unit, used to reduce the noise of the speech signal filtered by the adaptive filter through the neural network to obtain the clean speech signal of the wearer; a second transformation unit, used to perform short-time Fourier inverse transform on the clean time-frequency domain speech signal to obtain the wearer's speech time domain signal.

[0013] Optionally, the updating unit includes: an updating module, used to update the adaptive filter when the angle information is greater than the preset angle information, or the first speech energy is less than the second speech energy, or the difference between the first speech energy and the second speech energy is less than a preset speech energy value.

[0014] Optionally, the device further includes: a non-updating unit, configured to not update the adaptive filter when the angle information or the first speech energy and the second speech energy do not satisfy a preset condition.

[0015] According to a first aspect of an embodiment of the present application, a computer-readable storage medium is provided, characterized in that a computer program is stored in the storage medium, wherein the computer program is configured to execute the above-mentioned noise reduction method on the wearable device when running.

[0016] According to a first aspect of an embodiment of the present application, an electronic device is provided, comprising a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to run the computer program to execute the noise reduction method on the wearable device.

[0017] In an embodiment of the present invention, a mixed speech signal received by a dual microphone array is obtained, and the mixed speech signal is short-time Fourier transformed to obtain the time-frequency domain of the first channel data and the time-frequency domain of the second channel data; the time-frequency domain of the first channel data and the time-frequency domain of the second channel data are passed through an adaptive filter to obtain a near-end estimation signal; the near-end estimation signal is passed through a noise estimation algorithm to obtain noise energy; when the noise energy value is greater than or equal to a first threshold, the near-end estimation signal is passed through a neural network noise reduction and a post-processing algorithm to obtain a clean time-frequency domain speech signal of the wearer, wherein the wearer wears a wearable device; the clean time-frequency domain speech signal is short-time Fourier inverse transformed to obtain the wearer's speech time domain signal. That is, in the present invention, the dual microphone array, in a strong noise environment, adopts a neural network noise reduction and post-processing solution to make up for the shortcomings of the traditional signal processing algorithm; in a living environment, it is comprehensively determined whether the orientation information and transient energy meet the update conditions of the adaptive filter, and then the wearer's clean speech signal is obtained by combining the neural network noise reduction algorithm, which largely solves the troubles caused by mixed noise in complex life scenes, and thus solves the technical problem of low noise reduction accuracy of wearable devices in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0019] Figure 1 is a hardware structure block diagram of a mobile terminal of an optional noise reduction method on a wearable device according to an embodiment of the present invention;

[0020] Figure 2 is a flow chart of an optional noise reduction method on a wearable device according to an embodiment of the present invention;

[0021] Figure 3 is a schematic diagram of an optional noise reduction system on a wearable device according to an embodiment of the present invention;

[0022] Figure 4 is a schematic diagram of an optional noise reduction method on a wearable device according to an embodiment of the present invention;

[0023] Figure 5 is a diagram of an optional noise reduction device on a wearable device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0024] 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.

[0025] 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 sequence 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.

[0026] The noise reduction method embodiment on the wearable device provided in the embodiment of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Taking running on a mobile terminal as an example, Figure 1 1 is a hardware structure block diagram of a mobile terminal of a noise reduction method on a wearable device according to an embodiment of the present invention. Figure 1 As shown, the mobile terminal 10 may include one or more ( Figure 1 Only one is shown in the figure) a processor 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data. Optionally, the mobile terminal may also include a transmission device 106 and an input / output device 108 for communication functions. It can be understood by those skilled in the art that Figure 1 The structure shown is for illustration only and does not limit the structure of the mobile terminal. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.

[0027] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the noise reduction method on the wearable device in the embodiment of the present invention. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, to implement the above method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories can be connected to the mobile terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0028] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the mobile terminal 10. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0029] In this embodiment, a noise reduction method on a wearable device is also provided. Figure 2 is a flow chart of a noise reduction method on a wearable device according to an embodiment of the present invention. Figure 2 As shown, the noise reduction method on the wearable device includes the following steps:

[0030] Step S202, obtaining a mixed speech signal received by the dual microphone array, and performing short-time Fourier transform on the mixed speech signal to obtain time-frequency domain of first channel data and time-frequency domain of second channel data.

[0031] Step S204: Pass the time-frequency domain of the first channel data and the time-frequency domain of the second channel data through an adaptive filter to obtain a near-end estimation signal.

[0032] Step S206: The near-end estimated signal is passed through a noise estimation algorithm to obtain noise energy.

[0033] Step S208, when the noise energy value is greater than or equal to the first threshold, the near-end estimated signal is subjected to neural network noise reduction and post-processing algorithm to obtain a clean time-frequency domain speech signal of the wearer, wherein the wearer wears a wearable device.

[0034] Step S210, performing short-time inverse Fourier transform on the clean time-frequency domain speech signal to obtain the wearer's speech time-domain signal.

[0035] In this embodiment, the noise reduction method on the wearable device can be used to obtain the wearer's speech recognition result based on the processed speech time domain signal. That is, based on the noise-reduced speech signal, the semantics of the wearer can be recognized. It can include but is not limited to a dual microphone array, a plurality of microphone arrays greater than two microphones.

[0036] The distance between the above dual microphone arrays is D. Voice information from different directions can be obtained. That is, each microphone corresponds to a channel of voice data. Multiple microphone arrays greater than two microphones can be arranged equidistantly or randomly, and the wearer's voice information can be obtained from more directions.

[0037] The wearer's voice information obtained through the microphone array usually carries a lot of noise. The noise reduction method adopted in this embodiment can achieve efficient noise reduction.

[0038] In this embodiment, the above-mentioned wearable devices may include but are not limited to smart watches, smart bracelets, smart glasses, VR headsets, smart helmets, and smart sneakers.

[0039] In this embodiment, when the noise energy value is greater than or equal to the first threshold, it is judged as a strong noise environment, and the near-end estimated signal is subjected to neural network noise reduction and post-processing algorithm to obtain a clean time-frequency domain voice signal of the wearer, wherein the wearer wears a wearable device.

[0040] Among them, the execution entity of the noise reduction method on the above-mentioned wearable device may include but is not limited to a smart device, a server, a terminal device, etc.

[0041] Through the embodiment provided by the present application, the mixed voice signal received by the dual microphone array is obtained, and the mixed voice signal is short-time Fourier transformed to obtain the time-frequency domain of the first channel data and the time-frequency domain of the second channel data; the time-frequency domain of the first channel data and the time-frequency domain of the second channel data are passed through an adaptive filter to obtain a near-end estimation signal; the near-end estimation signal is passed through a noise estimation algorithm to obtain noise energy; when the noise energy value is greater than or equal to the first threshold, the near-end estimation signal is passed through a neural network noise reduction and a post-processing algorithm to obtain a clean time-frequency domain voice signal of the wearer, wherein the wearer wears a wearable device; the clean time-frequency domain voice signal is short-time Fourier inverse transformed to obtain the wearer's voice time domain signal. That is, in the present invention, the dual microphone array, in a strong noise environment, adopts a neural network noise reduction and post-processing solution to make up for the shortcomings of the traditional signal processing algorithm; in a living environment, it is comprehensively determined whether the azimuth information and transient energy meet the update conditions of the adaptive filter, and then the wearer's clean voice signal is obtained by combining the neural network noise reduction algorithm, which largely solves the troubles caused by mixed noise in complex life scenes, and thus solves the technical problem of low noise reduction accuracy of wearable devices in the prior art.

[0042] Optionally, the method may also include: when the noise energy value is less than a first threshold, obtaining angle information of each frame by azimuth estimation of the near-end estimation signal; estimating the first speech energy corresponding to the first channel data, and the second speech energy corresponding to the second channel data; sending the first speech energy and the second channel speech energy to the adaptive filter, and updating the adaptive filter when the angle information or the first speech energy and the second speech energy meet preset conditions; denoising the speech signal filtered by the adaptive filter through the neural network to obtain a clean speech signal of the wearer; performing short-time Fourier inverse transform on the clean time-frequency domain speech signal to obtain the wearer's speech time domain signal.

[0043] In this embodiment, when the noise energy value is less than the first threshold, it is determined that the wearer is in a living environment, and noise reduction is then performed in a corresponding manner.

[0044] Optionally, updating the adaptive filter when the angle information or the first speech energy and the second speech energy meet a preset condition may include: updating the adaptive filter when the angle information is greater than preset angle information, or the first speech energy is less than the second speech energy, or the difference between the first speech energy and the second speech energy is less than a preset speech energy value.

[0045] Optionally, the method may further include: if the angle information or the first speech energy and the second speech energy do not satisfy a preset condition, not updating the adaptive filter.

[0046] As an optional embodiment, the present application also provides a noise reduction method on a wearable device. Figure 3 As shown, a schematic diagram of the noise reduction system based on a wearable device.

[0047] like Figure 3 As shown, the noise reduction system on the wearable device includes a front-end signal processing module, a speech recognition module, and semantic recognition (the human ear obtains the semantics). The speech collected by the microphone is processed by the front-end signal processing algorithm, and after obtaining the speech signal of the target speaker, it is sent to the speech recognition module or the human ear to improve the speech quality.

[0048] Among them, the front-end signal processing part assumes that the target sound source A is located at a fixed position of the wearable device, and its azimuth angle is φ (such as 60 degrees). The present invention first performs echo elimination on the voice signal received by the microphone to obtain a near-end estimation signal, and then uses noise tracking to determine the usage scenario, and then automatically switches the processing mode. There are two processing methods: 1. Perform neural network noise reduction on the voice signal received by the microphone, and then use post-processing to obtain a clean voice signal; 2. First perform azimuth estimation and transient energy detection on the voice signal to control the adaptive filtering update logic, and then obtain the voice signal of the target speaker by combining the neural network noise reduction algorithm module.

[0049] The front-end signal processing module in the noise reduction system on the wearable device may include but is not limited to the above-mentioned noise reduction method on the wearable device.

[0050] like Figure 4 As shown, a schematic diagram of a noise reduction method on a wearable device.

[0051] Step 1: Set the threshold of noise estimation to The threshold for transient energy estimation is σ1;

[0052] Step 2, performing short-time Fourier transform on the mixed speech signals x1(t) and x2(t) received by the microphone to obtain the time-frequency domain form x1(t, k) of the first channel data and the time-frequency domain form x2(t, k) of the second channel data, where t represents the frame number and k represents the frequency;

[0053] Step 3, applying an adaptive filtering algorithm to the microphone signal to obtain a near-end estimation signal;

[0054] Step 4: The obtained near-end signal is subjected to a noise estimation algorithm to obtain the noise energy E n; If it is determined as a strong noise environment, and the first processing method is performed. Otherwise, it is determined as a living environment, and the second processing method is performed on the near-end signal;

[0055] Among them, the first processing method: passing the near-end estimated signal through a neural network noise reduction and a post-processing algorithm to obtain the clean time-frequency domain speech signal of the wearer.

[0056] The second processing method: obtaining the angle information of each frame by passing the near-end estimated signal through azimuth estimation Estimate that the speech energies of the first channel and the second channel are respectively and

[0057] Send channel 1 and channel 2 into the adaptive filtering algorithm. If or E1 < E2 or E1 - E2 < σ1, update the parameters of the adaptive filter. Otherwise, do not update the filter parameters;

[0058] Pass the speech signal filtered in the previous step through the neural network noise reduction algorithm to obtain the time-frequency domain form of the clean speech signal;

[0059] Step 5, perform the inverse short-time Fourier transform on the time-frequency domain form of the speech signal to obtain the final speech time-domain signal of the wearer.

[0060] In this embodiment, in a strong noise environment, the scheme of combining neural network noise reduction with post-processing makes up for the deficiencies of traditional signal processing algorithms; in a living environment, comprehensively determine whether the azimuth information and transient energy meet the update conditions of the adaptive filter, and then obtain the clean speech signal of the wearer by combining the neural network noise reduction algorithm, which largely solves the problem of the mixed noise in complex living scenarios.

[0061] The noise tracking algorithm is used to enable the device to automatically switch to different noise reduction algorithms in different scenarios, solving the problem that the device is limited to a single scenario.

[0062] Through the description of the above implementation manners, those skilled in the art can clearly understand that the method according to the above embodiments can be implemented by means of software plus a necessary general hardware platform. Of course, it can also be implemented by hardware, but in many cases, the former is a better implementation manner. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disc), and includes several instructions for causing a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in various embodiments of the present invention.

[0063] In this embodiment, a noise reduction device on a wearable device is also provided, which is used to implement the above-mentioned embodiments and preferred embodiments, and will not be repeated here. As used below, the term "module" can implement a combination of software and / or hardware of a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0064] Figure 5 is a structural block diagram of a noise reduction device on a wearable device according to an embodiment of the present invention, such as Figure 5 As shown, the noise reduction device on the wearable device includes:

[0065] The acquisition unit 51 is used to acquire the mixed speech signal received by the dual microphone array, and perform short-time Fourier transform on the mixed speech signal to obtain the time-frequency domain of the first channel data and the time-frequency domain of the second channel data.

[0066] The first filtering unit 53 is used for passing the time-frequency domain of the first channel data and the time-frequency domain of the second channel data through an adaptive filter to obtain a near-end estimation signal.

[0067] The first estimation unit 55 is used to obtain noise energy by subjecting the near-end estimation signal to a noise estimation algorithm.

[0068] The first noise reduction unit 57 is used to reduce noise of the near-end estimation signal through a neural network and a post-processing algorithm when the noise energy value is greater than or equal to a first threshold, so as to obtain a clean time-frequency domain speech signal of the wearer, wherein the wearer wears a wearable device.

[0069] The first transform unit 59 is used to perform short-time Fourier inverse transform on the clean time-frequency domain speech signal to obtain the wearer's speech time-domain signal.

[0070] Through the embodiments provided by the present application, the acquisition unit 51 acquires the mixed voice signal received by the dual microphone array, and performs short-time Fourier transform on the mixed voice signal to obtain the time-frequency domain of the first channel data and the time-frequency domain of the second channel data; the first filtering unit 53 passes the time-frequency domain of the first channel data and the time-frequency domain of the second channel data through an adaptive filter to obtain a near-end estimation signal; the first estimation unit 55 passes the near-end estimation signal through a noise estimation algorithm to obtain noise energy; the first noise reduction unit 57 passes the near-end estimation signal through a neural network noise reduction and a post-processing algorithm when the noise energy value is greater than or equal to the first threshold to obtain a clean time-frequency domain voice signal of the wearer, wherein the wearer wears a wearable device; the first transformation unit 59 performs short-time inverse Fourier transform on the clean time-frequency domain voice signal to obtain the wearer's voice time domain signal. That is, in the present invention, the dual-microphone array adopts a neural network noise reduction combined with post-processing in a strong noise environment, which makes up for the shortcomings of the traditional signal processing algorithm; in a living environment, it comprehensively determines whether the orientation information and transient energy meet the updating conditions of the adaptive filtering, and then obtains the wearer's clean voice signal by combining the neural network noise reduction algorithm, which largely solves the problems caused by mixed noise in complex life scenes, and further solves the technical problem of low noise reduction accuracy of wearable devices in the prior art.

[0071] Optionally, the device may also include: a second estimation unit, used to obtain the angle information of each frame by azimuth estimation of the near-end estimation signal when the noise energy value is less than the first threshold; a third estimation unit, used to estimate the first speech energy corresponding to the first channel data, and the second speech energy corresponding to the second channel data; an updating unit, used to send the first speech energy and the second channel speech energy into an adaptive filter, and update the adaptive filter when the angle information or the first speech energy and the second speech energy meet a preset condition; a second noise reduction unit, used to reduce the noise of the speech signal filtered by the adaptive filter through the neural network to obtain the clean speech signal of the wearer; a second transformation unit, used to perform short-time Fourier inverse transform on the clean time-frequency domain speech signal to obtain the wearer's speech time domain signal.

[0072] Optionally, the update unit may include: an update module, used to update the adaptive filter when the angle information is greater than the preset angle information, or the first speech energy is less than the second speech energy, or the difference between the first speech energy and the second speech energy is less than a preset speech energy value.

[0073] Optionally, the device may further include: a non-updating unit, configured to not update the adaptive filter when the angle information or the first speech energy and the second speech energy do not satisfy a preset condition.

[0074] It should be noted that the above modules can be implemented by software or hardware. For the latter, it can be implemented in the following ways, but not limited to: the above modules are all located in the same processor; or the above modules are located in different processors in any combination.

[0075] An embodiment of the present invention further provides a storage medium, in which a computer program is stored, wherein the computer program is configured to execute the steps of any of the above method embodiments when running.

[0076] Optionally, in this embodiment, the storage medium may be configured to store a computer program for performing the following steps:

[0077] S1, obtaining a mixed speech signal received by a dual microphone array, and performing short-time Fourier transform on the mixed speech signal to obtain a time-frequency domain of first channel data and a time-frequency domain of second channel data;

[0078] S2, passing the time-frequency domain of the first channel data and the time-frequency domain of the second channel data through an adaptive filter to obtain a near-end estimation signal;

[0079] S3, passing the near-end estimated signal through a noise estimation algorithm to obtain noise energy;

[0080] S4, when the noise energy value is greater than or equal to the first threshold, performing denoising and post-processing on the near-end estimated signal through a neural network to obtain a clean time-frequency domain speech signal of the wearer, wherein the wearer wears a wearable device;

[0081] S5, performing short-time inverse Fourier transform on the clean time-frequency domain speech signal to obtain the wearer's speech time domain signal.

[0082] Optionally, in this embodiment, the above-mentioned storage medium may include but is not limited to: a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk, and other media that can store computer programs.

[0083] An embodiment of the present invention further provides an electronic device, including a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0084] Optionally, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0085] Optionally, in this embodiment, the processor may be configured to perform the following steps through a computer program:

[0086] S1, obtaining a mixed speech signal received by a dual microphone array, and performing short-time Fourier transform on the mixed speech signal to obtain a time-frequency domain of first channel data and a time-frequency domain of second channel data;

[0087] S2, passing the time-frequency domain of the first channel data and the time-frequency domain of the second channel data through an adaptive filter to obtain a near-end estimation signal;

[0088] S3, passing the near-end estimated signal through a noise estimation algorithm to obtain noise energy;

[0089] S4, when the noise energy value is greater than or equal to the first threshold, performing denoising and post-processing on the near-end estimated signal through a neural network to obtain a clean time-frequency domain speech signal of the wearer, wherein the wearer wears a wearable device;

[0090] S5, performing short-time inverse Fourier transform on the clean time-frequency domain speech signal to obtain the wearer's speech time domain signal.

[0091] Optionally, the specific examples in this embodiment may refer to the examples described in the above embodiments and optional implementation modes, and this embodiment will not be described in detail here.

[0092] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, and optionally, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order than here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0093] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A noise reduction method on a wearable device, characterized in that: include: Acquire a mixed speech signal received by a dual microphone array, and perform short-time Fourier transform on the mixed speech signal to obtain a time-frequency domain of first channel data and a time-frequency domain of second channel data; Passing the time-frequency domain of the first channel data and the time-frequency domain of the second channel data through an adaptive filter to obtain a near-end estimation signal; Passing the near-end estimation signal through a noise estimation algorithm to obtain noise energy; When the noise energy value is greater than or equal to a first threshold, the proximal estimation signal is subjected to a neural network denoising and a post-processing algorithm to obtain a clean time-frequency domain speech signal of the wearer, wherein the wearer wears the wearable device; Performing a short-time inverse Fourier transform on the clean time-frequency domain speech signal to obtain the wearer's speech time-domain signal; The method further comprises: When the noise energy value is less than a first threshold, obtaining angle information of each frame by using the near-end estimation signal through azimuth estimation; estimating a first speech energy corresponding to the first channel data and a second speech energy corresponding to the second channel data; Sending the first speech energy and the second speech energy to the adaptive filter, and updating the adaptive filter when the angle information or the first speech energy and the second speech energy meet a preset condition; De-noising the speech signal filtered by the adaptive filter through the neural network to obtain a clean speech signal of the wearer; The clean time-frequency domain speech signal is subjected to short-time inverse Fourier transform to obtain the wearer's speech time-domain signal.

2. The method according to claim 1, characterized in that When the angle information or the first speech energy and the second speech energy meet a preset condition, updating the adaptive filter includes: When the angle information is greater than the preset angle information, or the first speech energy is less than the second speech energy, or the difference between the first speech energy and the second speech energy is less than the preset speech energy value, the adaptive filter is updated.

3. The method according to claim 1, characterized in that The method further comprises: When the angle information or the first speech energy and the second speech energy do not meet a preset condition, the adaptive filter is not updated.

4. A noise reduction device on a wearable device, characterized in that: include: An acquisition unit, used for acquiring a mixed speech signal received by the dual microphone array, and performing a short-time Fourier transform on the mixed speech signal to obtain a time-frequency domain of first channel data and a time-frequency domain of second channel data; A first filtering unit, configured to pass the time-frequency domain of the first channel data and the time-frequency domain of the second channel data through an adaptive filter to obtain a near-end estimation signal; A first estimation unit, configured to obtain noise energy by applying a noise estimation algorithm to the near-end estimation signal; a first noise reduction unit, configured to, when the noise energy value is greater than or equal to a first threshold, perform noise reduction on the proximal estimation signal through a neural network and a post-processing algorithm to obtain a clean time-frequency domain speech signal of a wearer, wherein the wearer wears the wearable device; A first transform unit, configured to perform a short-time inverse Fourier transform on the clean time-frequency domain speech signal to obtain a time-domain signal of the wearer's speech; A second estimation unit, configured to obtain angle information of each frame by performing azimuth estimation on the near-end estimation signal when the noise energy value is less than a first threshold; A third estimation unit, configured to estimate a first speech energy corresponding to the first channel data and a second speech energy corresponding to the second channel data; An updating unit, configured to send the first speech energy and the second speech energy to an adaptive filter, and update the adaptive filter when the angle information or the first speech energy and the second speech energy meet a preset condition; A second noise reduction unit, configured to reduce the noise of the speech signal filtered by the adaptive filter through the neural network to obtain a clean speech signal of the wearer; The second transformation unit is used to perform short-time Fourier inverse transformation on the clean time-frequency domain speech signal to obtain the wearer's speech time-domain signal.

5. The device according to claim 4, characterized in that The updating unit comprises: An updating module is used to update the adaptive filter when the angle information is greater than the preset angle information, or the first speech energy is less than the second speech energy, or the difference between the first speech energy and the second speech energy is less than the preset speech energy value.

6. The device according to claim 4, characterized in that The device also includes: A non-updating unit is used for not updating the adaptive filter when the angle information or the first speech energy and the second speech energy do not meet a preset condition.

7. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, wherein the computer program is configured to execute the method according to any one of claims 1 to 3 when executed.

8. An electronic device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to run the computer program to perform the method according to any one of claims 1 to 3.

Images