A method, device, system, medium and equipment for reducing speech noise

By using low-latency improved discrete cosine transform and spectral entropy calculation in Bluetooth devices, noise analysis and elimination are directly performed in the LC3 audio encoder, solving the high computing power requirements of Fourier transform and reducing power consumption and improve user experience.

CN114743559BActive Publication Date: 2024-11-08BEIJING BAIRUI INTERNET TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210579107.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-05-25
Publication Date
2024-11-08
Estimated Expiration
2042-05-25

AI Technical Summary

Technical Problem

In the prior art, Fourier transform and inverse Fourier transform have high requirements for computing power, resulting in increased power consumption of the system and reduced user experience.

Method used

Using low-latency improved discrete cosine transformation and spectral entropy calculation, noise cancellation is performed directly during the encoding and decoding of the LC3 audio encoder to avoid additional time-frequency conversion and noise analysis in the encoder at the transmitting and receiving end of Bluetooth. Frequency conversion and inverse conversion.

Benefits of technology

It saves the system's computing volume and code rate, improves coding efficiency, reduces system power consumption, and improves user experience.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114743559B_ABST
    Figure CN114743559B_ABST
Patent Text Reader

Abstract

The present application discloses a method, device, system, medium and equipment for speech noise reduction, which belongs to the field of speech coding and decoding technology. The method mainly includes: framing the speech information collected by the Bluetooth transmitter, performing low-delay improved discrete cosine transform on the single-frame speech information to obtain the single-frame spectrum coefficient; calculating the spectral entropy of the pseudo-spectrum corresponding to the single-frame speech information according to the single-frame spectrum coefficient; determining the noise in the speech information according to the spectral entropy and the preset threshold value; performing noise elimination on the noise, and performing the remaining encoding steps to obtain and output the bit stream; using the decoder of the Bluetooth receiver itself to decode the bit stream and output the noise-reduced speech information. The spectral coefficients are obtained by using the time-frequency conversion module of the audio encoder itself, and the spectral coefficients are subjected to noise analysis and noise elimination in the frequency domain to achieve noise reduction. By performing noise reduction in the encoding and decoding process, the effects of saving bit rate, reducing the amount of system calculations, and accelerating encoding efficiency are achieved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of speech encoding and decoding technology, and in particular to a speech noise reduction method, apparatus, system, medium and device. Background Technology

[0002] The mainstream Bluetooth audio encoders are as follows: SBC: Mandatory requirement of the A2DP protocol, the most widely used, and must be supported by all Bluetooth audio devices, but the sound quality is generally average; AAC-LC: Good sound quality and widely used, supported by many mainstream mobile phones, but compared to SBC, it has a larger memory footprint and higher computational complexity. Many Bluetooth devices are based on embedded platforms with limited battery capacity, poor processor processing power, and limited memory. Furthermore, its patent fees are high; aptX series: Good sound quality, but very high bitrate. aptX requires a bitrate of 384kbps, while apt... X-HD has a bitrate of 576kbps and is a Qualcomm proprietary technology, making it relatively closed. LDAC offers good sound quality but also has high bitrates of 330kbps, 660kbps, and 990kbps. Due to the complex wireless environment of Bluetooth devices, stably supporting such high bitrates is challenging, and LDAC is a Sony proprietary technology, also relatively closed. LHDC offers good sound quality but also has high bitrates, typically 400kbps, 600kbps, and 900kbps. Such high bitrates place high demands on Bluetooth baseband / RF design. For these reasons, the Bluetooth SIG, in collaboration with numerous manufacturers, launched LC3, primarily for Bluetooth Low Energy but also usable in Classic Bluetooth. It boasts advantages such as lower latency, higher sound quality and coding gain, and no patent fees in the Bluetooth field, attracting widespread attention from manufacturers.

[0003] However, noise reduction is necessary during Bluetooth voice communication to improve sound quality and user experience. In related technologies, spectral subtraction is commonly used for noise reduction, and its specific implementation scheme is as follows: Figure 1 As shown, Fourier transform is used to perform time-frequency transformation on the speech information, converting the speech information in the time domain into frequency domain data. Noise is estimated and eliminated in the frequency domain, and then the inverse Fourier transform is used to convert it back to the time domain to obtain the noise-reduced speech information. The noise-reduced speech information is transmitted between the Bluetooth transmitter and the Bluetooth receiver.

[0004] However, Figure 1 The Fourier transform and inverse Fourier transform in the technical solution have high requirements for computing power, and place high demands on the system's storage space and computing power. This solution increases the system's power consumption and reduces the user experience. Summary of the Invention

[0005] In view of the fact that existing technologies require high computing power for Fourier transform and inverse Fourier transform, which places high demands on system storage space and computing power, resulting in increased power consumption and reduced user experience, this application mainly provides a speech noise reduction method, device, system, medium and equipment.

[0006] To achieve the above objectives, this application adopts the following technical solution: providing a speech denoising method, comprising: segmenting the speech information acquired by the Bluetooth transmitter into frames according to the encoder of the Bluetooth transmitter itself, and performing low-latency improved discrete cosine transform on each of the obtained single-frame speech information to obtain the single-frame spectral coefficients in the frequency domain corresponding to the single-frame speech information; calculating the spectral entropy of the pseudo-spectrum corresponding to the single-frame speech information based on the single-frame spectral coefficients; determining the noise in the speech information based on the spectral entropy and a preset threshold value; performing noise cancellation on the noise-cancelled denoised spectral coefficients, and performing the remaining encoding steps on the noise-cancelled denoised spectral coefficients to obtain the bitstream corresponding to the noise-cancelled spectral coefficients and outputting it; receiving the bitstream through a Bluetooth receiver connected to the Bluetooth transmitter via Bluetooth, and decoding the bitstream using the decoder of the Bluetooth receiver itself until the inverse discrete cosine transform is output to denoise-reduced speech information.

[0007] Another technical solution adopted in this application is: providing a speech noise reduction device, comprising: a time-frequency transformation module, which is used to segment the speech information collected by the Bluetooth transmitter according to the encoder of the Bluetooth transmitter itself, and perform low-latency improved discrete cosine transform on the obtained single-frame speech information to obtain the single-frame spectral coefficients in the frequency domain corresponding to the single-frame speech information; a spectral entropy calculation module, which is used to calculate the spectral entropy of the pseudo-spectrum corresponding to the single-frame speech information according to the single-frame spectral coefficients; a noise determination module, which is used to determine the noise in the speech information according to the spectral entropy and a preset threshold value; and a noise cancellation module, which is used to cancel the noise, and perform the remaining encoding steps on the noise-cancelled speech information to obtain the code stream corresponding to the noise-cancelled speech information and output it; and a time-frequency inverse transformation module, which is used to receive the code stream from the Bluetooth receiver connected to the Bluetooth transmitter via Bluetooth, and use the decoder of the Bluetooth receiver itself to decode the code stream until the inverse discrete cosine transform is output to denoise-cancelled speech information.

[0008] Another technical solution adopted in this application is: providing a speech noise reduction system, which includes the speech noise reduction device described in Solution 2, wherein the speech noise reduction device includes a time-frequency transformation module, which is used to segment the speech information collected by the Bluetooth transmitter according to the encoder of the Bluetooth transmitter itself, and perform low-latency improved discrete cosine transform on the obtained single-frame speech information to obtain the single-frame spectral coefficients in the frequency domain corresponding to the single-frame speech information; a spectral entropy calculation module, which is used to calculate the spectral entropy of the pseudo-spectrum corresponding to the single-frame speech information according to the single-frame spectral coefficients; a noise determination module, which is used to determine the noise in the speech information according to the spectral entropy and a preset threshold value; and a noise cancellation module, which is used to cancel the noise, and perform the remaining encoding steps on the noise-cancelled denoised speech information to obtain the code stream corresponding to the denoised speech information and output it; and a time-frequency inverse transformation module, which is used to receive the code stream from the Bluetooth receiver connected to the Bluetooth transmitter via Bluetooth, and use the decoder of the Bluetooth receiver itself to decode the code stream until the discrete cosine transform is output to denoise the speech information.

[0009] Another technical solution adopted in this application is to provide a computer-readable storage medium storing computer instructions that are operated to perform the speech noise reduction method in Solution 1.

[0010] Another technical solution adopted in this application is: providing a computer device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores computer instructions executable by the at least one processor, and the at least one processor operates the computer instructions to execute the speech noise reduction method in Solution 1.

[0011] The beneficial effects that the technical solution of this application can achieve are as follows: Since time-frequency transformation and inverse time-frequency transformation already exist in the Bluetooth encoding and decoding process, it is not necessary to perform additional time-frequency transformation and inverse time-frequency transformation separately in order to perform noise reduction processing in the frequency domain. Instead, the spectral coefficients can be obtained directly by using the audio encoder's own time-frequency conversion module during the encoding and decoding process. Noise reduction can be achieved by performing noise analysis and noise cancellation on the spectral coefficients in the frequency domain. By performing noise reduction during the encoding and decoding process, the bit rate can be saved, the system's computational load can be reduced, and the encoding efficiency can be accelerated. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1This is a flowchart of an optional example of a speech noise reduction method according to this application;

[0014] Figure 2 This is a schematic diagram of an optional implementation of a speech noise reduction method according to this application;

[0015] Figure 3 This is a schematic diagram of an optional embodiment of a speech noise reduction device according to this application.

[0016] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0017] The preferred embodiments of this application will now be described in detail with reference to the accompanying drawings, so that the advantages and features of this application can be more easily understood by those skilled in the art, thereby providing a clearer and more definite definition of the scope of protection of this application.

[0018] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0019] During Bluetooth voice communication, noise reduction is necessary to improve sound quality and user experience. Spectral subtraction is commonly used for noise reduction in voice communication, and its specific implementation scheme is as follows: Figure 1 As shown, Fourier transform is used to perform time-frequency transformation on the speech information, converting the time-domain speech information into frequency-domain data. Noise estimation and removal are then performed in the frequency domain, and finally, an inverse Fourier transform is used to return to the time domain to obtain the denoised speech information. The denoised speech information transmitted between the Bluetooth transmitter and receiver is the denoised speech information. However, Figure 1 The Fourier transform and inverse Fourier transform in the technical solution have high requirements for computing power, and place high demands on the system's storage space and computing power. This solution increases the system's power consumption and reduces the user experience.

[0020] To address the problems existing in the prior art, this application mainly provides a speech noise reduction method, apparatus, system, medium, and device. The speech noise reduction method includes: segmenting the speech information acquired by the Bluetooth transmitter into frames using its own encoder; performing a low-latency improved discrete cosine transform on each frame of speech information to obtain the single-frame spectral coefficients in the corresponding frequency domain; calculating the spectral entropy of the pseudo-spectrum corresponding to the single-frame speech information based on the single-frame spectral coefficients; determining the noise in the speech information based on the spectral entropy and a preset threshold value; performing noise removal on the noise-reduced spectral coefficients; performing the remaining encoding steps on the noise-reduced spectral coefficients to obtain the corresponding bitstream and outputting it; receiving the bitstream through a Bluetooth receiver connected to the Bluetooth transmitter via Bluetooth, and decoding the bitstream using the Bluetooth receiver's own decoder until an inverse discrete cosine transform is output to produce noise-reduced speech information.

[0021] By using the time-frequency conversion module of the LC3 audio encoder to obtain spectral coefficients during the encoding and decoding process, and then performing noise analysis and noise cancellation on the spectral coefficients in the frequency domain, noise reduction can be achieved. By performing noise reduction during the encoding and decoding process, the bit rate can be saved, the amount of computation in the system can be reduced, and the encoding efficiency can be accelerated.

[0022] The technical solutions of this application and how they solve the aforementioned technical problems will be described in detail below with specific embodiments. The specific embodiments described below can be combined with each other to form new embodiments. The same or similar ideas or processes described in one embodiment may not be repeated in other embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0023] Figure 2 An alternative implementation of a speech noise reduction method according to this application is shown.

[0024] exist Figure 2 In the optional implementation shown, the speech noise reduction method mainly includes step S201, which involves dividing the speech information collected by the Bluetooth transmitter into frames according to the encoder of the Bluetooth transmitter itself, and performing low-latency improved discrete cosine transform on the obtained single-frame speech information to obtain the single-frame spectral coefficients in the frequency domain corresponding to the single-frame speech information.

[0025] In this optional implementation, during Bluetooth voice communication, the encoder of the Bluetooth transmitter segments the received voice information into multiple single-frame voice information, facilitating subsequent judgment of each single-frame voice information to identify noise frames. After acquiring the single-frame voice data, the low-latency improved discrete cosine transform module in the encoder of the Bluetooth transmitter directly performs time-frequency conversion on the single-frame voice information, obtaining the single-frame spectral coefficients in the frequency domain corresponding to each single-frame voice information. This provides the necessary foundation for subsequent steps such as calculating the spectral entropy of the pseudo-spectrum based on the spectral coefficients in the frequency domain. Furthermore, performing time-frequency conversion in the encoder allows noise reduction to be achieved with only one time-frequency conversion, greatly saving the system's computational load and accelerating the system's processing speed.

[0026] In an optional embodiment of this application, the speech information is framed according to the LC3 audio coding standard to obtain multiple single-frame audio information corresponding to the speech information; and the single-frame audio information is converted from the time domain to the frequency domain to obtain the single-frame spectral coefficients corresponding to the single-frame audio information. The calculation formula for the time domain to frequency domain conversion is as follows:

[0027] t(n) = x s (ZN F +n), for n=0…2·N F -1-Z

[0028] t(2N F -Z+n)=0, for n=0…Z-1

[0029]

[0030] N in the above formula F x represents the number of spectral coefficients in a single frame. s (n) represents the speech information of a single frame, and X(k) represents the spectral coefficients of a single frame.

[0031] exist Figure 2 In the optional embodiment shown, the speech denoising method further includes step S202, which calculates the spectral entropy of the pseudo-spectrum corresponding to the single-frame speech information based on the single-frame spectral coefficients.

[0032] In this optional implementation, since noise estimation and cancellation are performed using Fourier Transform (FFT) in the prior art, there is a relatively accurate correspondence between the spectral coefficients obtained by Fourier Transform (FFT) and the frequency of the speech information in the time domain. The spectral entropy calculated from these spectral coefficients can effectively distinguish between speech and noise. However, for spectral coefficients obtained using low-delay improved discrete cosine transform, there is a deviation between the spectral coefficients obtained by this method and the frequency of the speech information in the time domain. The spectral entropy calculated from these spectral coefficients cannot clearly distinguish between speech and noise. In order to avoid the above deviation, this scheme adds a pseudo-spectrum. By calculating the spectral entropy of the pseudo-spectrum, the above deviation is avoided, and the accuracy of noise judgment is guaranteed.

[0033] In an optional embodiment of this application, calculating the spectral entropy of the pseudospectrum corresponding to a single frame of speech information based on the single frame spectral coefficients further includes: calculating the pseudospectrum based on the single frame spectral coefficients and obtaining the pseudospectral coefficients corresponding to the pseudospectrum; dividing the pseudospectral coefficients into pseudospectral sub-bands and obtaining the pseudospectral coefficients of the pseudospectral sub-bands corresponding to multiple pseudospectral sub-bands respectively; calculating the pseudospectral energy of the pseudospectral sub-bands based on the pseudospectral coefficients of the pseudospectral sub-bands respectively, and then obtaining the total energy of the multiple pseudospectral sub-bands; and calculating the spectral entropy based on the pseudospectral energy and the total energy.

[0034] In this optional embodiment, the pseudo-spectrum corresponding to the single-frame spectral coefficient is calculated based on the single-frame spectral coefficient, and the pseudo-spectral coefficient corresponding to the pseudo-spectral coefficient is obtained; the pseudo-spectral coefficient is divided into pseudo-spectral sub-bands according to a preset pseudo-spectral sub-band division standard, and the pseudo-spectral coefficients corresponding to multiple pseudo-spectral sub-bands are obtained respectively, wherein the preset pseudo-spectral sub-band division standard includes but is not limited to average division; the pseudo-spectral energy of each pseudo-spectral sub-band and the total energy of multiple pseudo-spectral sub-bands are calculated based on the pseudo-spectral coefficients of the pseudo-spectral sub-bands, and the energy probability corresponding to each pseudo-spectral sub-band is calculated using the pseudo-spectral energy of each pseudo-spectral sub-band and the total energy of multiple pseudo-spectral sub-bands; the spectral entropy of the pseudo-spectrum corresponding to the single-frame spectral coefficient is calculated based on the energy probability corresponding to each pseudo-spectral sub-band, providing a necessary basis for ensuring the subsequent distinction between speech and noise.

[0035] In an optional example of this application, when the sampling rate of the speech information is 16kHz and the frame length is 10ms, the pseudospectrum corresponding to the spectral coefficients of a single frame is calculated using the following formula:

[0036]

[0037] Where X(k) = 0, when k = -1 or N F When, N in the above formula F X(k) represents the number of spectral coefficients in a single frame. pseudo (k) represents the pseudo-spectral coefficients of the pseudo-spectrum corresponding to the single-frame spectral coefficients. The spectral coefficients of the pseudo-spectrums calculated using the above formula are X... pseudo (0), Xpseudo (1), ...X pseudo (159); The spectral coefficients of the pseudospectrum are divided into 16 pseudospectral subbands, each with 10 pseudospectral coefficients. The pseudospectral coefficients of the first pseudospectral subband are as follows: X pseudo (0), X pseudo (1), ...X pseudo (9); The pseudo-spectral coefficients of the second pseudo-spectral subband are respectively: X pseudo (10), X pseudo (11), ...X pssudo (19); The pseudo-spectral coefficients of the 16th pseudo-spectral subband are: X pseudo (150), X pseudo (151), ...X pseudo (159).

[0038] Based on the pseudospectral coefficients of the pseudospectral subbands obtained above, the pseudospectral energy corresponding to each pseudospectral subband is calculated as follows:

[0039]

[0040] In the above formula, SUBBAND_ENERYG(sb) is the pseudospectral energy of a single pseudospectral subband, SUBBAND_NUM is the number of pseudospectral subbands (16 in this example), SUBBAND_WIDTH is the width of the pseudospectral subband (10 in this example), and X... pseudo (k) represents the pseudo-spectral coefficients corresponding to the pseudo-spectral coefficients of a single frame.

[0041] The total energy of the pseudo-spectral subbands corresponding to the pseudo-spectral spectrum is calculated based on the pseudo-spectral energy of each pseudo-spectral subband obtained from the above formula, as follows:

[0042]

[0043] In the above formula, SUBBAND_ENERYG(k) is the pseudo-spectral energy of a single pseudo-spectral subband, SUBBAND_NUM is the number of pseudo-spectral subbands (16 in this example), SUBBAND_WIDTH is the width of the pseudo-spectral subband (10 in this example), and SUBBAND_ENERYG_TOTAL is the total energy of the multiple pseudo-spectral subbands corresponding to this pseudo-spectrum.

[0044] Based on the pseudo-spectral energy of the first pseudo-spectral subband obtained from the above formula and the total energy of multiple pseudo-spectral subbands corresponding to that pseudo-spectral subband, the energy probability of each pseudo-spectral subband is calculated as follows:

[0045] SUBBAND_ENERYG_PROB(k)

[0046] =SUBBAND_ENERYG(k) / SUBBAND_ENERYG_TOTAL

[0047] for k=1…SUBBAND_NUM

[0048] In the above formula, SUBBAND_ENERYG(k) is the pseudo-spectral energy of a single pseudo-spectral subband, SUBBAND_NUM is the number of pseudo-spectral subbands (16 in this example), SUBBAND_WIDTH is the width of the pseudo-spectral subband (10 in this example), SUBBAND_ENERYG_TOTAL is the total energy of the multiple pseudo-spectral subbands corresponding to this pseudo-spectrum, and SUBBAND_ENERYG_PROB(k) is the energy probability of a single pseudo-spectral subband.

[0049] The spectral entropy of the pseudospectral corresponding to the spectral coefficients of a single frame is calculated based on the energy probability of each pseudospectral subband obtained from the above formula, as follows:

[0050]

[0051] In this example, the number of pseudo-spectral subbands is 16, SUBBAND_WIDTH is the width of the pseudo-spectral subband, which is 10 in this example, and PSEUDO_SPEC_ENTROPY is the spectral entropy of the pseudo-spectrum corresponding to the spectral coefficients of a single frame.

[0052] exist Figure 2 In the optional implementation shown, the speech noise reduction method further includes step S203, which determines the noise in the speech information based on the spectral entropy and a preset threshold value.

[0053] In this optional implementation, since the spectral entropy of speech and noise differs significantly, a preset threshold value can be set. Speech and noise can be distinguished based on the relationship between the spectral entropy and the preset threshold value, thereby identifying noise in a single frame of speech information. This ensures the accuracy of noise identification and provides the necessary foundation for subsequent noise cancellation.

[0054] In an optional embodiment of this application, determining noise in the speech information based on spectral entropy and a preset threshold further includes: setting a preset threshold; when the spectral entropy is greater than the preset threshold, determining the single frame of speech information corresponding to the pseudo-spectrum as noise.

[0055] In this optional embodiment, based on the prior analysis of noise and speech, it is known that the spectral entropy of noise is greater than that of speech. Therefore, after setting a preset threshold, the single-frame speech information corresponding to the spectral entropy greater than the preset threshold is determined as noise, so that the accuracy of the determined noise is guaranteed.

[0056] In an optional example of this application, the preset threshold value can be 0.8, and the single-frame speech corresponding to the pseudo-spectrum with a spectral entropy greater than 0.8 is determined as noise, while the single-frame speech corresponding to the pseudo-spectrum with a spectral entropy less than or equal to 0.8 is determined as speech.

[0057] In an optional embodiment of this application, when the spectral entropy is greater than a preset threshold, the method further includes: calculating the energy estimate corresponding to each pseudo-spectral subband in the noise according to a preset energy smoothing factor.

[0058] In this optional embodiment, the subband energy corresponding to each subband in the single-frame spectral coefficients of the noise is calculated according to the standard specification. Based on the subband energy and the energy smoothing factor, the energy estimate corresponding to each pseudo-spectral subband in the noise is calculated, providing a basis for the subsequent noise elimination steps.

[0059] In an optional example of this application, when the sampling rate of the speech information is 16kHz and the frame length is 10ms, the single-frame spectral coefficients are calculated. After obtaining the single-frame spectral coefficients, the standard specification calculates the sub-band energy in the single-frame spectral coefficients, and the calculation formula is as follows:

[0060]

[0061] Wherein, N in the above formula F X(k) represents the number of spectral coefficients in a single frame, and I represents the number of spectral coefficients in a single frame. fs () represents the sub-band sequence number, E B (b) represents the subband energy of a single subband.

[0062] Based on the subband energy obtained from the above formula and the preset energy smoothing factor, the energy estimate corresponding to each pseudo-spectral subband in the noise is calculated as follows:

[0063] N(k,b)=max(Nmin,β*N(k-1,b)+(1-β)*√E B (b) for b=0…N b -1

[0064] In the above formula, β is the energy smoothing factor, and its value ranges from [0, 1]; Nmin is the minimum energy, and its normalized value can be 0.05; N(k, b) is the energy estimate of the b-th subband noise in the k-th frame.

[0065] exist Figure 2 In the optional embodiment shown, the speech denoising method further includes step S204, which involves noise removal, and then performing the remaining encoding steps on the noise-removed spectral coefficients to obtain the bitstream corresponding to the noise-removed spectral coefficients and outputting it.

[0066] In this optional embodiment, the noise determined above is eliminated. After elimination, the remaining spectral coefficients corresponding to the remaining speech information, i.e. the denoised speech spectral coefficients, are then subjected to the remaining encoding steps until the code stream encapsulation step obtains the code stream corresponding to the denoised spectral coefficients, and the code stream is output to the Bluetooth transmitter; thus providing a basis for subsequent inverse time-frequency transformation.

[0067] In one optional embodiment of this application, noise cancellation further includes: updating the spectral coefficients of each single frame according to a preset pseudo-spectral subband update formula, and obtaining the single-frame updated spectral coefficients corresponding to the single-frame spectral coefficients.

[0068] In this optional embodiment, the noise elimination method in this solution is spectral subtraction, that is, the spectral coefficients of each single frame are updated using a preset pseudo-spectral subband update formula, so that the noise is suppressed and the noise reduction is achieved.

[0069] In an optional embodiment of this application, after determining the noise according to the above steps, the spectral coefficients of each single frame are updated, wherein the preset pseudo-spectral subband update formula is as follows:

[0070]

[0071] In the above formula, sign() extracts the sequence number of the corresponding spectral coefficient, μ is the noise energy adjustment factor, and N(b) is the sub-band-based noise average energy estimate. For single-frame updated spectral coefficients; where the average energy estimate deviates slightly from the actual energy estimate, the typical value of μ is [0.9, 1], and optionally, μ can be 0.94.

[0072] In one optional embodiment of this application, noise cancellation further includes: noise suppression of the single-frame spectral coefficients corresponding to the noise.

[0073] In this optional embodiment, after performing the above-mentioned noise cancellation steps, in order to avoid insufficient noise reduction effect and residual noise still existing after noise reduction, residual noise suppression is performed on the single-frame spectral coefficients after noise reduction to improve the noise reduction effect and enhance the user experience.

[0074] In an optional embodiment of this application, the updated single-frame spectral coefficients are suppressed using residual noise suppression coefficients, and the calculation formula is as follows:

[0075]

[0076] In the above formula, γ is the residual noise suppression coefficient. For updating the spectral coefficients in a single frame, preferably, γ can be 0.5.

[0077] exist Figure 2 In the optional embodiment shown, the speech noise reduction method further includes step S205, receiving the bit stream from the Bluetooth receiver connected to the Bluetooth transmitter via Bluetooth, and decoding the bit stream using the Bluetooth receiver's own decoder until the inverse discrete cosine transform is output to output noise-reduced speech information.

[0078] In this optional implementation, the Bluetooth receiver that communicates with the Bluetooth transmitter receives the bitstream and decodes the bitstream using its own decoder to obtain noise-reduced voice information, thus completing the entire noise reduction process and obtaining low-noise or even noise-free voice information, thereby improving the user experience.

[0079] Figure 3 An alternative embodiment of a speech noise reduction device according to this application is shown.

[0080] exist Figure 3 In the optional embodiment shown, the voice noise reduction device mainly includes: a time-frequency transformation module 301, which is used to frame the voice information collected by the Bluetooth transmitter according to the encoder of the Bluetooth transmitter itself, and to perform low-latency improved discrete cosine transform on the framed single-frame voice information to obtain the single-frame spectral coefficients in the frequency domain corresponding to the single-frame voice information; a spectral entropy calculation module 302, which is used to calculate the spectral entropy of the pseudo-spectrum corresponding to the single-frame voice information according to the single-frame spectral coefficients; a noise determination module 303, which is used to determine the noise in the voice information according to the spectral entropy and a preset threshold value; a noise cancellation module 304, which is used to cancel the noise, and to perform the remaining encoding steps on the noise-cancelled voice information to obtain the code stream corresponding to the noise-cancelled voice information and output it; and a time-frequency inverse transformation module 305, which is used to receive the code stream from the Bluetooth receiver connected to the Bluetooth transmitter via Bluetooth, and to decode the code stream using the decoder of the Bluetooth receiver itself until the discrete cosine transform is output to denoise-cancelled voice information.

[0081] In one optional embodiment of this application, the functional modules of the speech noise reduction device of this application may be directly in hardware, in software modules executed by a processor, or in a combination of both.

[0082] Software modules may reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in this art. An exemplary storage medium is coupled to the processor, enabling the processor to read information from and write information to the storage medium.

[0083] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic, discrete hardware components, or any combination thereof. A general-purpose processor can be a microprocessor, but alternatively, it can be any conventional processor, controller, microcontroller, or state machine. The processor can also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors incorporating a DSP core, or any other such configuration. Alternatively, the storage medium can be integrated with the processor. The processor and storage medium can reside in an ASIC. The ASIC can reside in the user terminal. Alternatively, the processor and storage medium can reside as discrete components in the user terminal.

[0084] The speech noise reduction device provided in this application can be used to perform the speech noise reduction method described in any of the above embodiments. Its implementation principle and technical effect are similar, and will not be repeated here.

[0085] In another optional embodiment of this application, a speech noise reduction system includes the speech noise reduction device described in the above scheme. The speech noise reduction device includes a time-frequency transformation module, which is used to frame the speech information collected by the Bluetooth transmitter according to the encoder of the Bluetooth transmitter itself, and to perform low-latency improved discrete cosine transform on the framed single-frame speech information to obtain the single-frame spectral coefficients in the frequency domain corresponding to the single-frame speech information; a spectral entropy calculation module, which is used to calculate the spectral entropy of the pseudo-spectrum corresponding to the single-frame speech information according to the single-frame spectral coefficients; a noise determination module, which is used to determine the noise in the speech information according to the spectral entropy and a preset threshold value; a noise cancellation module, which is used to cancel the noise, and to perform the remaining encoding steps on the noise-cancelled speech information to obtain the code stream corresponding to the noise-cancelled speech information and output it; and a time-frequency inverse transformation module, which is used to receive the code stream from the Bluetooth receiver connected to the Bluetooth transmitter via Bluetooth, and to decode the code stream using the decoder of the Bluetooth receiver itself until the inverse discrete cosine transform is output to denoise-cancelled speech information.

[0086] The speech noise reduction system provided in this application can be used to perform the speech noise reduction method described in any of the above embodiments. Its implementation principle and technical effect are similar, and will not be repeated here.

[0087] In another alternative embodiment of this application, a computer-readable storage medium is provided that stores computer instructions which are operated to perform the speech noise reduction method described in the above embodiments.

[0088] In one optional embodiment of this application, a computer device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores computer instructions executable by the at least one processor, and the at least one processor operates the computer instructions to perform the speech noise reduction method described in the above embodiments.

[0089] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0090] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0091] The above description is merely an embodiment of this application and does not limit the patent scope of this application. Any equivalent structural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A speech noise reduction method, characterized in that: include: The voice information collected by the Bluetooth transmitter is divided into frames according to the encoder of the Bluetooth transmitter itself, and the obtained single-frame voice information is subjected to low-delay improved discrete cosine transform respectively to obtain the single-frame spectrum coefficient of the frequency domain corresponding to the single-frame voice information; Calculate the spectral entropy of the pseudo spectrum corresponding to the single frame speech information according to the single frame spectrum coefficient; Determine the noise in the speech information according to the spectral entropy and a preset threshold value; Eliminating the noise, and performing the remaining encoding steps on the noise reduction spectrum coefficients obtained by the noise elimination, obtaining a code stream corresponding to the noise reduction spectrum coefficients and outputting the code stream; as well as The code stream is received by a Bluetooth receiving end connected to the Bluetooth transmitting end, and the code stream is decoded by a decoder of the Bluetooth receiving end itself until an inverse discrete cosine transform is performed to output noise reduction voice information.

2. The speech noise reduction method according to claim 1, characterized in that: The step of calculating the spectral entropy of the pseudo spectrum corresponding to the single frame speech information according to the single frame spectrum coefficient further includes: Calculate the pseudo spectrum according to the single-frame spectrum coefficients, and obtain pseudo spectrum coefficients corresponding to the pseudo spectrum; Divide the pseudo-spectrum coefficients into pseudo-spectrum sub-bands to obtain pseudo-spectrum sub-band pseudo-spectrum coefficients corresponding to a plurality of pseudo-spectrum sub-bands respectively; Calculate the pseudo-spectral energies of the pseudo-spectral sub-bands according to the pseudo-spectral coefficients of the pseudo-spectral sub-bands, and then obtain the total energy of the plurality of pseudo-spectral sub-bands; The spectral entropy is obtained by calculation according to the pseudo-spectral energy and the total energy.

3. The speech noise reduction method according to claim 1, characterized in that: The step of determining the noise in the speech information according to the spectral entropy and the preset threshold value further includes: Presetting the preset threshold value; When the spectrum entropy is greater than a preset threshold value, the single-frame speech information corresponding to the pseudo-spectrum is determined to be noise.

4. The speech noise reduction method according to claim 3, characterized in that: When the spectral entropy is greater than a preset threshold value, the method further includes: According to a preset energy smoothing factor, energy estimates corresponding to each pseudo-spectrum subband in the noise are calculated.

5. The speech noise reduction method according to claim 1, characterized in that: The performing noise elimination on the noise further comprises: Each of the single-frame spectrum coefficients is updated according to a preset pseudo-spectrum subband updating formula to obtain a single-frame updated spectrum coefficient corresponding to the single-frame spectrum coefficient.

6. The speech noise reduction method according to claim 5, characterized in that: The performing noise elimination on the noise further comprises: Noise suppression is performed on the single-frame spectrum coefficients corresponding to the noise.

7. A speech noise reduction device, characterized in that: include: A time-frequency conversion module is used to frame the voice information collected by the Bluetooth transmitter according to the encoder of the Bluetooth transmitter itself, and perform low-delay improved discrete cosine transform on the obtained single-frame voice information to obtain the single-frame spectrum coefficient of the frequency domain corresponding to the single-frame voice information; A spectral entropy calculation module, which is used to calculate the spectral entropy of the pseudo spectrum corresponding to the single frame speech information according to the single frame spectrum coefficient; A noise determination module, which is used to determine the noise in the speech information according to the spectral entropy and a preset threshold value; A noise elimination module, which is used to eliminate the noise, and perform the remaining encoding steps on the noise-reduced voice information after the noise elimination, obtain the code stream corresponding to the noise-reduced voice information and output it; as well as The time-frequency inverse transform module is used to receive the code stream according to the Bluetooth receiving end connected to the Bluetooth transmitting end, and use the decoder of the Bluetooth receiving end itself to decode the code stream until the inverse discrete cosine transform outputs the noise reduction voice information.

8. A speech noise reduction system, characterized in that: The speech noise reduction system comprises the speech noise reduction device as claimed in claim 7.

9. A computer-readable storage medium storing computer instructions, characterized in that: The computer instructions are operated to execute the speech noise reduction method according to any one of claims 1-6.

10. A computer device, characterized in that: include: at least one processor; as well as a memory in communication with the at least one processor; The memory stores computer instructions that can be executed by the at least one processor, and the at least one processor operates the computer instructions to execute the speech noise reduction method as described in any one of claims 1-6.