Bluetooth transmitting terminal voice noise reduction method and device based on non-negative matrix factorization, and medium

By using low-latency improved discrete cosine transform and subband energy non-negative matrix decomposition on the Bluetooth transmitter, the problems of large matrix size and poor noise reduction quality in the prior art are solved, and low-latency and efficient Bluetooth voice noise reduction are achieved. in multiple Bluetooth protocols.

CN119091898APending Publication Date: 2024-12-06SHENZHEN BARROT TECH LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411045520.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-01
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

In the prior art, non-negative matrix decomposition uses the amplitude spectrum output by FFT in Bluetooth audio denoising, resulting in large matrix size and complex calculations. The difference between the noisy speech phase spectrum and the pure speech phase spectrum leads to a decrease in the noise reduction quality, which increases the algorithm delay. and implementation difficulty.

Method used

The spectrum coefficient is obtained by using low-latency improved discrete cosine transformation at the Bluetooth transmitter, and non-negative matrix decomposition is performed based on the subband energy, reducing the matrix size, avoiding the use of noise-free speech phase spectrum for reconstruction, ensuring sound quality and reducing the computational volume.

Benefits of technology

It realizes low-latency voice noise reduction at the Bluetooth transmitter, ensures sound quality, reduces matrix size and computing volume, and is suitable for classic Bluetooth and low-power Bluetooth, reducing sound quality loss and algorithm complexity.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119091898A_ABST
    Figure CN119091898A_ABST
Patent Text Reader

Abstract

The invention discloses a Bluetooth transmitting end voice noise reduction method and device based on non-negative matrix factorization, a storage medium and equipment, and belongs to the technical field of Bluetooth audios, and the method comprises the steps: inputting monaural voice PCM data at a Bluetooth transmitting end, and executing low-delay improved discrete cosine transform to obtain an MDCT spectral coefficient; calculating sub-band energy according to the MDCT spectral coefficient, and constructing a sub-band energy observation matrix; performing non-negative matrix factorization on the sub-band energy observation matrix, and calculating a voice sub-band energy value and a noise sub-band energy value; calculating a sub-band noise reduction gain according to the voice sub-band energy value and the noise sub-band energy value, and performing interpolation on the sub-band noise reduction gain to obtain all frequency bin gains; obtaining a noise reduction spectral coefficient according to the frequency bin gain and the MDCT spectral coefficient; and continuously executing a coding process on the noise reduction spectrum coefficient to obtain a noise reduction voice code stream. According to the invention, the non-negative matrix factorization is executed at the Bluetooth transmitting end based on the sub-band energy to realize noise reduction, the tone quality is ensured, and the operand is greatly reduced.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application belongs to the field of Bluetooth audio technology, and in particular, relates to a Bluetooth transmitter voice noise reduction method, device, storage medium and equipment based on non-negative matrix decomposition. Background Art

[0002] The current mainstream Bluetooth audio encoders include: SBC: mandatory requirement of A2DP protocol, the most widely used, and must be supported by all Bluetooth audio devices, but the sound quality is average; AAC-LC: good sound quality and widely used, many mainstream mobile phones support it, but compared with SBC, it occupies a large amount of memory and has high computational complexity. Many Bluetooth devices are based on embedded platforms with limited battery capacity, poor processor computing power and limited memory, and its patent fees are high; aptX series: good sound quality, but high bit rate, aptX requires a bit rate of 384kbps, while aptX -HD has a bit rate of 576kbps, and it is a unique technology of Qualcomm and is relatively closed; LDAC: The sound quality is good, but the bit rate is also very high, which are 330kbps, 660kbps and 990kbps respectively. Since the wireless environment of Bluetooth devices is particularly complex, it is difficult to stably support such a high bit rate, and it is a unique technology of Sony and is also very closed; LHDC: The sound quality is good, but the bit rate is also very high, typically including 400kbps, 600kbps and 900kbps. Such a high bit rate puts high demands on the baseband / RF design of Bluetooth. Based on the above reasons, the Bluetooth International Alliance Bluetooth Sig and many manufacturers launched LC3, which is mainly for low-power Bluetooth and can also be used for classic Bluetooth. It has the advantages of lower latency, higher sound quality and coding gain, and no patent fees in the Bluetooth field, and has attracted the attention of many manufacturers.

[0003] In many Bluetooth applications, such as Bluetooth calls, Bluetooth microphones, and recordings, noise reduction is required.

[0004] Non-negative matrix factorization (NMF) makes all matrix components after decomposition non-negative values ​​and realizes non-linear dimensionality reduction. NMF has gradually become one of the most popular multi-dimensional data processing tools in research fields such as signal processing, biomedical engineering, pattern recognition, computer vision and image engineering. Non-negative matrix factorization has many applications in signal separation applications. In the prior art, a method of using non-negative matrix factorization to achieve speech denoising includes offline training process and real-time denoising process, which is briefly described as follows:

[0005] (1) Offline training, including: selecting clean speech and noise to construct a training set; extracting features from the training set, i.e., performing Fourier transform and obtaining the amplitude spectrum; and using the amplitude spectrum to construct the clean speech amplitude spectrum matrix V speech Sum noise amplitude spectrum matrix V noise ; Perform non-negative matrix decomposition on the amplitude spectrum to obtain the basic matrix W speech and W noise .

[0006] (2) Real-time noise reduction, including: extracting features from noisy speech to obtain the amplitude spectrum and constructing the amplitude spectrum observation matrix V noisy Based on the basic matrix of the training output, the activation matrix H′ is obtained by performing non-negative matrix decomposition on the noisy speech amplitude spectrum matrix speech and H′ noise ; Calculate the amplitude spectrum of speech and noise; Estimate the signal-to-noise ratio; Construct a Wiener filter; Filter the noisy amplitude spectrum; Combine the phase spectrum of the noisy speech signal with the filtered amplitude spectrum to perform an inverse Fourier transform to obtain the denoised speech.

[0007] The shortcomings of the above method are: non-negative matrix decomposition uses the amplitude spectrum based on FFT output, which leads to large sizes of the amplitude spectrum matrix, the decomposed basic matrix and the activation matrix. When the matrix is ​​decomposed, the larger the size, the more complex the operation, which limits its scope of application; when performing inverse Fourier transform on the amplitude spectrum of the denoised speech, the phase spectrum of the noisy speech is used. Since the phase spectrum of the noisy speech is different from the amplitude spectrum of the pure speech, this also reduces the quality of the speech after denoising and reconstruction; the existing technology will introduce algorithm delays, increasing the difficulty of application implementation. Summary of the invention

[0008] In response to the above-mentioned technical problems existing in the prior art, the present application provides a Bluetooth transmitter speech noise reduction method, device, storage medium and equipment based on non-negative matrix decomposition. At the Bluetooth transmitter, the spectral coefficients are first obtained based on the low-latency improved discrete cosine transform, and then non-negative matrix decomposition is performed based on the sub-band energy to achieve noise reduction, which ensures the sound quality, reduces the size of the matrix, and greatly reduces the amount of calculation.

[0009] In order to achieve the above-mentioned purpose, the first technical solution adopted in the present application is: to provide a Bluetooth transmitter speech noise reduction method based on non-negative matrix decomposition, including: at the Bluetooth transmitter, inputting monophonic speech PCM data, and performing a low-delay improved discrete cosine transform to obtain MDCT spectrum coefficients; calculating subband energy according to the MDCT spectrum coefficients, and constructing a subband energy observation matrix; performing non-negative matrix decomposition on the subband energy observation matrix, and calculating the speech subband energy value and the noise subband energy value; calculating the subband noise reduction gain according to the speech subband energy value and the noise subband energy value, and performing interpolation on the subband noise reduction gain to obtain all frequency bin gains; obtaining the noise reduction spectrum coefficient according to the frequency bin gain and the MDCT spectrum coefficient; and continuing to perform the encoding process on the noise reduction spectrum coefficient to obtain a noise-reduced speech code stream.

[0010] The second technical solution adopted in the present application is: to provide a Bluetooth transmitter speech noise reduction device based on non-negative matrix decomposition, including: a module for inputting monophonic speech PCM data at the Bluetooth transmitter, and performing a low-delay improved discrete cosine transform to obtain MDCT spectrum coefficients; a module for calculating subband energy according to the MDCT spectrum coefficients, and constructing a subband energy observation matrix; a module for performing non-negative matrix decomposition on the subband energy observation matrix, and calculating speech subband energy values ​​and noise subband energy values; a module for calculating subband noise reduction gains according to speech subband energy values ​​and noise subband energy values, and performing interpolation on the subband noise reduction gains to obtain all frequency bin gains; a module for obtaining noise reduction spectrum coefficients according to frequency bin gains and MDCT spectrum coefficients; a module for continuing the encoding process on the noise reduction spectrum coefficients to obtain a noise reduction speech code stream.

[0011] The third technical solution adopted in the present application is: providing a computer-readable storage medium storing computer instructions, wherein the computer instructions are operated to execute the Bluetooth transmitter speech noise reduction method based on non-negative matrix decomposition in solution one.

[0012] The fourth technical solution adopted in the present application is: to provide a computer device, which includes a processor and a memory, the memory stores computer instructions, wherein the processor operates the computer instructions to execute the Bluetooth transmitter speech noise reduction method based on non-negative matrix decomposition in solution one.

[0013] The beneficial effects that can be achieved by the technical solution of the present application are: the technical solution of the present application can be applied to both classic Bluetooth (BR, EDR) and low-power Bluetooth (LE, LE Audio). At the Bluetooth transmitter, the MDCT spectrum coefficients are first obtained based on the low-latency improved discrete cosine transform, thereby avoiding the problem of reduced sound quality caused by the need to use the phase of noisy speech for reconstruction in the prior art. Then, non-negative matrix decomposition is performed based on the sub-band energy spectrum to achieve noise reduction, thereby ensuring sound quality, reducing the size of the matrix, and greatly reducing the amount of calculation. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a flowchart of a specific implementation method of the Bluetooth transmitter voice noise reduction method based on non-negative matrix decomposition of the present application;

[0015] Figure 2 It is a schematic diagram of a specific example of speech noise reduction at a Bluetooth transmitter based on non-negative matrix decomposition in the present application;

[0016] Figure 3 It is a schematic diagram of a specific implementation of the Bluetooth transmitter voice noise reduction device based on non-negative matrix decomposition of the present application. DETAILED DESCRIPTION

[0017] The preferred embodiments of the present application are described in detail below in conjunction with the accompanying drawings so that the advantages and features of the present application can be more easily understood by those skilled in the art, thereby making a clearer and more definite definition of the protection scope of the present application.

[0018] The inventive concept of this application is as follows:

[0019] 1. At the Bluetooth transmitter, encode the voice signal and generate a noise-reduced bit stream at the same time;

[0020] 2. Send the generated code stream to the receiving end;

[0021] 3. The receiving end decodes the PCM signal and outputs the noise-reduced speech.

[0022] The specific implementation of the present application is described in detail below using the 48 kHz sampling rate as an example. The principles of other sampling rates are similar to those of the 48 kHz sampling rate and will not be repeated here.

[0023] Figure 1 It is a flowchart of a specific implementation method of the Bluetooth transmitter voice noise reduction method based on non-negative matrix decomposition of the present application.

[0024] exist Figure 1 In a specific embodiment shown, the Bluetooth transmitter speech noise reduction method based on non-negative matrix decomposition of the present application includes process S101, at the Bluetooth transmitter, monophonic speech PCM data is input, and a low-delay improved discrete cosine transform is performed to obtain MDCT spectrum coefficients.

[0025] In a specific embodiment of the present application, monophonic speech PCM data is input, and a low-delay improved discrete cosine transform is performed to obtain MDCT spectral coefficients, including: performing framing and windowing on the monophonic speech PCM data to obtain windowed audio frames; and performing a low-delay improved discrete cosine transform on the windowed audio frames to obtain MDCT spectral coefficients.

[0026] Specifically, firstly, monophonic speech PCM data is input, then frame division and windowing are performed on the monophonic speech PCM data to obtain windowed audio frames, and then low-delay improved discrete cosine transform is performed on the windowed audio frames to obtain MDCT spectrum coefficients.

[0027] In a specific example of the present application, taking 48kHz sampling rate and 10ms frame length as an example, the MDCT spectrum coefficient is:

[0028] X(k), k=0,…,N F -1, where N F =480

[0029] exist Figure 1 In a specific implementation shown, the Bluetooth transmitter speech noise reduction method based on non-negative matrix decomposition of the present application includes process S102, calculating sub-band energy according to MDCT spectrum coefficients and constructing a sub-band energy observation matrix.

[0030] In a specific embodiment of the present application, subband energy is calculated based on MDCT spectrum coefficients, including: dividing the MDCT spectrum coefficients into subbands to obtain a frequency bin index table corresponding to each subband; and obtaining subband energy based on the MDCT spectrum coefficients and the frequency bin index table.

[0031] Specifically, the MDCT spectrum coefficients are first divided into subbands, and the above MDCT spectrum coefficients are divided into 64 subbands, and the subband sequence number is recorded as b = 0, 1, 2, ..., 63. This application takes the division of the MDCT spectrum coefficients into 64 subbands as an example. In practical applications, the subband size can be flexibly configured according to the needs to take into account the requirements of noise reduction performance and computational complexity, and then the frequency bin index table corresponding to each subband is obtained, as follows:

[0032] Index

[64] =[0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,20,22,24,26,28,30,32,34,36,39,42,45,48,51,55,59,63, 67,71,76,81,86,92,98,105,112,119,127,135,144,154,164,175,186,198,211,225,240,256,273,291,310,330,352,375,400]

[0033] Then, the subband energy is calculated based on the MDCT spectrum coefficients and the frequency bin index table:

[0034]

[0035] After obtaining the subband energy, construct the subband energy observation matrix V noisy .

[0036] exist Figure 1 In a specific embodiment shown, the Bluetooth transmitter speech noise reduction method based on non-negative matrix decomposition of the present application includes process S103, performing non-negative matrix decomposition on the sub-band energy observation matrix, and calculating the speech sub-band energy value and the noise sub-band energy value.

[0037] In a specific embodiment of the present application, non-negative matrix decomposition is performed on the sub-band energy observation matrix, including: fixing the speech basic matrix and the noise basic matrix respectively to perform non-negative matrix decomposition on the sub-band energy observation matrix to obtain a speech activation matrix and a noise activation matrix.

[0038] Specifically, the fixed speech basic matrix W speech , perform non-negative matrix decomposition on the subband energy observation matrix to obtain the speech activation matrix H′ speech .

[0039] Then, fix the noise basis matrix W noise , perform non-negative matrix factorization on the subband energy observation matrix to obtain the noise activation matrix H′ noise .

[0040] Next, the speech basic matrix W speech and the noise basis matrix W noise The training method is briefly described as follows:

[0041] (1) Selecting clean speech and noise as training sets (the following takes the training process of clean speech as an example, the training process of noise is similar and will not be repeated);

[0042] (2) performing framing, windowing and low-delay modified discrete cosine transform on the clean speech to obtain spectral coefficients;

[0043] (3) Calculate the subband energy and construct the observation matrix V;

[0044] (4) Perform non-negative matrix decomposition on the observation matrix V to obtain the basic matrix W (i.e., the speech basic matrix W speech ) and the activation matrix H.

[0045] For a non-negative matrix V, non-negative matrix decomposition is to find the optimal solution so that the product of W and H obtained by decomposition is an approximation of V. The steps of non-negative matrix decomposition are briefly described as follows:

[0046] (1) Construct the observation matrix V KL , there are K rows and L columns in total, each column is the amplitude spectrum of a frame, and there are L columns in total.

[0047] (2) Initialize the basic matrix WKM .

[0048] (3) Initialize the activation matrix H ML In a typical scenario, the above parameters can be K=480, M=240, and L=3~5.

[0049] (4) Iterative solution based on KL divergence criterion is briefly described as follows:

[0050]

[0051] in represents element-wise multiplication, T represents the transpose operation of the matrix, and I represents the identity matrix.

[0052] During the iteration process, the KL divergence of V and WH is calculated, and the process stops when it is less than the preset threshold. In order to avoid long calculation time caused by abnormal conditions, the number of iterations can also be preset, and the process stops when the preset number is reached, thereby obtaining the basis matrix and activation matrix.

[0053] In a specific embodiment of the present application, the speech subband energy value and the noise subband energy value are calculated, including: multiplying the speech basic matrix with the speech activation matrix to obtain the speech subband energy value; and multiplying the noise basic matrix with the noise activation matrix to obtain the noise subband energy value.

[0054] Specifically, calculate the speech subband energy value:

[0055] V′ speech (b) = W speech H′ speech

[0056] Calculate the noise subband energy value:

[0057] N′ noise (b) = W noise H′ noise

[0058] exist Figure 1 In a specific embodiment shown, the Bluetooth transmitter speech noise reduction method based on non-negative matrix decomposition of the present application includes process S104, calculating the sub-band noise reduction gain according to the speech sub-band energy value and the noise sub-band energy value, and performing interpolation on the sub-band noise reduction gain to obtain all frequency bin gains.

[0059] In this specific implementation, first, the sub-band noise reduction gain is calculated:

[0060]

[0061] Then the sub-band noise reduction gains are interpolated to obtain all frequency bin gains.

[0062] In a specific embodiment of the present application, interpolation is performed on the sub-band noise reduction gain to obtain all frequency bin gains, including: performing a predetermined operation on the sub-band noise reduction gain and a waveform symmetrical about the sub-band center frequency bin to obtain all frequency bin gains.

[0063] Specifically, all frequency bin gains are:

[0064]

[0065] Among them, Weight b (k) is a waveform symmetrical about the sub-band center frequency bin, which satisfies the following conditions:

[0066]

[0067] exist Figure 1 In a specific implementation manner shown, the Bluetooth transmitter speech noise reduction method based on non-negative matrix decomposition of the present application includes process S105, obtaining noise reduction spectrum coefficients according to frequency bin gains and MDCT spectrum coefficients.

[0068] In a specific embodiment of the present application, the noise reduction spectrum coefficients are obtained according to the frequency bin gains and the MDCT spectrum coefficients, including: multiplying the frequency bin gains and the MDCT spectrum coefficients to obtain the noise reduction spectrum coefficients.

[0069] Specifically, the frequency bin gain is applied to the MDCT spectrum coefficients to obtain the noise reduction spectrum coefficients:

[0070] X ′ (k)=X(k)Gain f ′ req (k)

[0071] exist Figure 1 In a specific implementation manner shown, the Bluetooth transmitter speech noise reduction method based on non-negative matrix decomposition of the present application includes process S106, which continues to perform the encoding process on the noise reduction spectrum coefficients to obtain a noise reduction speech code stream.

[0072] In a specific example of the present application, continuing the encoding process includes: performing transform domain noise shaping, time domain noise shaping, quantization, noise level estimation, arithmetic coding and residual coding, and then outputting a noise-reduced speech code stream.

[0073] In a specific embodiment of the present application, the Bluetooth transmitter speech noise reduction method based on non-negative matrix decomposition of the present application also includes: sending the noise reduction speech code stream to the Bluetooth receiver via Bluetooth wireless transmission.

[0074] In a specific embodiment of the present application, the Bluetooth transmitter speech noise reduction method based on non-negative matrix decomposition of the present application also includes: at the Bluetooth receiving end, decoding the received noise reduction speech code stream and outputting the noise-reduced PCM data.

[0075] Figure 2 It is a schematic diagram of a specific example of speech noise reduction at a Bluetooth transmitter based on non-negative matrix decomposition in the present application.

[0076] like Figure 2 A specific example shown, the speech noise reduction process of the Bluetooth transmitter based on non-negative matrix decomposition of the present application is: first, at the Bluetooth transmitter, input audio data (PCM), perform framing and windowing operations, then perform low-delay improved discrete cosine transform to obtain MDCT spectrum coefficients, then calculate the subband energy and form a subband energy observation matrix, perform non-negative matrix decomposition to obtain the activation matrix of speech and noise, then calculate the subband energy of speech and noise, thereby calculating the subband gain and interpolation, then apply the gain to the MDCT spectrum coefficients, thereby obtaining the spectrum coefficients after noise reduction, and then continue to perform transform domain noise shaping, time domain noise shaping, quantization, noise level estimation, arithmetic coding and residual coding on the spectrum coefficients after noise reduction, and finally output the speech code stream after noise reduction. This example is at the Bluetooth transmitter, based on the spectrum coefficients obtained by discrete cosine transform, and then implements non-negative matrix decomposition and noise reduction on the subband, and its advantages are:

[0077] (1) Reduce the size of the matrix. For example, for a codec similar to the sound quality, the amplitude spectrum coefficients of each frame output by Fourier transform are 2N, and the coefficients of the amplitude spectrum using discrete cosine transform are N. If sub-band division is used, the number is even smaller, typically 960 Fourier transform spectrum coefficients, 480 discrete cosine transform spectrum coefficients and 64 Bark sub-bands;

[0078] (2) Based on the above (1), the amount of calculation is greatly reduced;

[0079] (3) The inverse Fourier transform requires a phase spectrum, but the phase spectrum of pure speech cannot be obtained, so only the phase spectrum of noisy speech can be used. However, the processing based on the low-latency improved discrete cosine transform in the present application does not require a phase spectrum during the inverse transform, thereby avoiding the damage of the noisy speech phase spectrum to the reconstructed speech.

[0080] In the Bluetooth transmitter speech noise reduction method based on non-negative matrix decomposition of the present application, at the Bluetooth transmitter, the MDCT spectrum coefficients are first obtained based on the low-latency improved discrete cosine transform, avoiding the problem of reduced sound quality caused by the need to use the phase of noisy speech for reconstruction in the prior art, and then non-negative matrix decomposition is performed based on the sub-band energy spectrum to achieve noise reduction, which ensures sound quality, reduces the size of the matrix, and can greatly reduce the amount of calculation.

[0081] Figure 3 It is a schematic diagram of a specific implementation of the Bluetooth transmitter voice noise reduction device based on non-negative matrix decomposition of the present application.

[0082] exist Figure 3 In a specific embodiment shown, the Bluetooth transmitter speech noise reduction device based on non-negative matrix decomposition of the present application includes: a module 301 for inputting monophonic speech PCM data at the Bluetooth transmitter and performing a low-delay improved discrete cosine transform to obtain MDCT spectrum coefficients; a module 302 for calculating subband energy according to the MDCT spectrum coefficients and constructing a subband energy observation matrix; a module 303 for performing non-negative matrix decomposition on the subband energy observation matrix and calculating speech subband energy values ​​and noise subband energy values; a module 304 for calculating subband noise reduction gains according to speech subband energy values ​​and noise subband energy values, and performing interpolation on the subband noise reduction gains to obtain all frequency bin gains; a module 305 for obtaining noise reduction spectrum coefficients according to frequency bin gains and MDCT spectrum coefficients; and a module 306 for continuing to perform the encoding process on the noise reduction spectrum coefficients to obtain a noise reduction speech code stream.

[0083] In a specific embodiment of the present application, monophonic speech PCM data is input, and a low-delay improved discrete cosine transform is performed to obtain MDCT spectral coefficients, including: performing framing and windowing on the monophonic speech PCM data to obtain windowed audio frames; and performing a low-delay improved discrete cosine transform on the windowed audio frames to obtain MDCT spectral coefficients.

[0084] In a specific embodiment of the present application, subband energy is calculated based on MDCT spectrum coefficients, including: dividing the MDCT spectrum coefficients into subbands to obtain a frequency bin index table corresponding to each subband; and obtaining subband energy based on the MDCT spectrum coefficients and the frequency bin index table.

[0085] In a specific embodiment of the present application, non-negative matrix decomposition is performed on the sub-band energy observation matrix, including: fixing the speech basic matrix and the noise basic matrix respectively to perform non-negative matrix decomposition on the sub-band energy observation matrix to obtain a speech activation matrix and a noise activation matrix.

[0086] In a specific embodiment of the present application, the speech subband energy value and the noise subband energy value are calculated, including: multiplying the speech basic matrix with the speech activation matrix to obtain the speech subband energy value; and multiplying the noise basic matrix with the noise activation matrix to obtain the noise subband energy value.

[0087] In a specific embodiment of the present application, interpolation is performed on the sub-band noise reduction gain to obtain all frequency bin gains, including: performing a predetermined operation on the sub-band noise reduction gain and a waveform symmetrical about the sub-band center frequency bin to obtain all frequency bin gains.

[0088] In a specific embodiment of the present application, the noise reduction spectrum coefficients are obtained according to the frequency bin gains and the MDCT spectrum coefficients, including: multiplying the frequency bin gains and the MDCT spectrum coefficients to obtain the noise reduction spectrum coefficients.

[0089] In a specific embodiment of the present application, the Bluetooth transmitter speech noise reduction device based on non-negative matrix decomposition of the present application also includes: a module for sending the noise reduction speech code stream to the Bluetooth receiver via Bluetooth wireless transmission.

[0090] In a specific embodiment of the present application, the Bluetooth transmitter speech noise reduction device based on non-negative matrix decomposition of the present application also includes: a module for decoding the received noise reduction speech code stream at the Bluetooth receiving end and outputting the noise-reduced PCM data.

[0091] In the Bluetooth transmitter speech noise reduction device based on non-negative matrix decomposition of the present application, at the Bluetooth transmitter, the MDCT spectrum coefficients are first obtained based on the low-latency improved discrete cosine transform, avoiding the problem of reduced sound quality caused by the need to use the phase of noisy speech for reconstruction in the prior art, and then non-negative matrix decomposition is performed based on the sub-band energy spectrum to achieve noise reduction, which ensures sound quality, reduces the size of the matrix, and can greatly reduce the amount of calculation.

[0092] In a specific embodiment of the present application, a computer-readable storage medium stores computer instructions, which are operated to execute the Bluetooth transmitter voice noise reduction method based on non-negative matrix decomposition described in any embodiment. The storage medium can be directly in hardware, in a software module executed by a processor, or in a combination of the two.

[0093] The software modules may reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, removable disk, CD-ROM or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor so that the processor can read information from and write information to the storage medium.

[0094] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA), or other programmable logic devices, discrete gates or transistor logic, discrete hardware components, or any combination thereof. A general-purpose processor may be a microprocessor, but in an alternative, the processor may be any conventional processor, controller, microcontroller, or state machine. The processor may also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors combined with a DSP core, or any other such configuration. In an alternative, the storage medium may be integrated with the processor. The processor and the storage medium may reside in an ASIC. The ASIC may reside in a user terminal. In an alternative, the processor and the storage medium may reside in a user terminal as discrete components.

[0095] In a specific embodiment of the present application, a computer device includes a processor and a memory, wherein the memory stores computer instructions, wherein the processor operates the computer instructions to execute the Bluetooth transmitter speech noise reduction method based on non-negative matrix decomposition described in any embodiment.

[0096] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0097] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0098] The above description is only an embodiment of the present application and does not limit the patent scope of the present application. Any equivalent structural transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A Bluetooth transmitter speech noise reduction method based on non-negative matrix decomposition, characterized in that: include: At the Bluetooth transmitter, monophonic voice PCM data is input and low-latency modified discrete cosine transform is performed to obtain MDCT spectrum coefficients; Calculating subband energy according to the MDCT spectrum coefficients and constructing a subband energy observation matrix; Performing non-negative matrix decomposition on the sub-band energy observation matrix, and calculating speech sub-band energy values ​​and noise sub-band energy values; Calculating a sub-band noise reduction gain according to the speech sub-band energy value and the noise sub-band energy value, and performing interpolation on the sub-band noise reduction gain to obtain all frequency bin gains; Obtaining noise reduction spectrum coefficients according to the frequency bin gain and the MDCT spectrum coefficients; as well as The encoding process is continued on the noise reduction spectrum coefficients to obtain a noise reduction speech code stream.

2. The method for reducing the noise of a Bluetooth transmitter based on non-negative matrix decomposition according to claim 1, wherein: The input monophonic speech PCM data and the low-delay improved discrete cosine transform are performed to obtain MDCT spectrum coefficients, including: Performing framing and windowing on the monophonic voice PCM data to obtain a windowed audio frame; and A low-delay modified discrete cosine transform is performed on the windowed audio frame to obtain the MDCT spectrum coefficients.

3. The method for reducing noise at the Bluetooth transmitter based on non-negative matrix decomposition according to claim 1, characterized in that: The calculating sub-band energy according to the MDCT spectrum coefficients comprises: Dividing the MDCT spectrum coefficients into sub-bands to obtain a frequency bin index table corresponding to each of the sub-bands; and The subband energy is obtained according to the MDCT spectrum coefficients and the frequency bin index table.

4. The method for reducing the noise of a Bluetooth transmitter based on non-negative matrix decomposition according to claim 1, wherein: The performing non-negative matrix decomposition on the sub-band energy observation matrix comprises: The speech basic matrix and the noise basic matrix are fixed respectively, and non-negative matrix decomposition is performed on the sub-band energy observation matrix to obtain a speech activation matrix and a noise activation matrix.

5. The method for reducing the noise of a Bluetooth transmitter based on non-negative matrix decomposition according to claim 4, characterized in that: The calculating of the speech sub-band energy value and the noise sub-band energy value comprises: Multiplying the speech basic matrix by the speech activation matrix to obtain the speech sub-band energy value; and The noise basic matrix is ​​multiplied by the noise activation matrix to obtain the noise subband energy value.

6. The method for reducing the noise of a Bluetooth transmitter based on non-negative matrix decomposition according to claim 1, characterized in that: The step of obtaining the noise reduction spectrum coefficient according to the frequency bin gain and the MDCT spectrum coefficient comprises: The frequency bin gain is multiplied by the MDCT spectrum coefficient to obtain the noise reduction spectrum coefficient.

7. The method for reducing speech noise at a Bluetooth transmitter based on non-negative matrix decomposition according to claim 3, characterized in that: The interpolation of the sub-band noise reduction gain to obtain all frequency bin gains includes: The sub-band noise reduction gain is subjected to a predetermined operation with a waveform symmetrical with respect to the sub-band center frequency bin to obtain all the frequency bin gains.

8. A Bluetooth transmitter speech noise reduction device based on non-negative matrix decomposition, characterized in that: include: A module for inputting monophonic voice PCM data at the Bluetooth transmitter and performing low-latency modified discrete cosine transform to obtain MDCT spectrum coefficients; A module for calculating sub-band energy according to the MDCT spectrum coefficients and constructing a sub-band energy observation matrix; A module for performing non-negative matrix decomposition on the sub-band energy observation matrix and calculating a speech sub-band energy value and a noise sub-band energy value; A module for calculating a sub-band noise reduction gain according to the speech sub-band energy value and the noise sub-band energy value, and performing interpolation on the sub-band noise reduction gain to obtain all frequency bin gains; A module for obtaining noise reduction spectrum coefficients according to the frequency bin gains and the MDCT spectrum coefficients; A module for continuing to perform the encoding process on the noise reduction spectrum coefficients to obtain a noise reduction speech code stream.

9. A computer-readable storage medium storing computer instructions, wherein the computer instructions are operated to execute the Bluetooth transmitter speech noise reduction method based on non-negative matrix decomposition according to any one of claims 1 to 7.

10. A computer device comprising a processor and a memory, wherein the memory stores computer instructions, wherein the processor operates the computer instructions to execute the Bluetooth transmitter speech noise reduction method based on non-negative matrix decomposition according to any one of claims 1-7.