CN117059114A - Voice processing method and related device - Google Patents
Voice processing method and related device Download PDFInfo
- Publication number
- CN117059114A CN117059114A CN202311086762.3A CN202311086762A CN117059114A CN 117059114 A CN117059114 A CN 117059114A CN 202311086762 A CN202311086762 A CN 202311086762A CN 117059114 A CN117059114 A CN 117059114A
- Authority
- CN
- China
- Prior art keywords
- time
- characteristic data
- voice
- frequency signal
- dimension
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000003672 processing method Methods 0.000 title abstract description 25
- 238000012545 processing Methods 0.000 claims abstract description 136
- 238000006243 chemical reaction Methods 0.000 claims abstract description 51
- 230000015572 biosynthetic process Effects 0.000 claims abstract description 21
- 238000003786 synthesis reaction Methods 0.000 claims abstract description 21
- 238000000605 extraction Methods 0.000 claims abstract description 16
- 238000000034 method Methods 0.000 claims description 48
- 238000012549 training Methods 0.000 claims description 35
- 238000004590 computer program Methods 0.000 claims description 22
- 238000007781 pre-processing Methods 0.000 claims description 19
- 238000009432 framing Methods 0.000 claims description 14
- 230000009466 transformation Effects 0.000 claims description 8
- 238000011084 recovery Methods 0.000 claims description 2
- 238000001228 spectrum Methods 0.000 description 19
- 230000000694 effects Effects 0.000 description 14
- 238000004891 communication Methods 0.000 description 11
- 230000006870 function Effects 0.000 description 10
- 230000008569 process Effects 0.000 description 9
- 238000004422 calculation algorithm Methods 0.000 description 8
- 238000010586 diagram Methods 0.000 description 8
- 239000000047 product Substances 0.000 description 6
- 230000001131 transforming effect Effects 0.000 description 5
- 230000001360 synchronised effect Effects 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000007726 management method Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 239000000835 fiber Substances 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 230000017105 transposition Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Speech or voice signal processing techniques to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/27—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the analysis technique
- G10L25/30—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the analysis technique using neural networks
Landscapes
- Engineering & Computer Science (AREA)
- Human Computer Interaction (AREA)
- Signal Processing (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Computational Linguistics (AREA)
- Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Quality & Reliability (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Telephone Function (AREA)
Abstract
The application provides a voice processing method and a related device, firstly, a first voice signal is preprocessed to obtain first time-frequency signal characteristic data; then, carrying out feature extraction on the first time-frequency signal feature data through a trained voice processing model to obtain first feature data; performing dimension conversion on the first characteristic data through the trained voice processing model to obtain second characteristic data; decoding the second characteristic data through the trained voice processing model to obtain second time-frequency signal characteristic data, wherein the dimension of the second time-frequency signal characteristic data is the same as that of the first time-frequency signal characteristic data; finally, the second time-frequency signal characteristic data is subjected to voice synthesis to obtain a second voice signal, wherein the second voice signal is the voice signal after the first voice signal is enhanced. The information of the voice signal can not be omitted, and the accuracy of voice processing is improved.
Description
Technical Field
The application relates to the technical field of computers, in particular to a voice processing method and a related device.
Background
Along with the development of technology, the method for enhancing the voice is more and more detailed, and the current mainstream voice enhancement algorithm based on deep learning can be divided into two fields of time domain and time-frequency domain, but how to improve the effect of voice processing is still a problem to be solved.
Disclosure of Invention
In view of this, the present application provides a voice processing method and related apparatus, which can extract features of a voice signal from a time dimension and a frequency dimension through a trained voice processing model, and enhance the voice signal, thereby improving the effect of voice processing.
In a first aspect, an embodiment of the present application provides a method for processing speech, where the method includes:
preprocessing the first voice signal to obtain first time-frequency signal characteristic data;
performing feature extraction on the first time-frequency signal feature data through a trained voice processing model to obtain first feature data, wherein the first feature data comprises features of frequency dimension and features of time dimension;
performing dimension conversion on the first characteristic data through the trained voice processing model to obtain second characteristic data;
decoding the second characteristic data through the trained voice processing model to obtain second time-frequency signal characteristic data, wherein the dimension of the second time-frequency signal characteristic data is the same as that of the first time-frequency signal characteristic data;
And performing voice synthesis on the second time-frequency signal characteristic data to obtain a second voice signal, wherein the second voice signal is the voice signal after the first voice signal is enhanced.
In a second aspect, an embodiment of the present application provides a speech processing apparatus, including:
the preprocessing unit is used for preprocessing the first voice signal to obtain first time-frequency signal characteristic data;
the voice model unit is used for carrying out feature extraction on the first time-frequency signal feature data through a trained voice processing model to obtain first feature data, wherein the first feature data comprises features of frequency dimension and features of time dimension; performing dimension conversion on the first characteristic data through the trained voice processing model to obtain second characteristic data; decoding the second characteristic data through the trained voice processing model to obtain second time-frequency signal characteristic data, wherein the dimension of the second time-frequency signal characteristic data is the same as that of the first time-frequency signal characteristic data;
and the voice synthesis unit is used for performing voice synthesis on the second time-frequency signal characteristic data to obtain a second voice signal, wherein the second voice signal is the voice signal after the first voice signal is enhanced.
In a third aspect, an embodiment of the present application provides an electronic device, including a processor, a memory, a communication interface, and one or more programs, where the one or more programs are stored in the memory and configured to be executed by the processor, the programs including instructions for performing steps in any of the methods of the first aspect of the embodiments of the present application.
In a fourth aspect, embodiments of the present application provide a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program for electronic data exchange, wherein the computer program causes a computer to perform part or all of the steps as described in any of the methods of the first aspect of the embodiments of the present application.
In a fifth aspect, embodiments of the present application provide a computer program product, wherein the computer program product comprises a non-transitory computer readable storage medium storing a computer program operable to cause a computer to perform some or all of the steps described in any of the methods of the first aspect of the embodiments of the present application. The computer program product may be a software installation package.
It can be seen that, through the above-mentioned voice processing method and related apparatus, first, a first voice signal is preprocessed to obtain first time-frequency signal feature data; then, carrying out feature extraction on the first time-frequency signal feature data through a trained voice processing model to obtain first feature data, wherein the first feature data comprises features of frequency dimension and features of time dimension; performing dimension conversion on the first characteristic data through the trained voice processing model to obtain second characteristic data; decoding the second characteristic data through the trained voice processing model to obtain second time-frequency signal characteristic data, wherein the dimension of the second time-frequency signal characteristic data is the same as that of the first time-frequency signal characteristic data; finally, the second time-frequency signal characteristic data is subjected to voice synthesis to obtain a second voice signal, wherein the second voice signal is the voice signal after the first voice signal is enhanced. The information of the voice signal can not be omitted, and the accuracy of voice processing is improved.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for the description of the embodiments will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present application, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a system architecture diagram of a speech processing method according to an embodiment of the present application;
fig. 2 is a schematic structural diagram of an electronic device according to an embodiment of the present application;
fig. 3 is a schematic flow chart of a voice processing method according to an embodiment of the present application;
FIG. 4 is a flowchart of another speech processing method according to an embodiment of the present application;
fig. 5 is a flow chart of a voice processing method with a combined architecture according to an embodiment of the present application;
FIG. 6 is a block diagram illustrating functional units of a speech processing device according to an embodiment of the present application;
fig. 7 is a block diagram illustrating functional units of another speech processing device according to an embodiment of the present application.
Detailed Description
In order that those skilled in the art will better understand the present application, a technical solution in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in which it is apparent that the described embodiments are only some embodiments of the present application, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.
The terms first, second and the like in the description and in the claims and in the above-described figures are used for distinguishing between different objects and not necessarily for describing a sequential or chronological order. Furthermore, the terms "comprise" and "have," as well as any variations thereof, are intended to cover a non-exclusive inclusion. For example, a process, method, system, article, or apparatus that comprises a list of steps or elements is not limited to only those listed steps or elements but may include other steps or elements not listed or inherent to such process, method, article, or apparatus.
It should be understood that the term "and/or" is merely an association relationship describing the associated object, and means that three relationships may exist, for example, a and/or B may mean: a exists alone, A and B exist together, and B exists alone. In this context, the character "/" indicates that the front and rear associated objects are an "or" relationship. The term "plurality" as used in the embodiments of the present application means two or more.
"at least one" or the like in the embodiments of the present application means any combination of these items, including any combination of single item(s) or plural items(s), meaning one or more, and plural means two or more. For example, at least one (one) of a, b or c may represent the following seven cases: a, b, c, a and b, a and c, b and c, a, b and c. Wherein each of a, b, c may be an element or a set comprising one or more elements.
The "connection" in the embodiment of the present application refers to various connection manners such as direct connection or indirect connection, so as to implement communication between devices, which is not limited in the embodiment of the present application.
Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the application. The appearances of such phrases in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. Those of skill in the art will explicitly and implicitly appreciate that the embodiments described herein may be combined with other embodiments.
The following describes related content, concepts, meanings, technical problems, technical schemes, beneficial effects and the like related to the embodiment of the application.
Related terms:
speech enhancement refers to a technique for extracting useful speech signals from noise background, suppressing and reducing noise interference when speech signals are disturbed or even submerged by various kinds of noise.
At present, the time domain algorithm can only show the change of amplitude with time, and the information of the frequency band is ignored. The time-frequency domain algorithm can obtain the information of the frequency of the voice signal changing along with time, but effectively models the dependence of time and frequency at the same time, so that the summary of all the amplitude and complex spectrogram information is still a problem to be solved by the algorithm. Furthermore, in the frequency domain loss function calculation process used by the current algorithm, only one dimension frequency domain signal can be obtained for the same voice signal, which is not beneficial to the stable training of the network on complex conditions and is also not beneficial to recovering the perception characteristics of the enhanced signals in the voice signal.
In order to solve the above problems, the embodiment of the application provides a voice processing method, which can construct a voice processing model with better effect, extract the characteristics of voice signals from time dimension and frequency dimension, enhance the voice signals and improve the effect of voice processing.
First, referring to fig. 1, a system architecture of a speech processing method according to an embodiment of the present application is described, and fig. 1 is a system architecture diagram of a speech processing method according to an embodiment of the present application, including a speech processing model 110, a speech synthesis module 120, and a model training module 130.
The speech processing model 110 includes an encoder 111, a dual-path modeling module 112, and a decoder 113, where the encoder 111 is configured to extract feature data of the preprocessed speech signal, the dual-path modeling module 112 is configured to process the feature data from the encoder 111 in a time dimension and a frequency dimension to make full use of a harmonic spectrum structure in the speech signal, and the decoder 113 is configured to decode the feature data processed by the dual-path modeling module 112 to restore the dimension of the feature data to the dimension of the preprocessed feature data.
The speech synthesis module 120 is configured to obtain an enhanced speech signal from the feature data from the decoder 113 through speech synthesis.
The model training module 130 is configured to determine a multi-dimensional loss score according to a difference between an enhanced speech signal and a clean speech signal in multi-dimensions, where the clean speech signal is an ideal speech signal enhanced by an unprocessed speech signal, and feed back the multi-dimensional loss score to the speech processing model 110, and the speech processing model 110 may perform parameter adjustment based on the loss score, and iterate step by step until the loss score output by the model training module 130 converges or is lower than a certain value. It will be appreciated that the model training module 130 is a speech signal for making the speech signal outputted by the speech processing module 110 and synthesized by the speech synthesis module 120 as close to the ideal state as possible.
Therefore, the performance of the final voice processing model can be greatly improved by adjusting the parameters of the voice processing model through the loss fraction with multiple dimensions, and the accuracy of voice enhancement can be greatly improved by extracting the characteristics of the voice signal from the time dimension and the frequency dimension.
An electronic device according to an embodiment of the present application will be described with reference to fig. 2, and fig. 2 is a schematic structural diagram of an electronic device according to an embodiment of the present application, as shown in fig. 2, where the electronic device includes one or more application processors 220, a memory 230, a communication module 240, and one or more programs 231, and the application processor 220 is communicatively connected to the memory 230 and the communication module 240 through an internal communication bus.
Wherein the one or more programs 231 are stored in the memory 230 and configured to be executed by the application processor 220, the one or more programs 231 comprising instructions for performing any of the steps of the method embodiments described above.
The Application processor 220 may be, for example, a central processing unit (Central Processing Unit, CPU), a general purpose processor, a digital signal processor (Digital Signal Processor, DSP), an Application-specific integrated circuit (ASIC), a field programmable gate array (Field Programmable Gate Array, FPGA) or other programmable logic device, a transistor logic device, a hardware component, or any combination thereof. Which may implement or perform the various exemplary logic blocks, units and circuits described in connection with this disclosure. The application processor 220 may also be a combination that implements computing functionality, such as a combination comprising one or more microprocessors, a combination of a DSP and a microprocessor, or the like. The communication unit may be a communication module 240, a transceiver, a transceiving circuit, etc., and the storage unit may be a memory 230.
Memory 230 may be volatile memory or nonvolatile memory, or may include both volatile and nonvolatile memory. The nonvolatile memory may be a read-only memory (ROM), a Programmable ROM (PROM), an Erasable PROM (EPROM), an electrically Erasable EPROM (EEPROM), or a flash memory. The volatile memory may be random access memory (random access memory, RAM) which acts as an external cache. By way of example but not limitation, many forms of random access memory (random access memory, RAM) are available, such as Static RAM (SRAM), dynamic Random Access Memory (DRAM), synchronous Dynamic Random Access Memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced Synchronous Dynamic Random Access Memory (ESDRAM), synchronous Link DRAM (SLDRAM), and direct memory bus RAM (DR RAM).
It is understood that the electronic device 20 may include more or less structural elements than those in the above-described block diagrams, for example, including a power module, a physical key, a Wi-Fi module, a speaker, a bluetooth module, a sensor, a display module, etc., and is not limited herein, and the system architecture of the above-described voice processing method may be applied to the electronic device 20.
After understanding the software and hardware architecture of the embodiment of the present application, a speech processing method in the embodiment of the present application is described below with reference to fig. 3, and fig. 3 is a schematic flow chart of a speech processing method provided by the embodiment of the present application, which specifically includes the following steps:
in step 301, the first speech signal is preprocessed to obtain first time-frequency signal feature data.
The first voice signal may be a voice signal with noise, and the first time-frequency signal characteristic data may be a complex spectrum, a magnitude spectrum, and the like, which are not limited herein.
In one possible embodiment, the first speech signal may be framed, windowed, and fourier transformed to obtain the first time-frequency signal feature data, where the fourier transform may be a short-time fourier transform, and specifically, a coefficient set of the framing, windowing, and fourier transform may be configured as a 1 = { win_len, win_shift, fft_num }, where win_len is the window length, win_shift is the window moving step length, fft_num is the number of samples of the fourier transform, and then framing, windowing, and short-time fourier transforming the first speech signal by the coefficient set to obtain a complex spectrum S 1 (k, λ), where k is a frequency index and λ is a frame index.
By preprocessing the first speech signal to obtain first time-frequency signal feature data, data support can be provided for subsequent extracted features.
And step 302, performing feature extraction on the first time-frequency signal feature data through a trained voice processing model to obtain first feature data.
Wherein the first characteristic data includes a characteristic of a frequency dimension and a characteristic of a time dimension.
The trained speech processing model may include an encoder, where the encoder may employ a convolutional neural network structure, a cyclic neural network structure, a hybrid network former structure, and the like, and is configured to perform feature extraction on first time-frequency signal feature data to obtain first feature data, where the first feature data may be a compressed feature map, and reflect features of a frequency dimension and features of a time dimension of the first speech signal.
Therefore, the feature extraction is carried out on the first time-frequency signal feature data through the trained voice processing model to obtain the first feature data, so that the harmonic spectrum structure in the first voice signal can be fully utilized, and the voice processing effect is improved.
And 303, performing dimension conversion on the first feature data through the trained voice processing model to obtain second feature data.
The dimension conversion may be a processing manner such as transposition.
The first feature data may be subjected to a first dimension conversion by the trained speech processing model to obtain a first conversion feature, and then the first conversion feature may be subjected to a second dimension conversion by the trained speech processing model to obtain the second feature data, where the first dimension conversion is used to indicate a frequency dimension conversion or a time dimension conversion, the second dimension conversion is used to indicate a frequency dimension conversion or a time dimension conversion, and the first dimension conversion is different from the second dimension conversion. Namely, when the first dimension is converted into the frequency dimension, the second dimension is converted into the time dimension; when the first dimension is converted into the time dimension, the second dimension is converted into the frequency dimension.
Specifically, the trained speech processing model may include a dual-path modeling module, where the dual-path modeling module includes a frequency dimension module and a time dimension module, the frequency dimension module is used for performing frequency dimension conversion, and the time dimension module is used for performing time dimension conversion, where the frequency dimension module may employ a long-short-term memory network structure, a feedforward sequence memory network structure, a time convolution network structure, and the time dimension module may also employ a long-short-term memory network structure, a feedforward sequence memory network structure, a time convolution network structure, and the like.
In one possible embodiment, the frequency dimension module is disposed before and the time dimension module is disposed after, and the frequency dimension module performs frequency dimension conversion on the first feature data to obtain a first conversion feature, and then the time dimension module performs time dimension conversion on the first conversion feature to obtain second feature data. For example, the frequency dimension module first compresses the feature map Z from the encoder 1 (F, T) frequency dimension conversion to Z 2 (T, F) then converting the time dimension by the time dimension module to obtain Z identical to the dimension of the compressed feature map 3 (F,T)。
In one possible embodiment, the time dimension module is disposed before and the frequency dimension module is disposed after, and the time dimension module performs a time dimension conversion on the first feature data to obtain a first conversion feature, and then the frequency dimension module performs a frequency dimension conversion on the first conversion feature to obtain second feature data. For example, the time dimension module converts the compressed feature map z_1 (F, T) from the encoder to z_2 (T, F) in the time dimension, and then converts the compressed feature map z_3 (F, T) in the frequency dimension to z_3 (F, T) in the same dimension as the compressed feature map. Thus, the flexibility of data processing can be improved under the condition of ensuring the output quality of the model.
Therefore, the first characteristic data is subjected to dimension conversion through the trained voice processing model to obtain second characteristic data, the characteristics of the voice signals can be extracted from the time dimension and the frequency dimension, and the voice processing effect is improved.
And step 304, performing decoding processing on the second characteristic data through the trained voice processing model to obtain second time-frequency signal characteristic data.
The dimensions of the second time-frequency signal feature data are the same as those of the first time-frequency signal feature data, and the second time-frequency signal feature data may correspond to the first time-frequency signal feature data, and may be a complex spectrum, a magnitude spectrum, or the like, which is not specifically limited herein.
And performing dimension recovery on the second characteristic data through the trained voice processing model to obtain first mask data, and performing multiplication processing on the first mask data and the first time-frequency signal characteristic data through the trained voice processing model to obtain the second time-frequency signal characteristic data.
Specifically, the trained speech processing model includes a decoder that can restore the dimension of the second feature data to the dimension of the first time-frequency signal feature data, taking the first time-frequency signal feature data as a complex spectrum as an example, and can output a complex mask M (k, λ) and then convert the first complex spectrum S 1 Multiplying (k, lambda) by the complex mask to obtain a second complex spectrum S 2 (k,λ)＝M(k,λ)×S 1 (k, λ) and will not be described in detail herein.
Therefore, the second characteristic data is decoded through the trained voice processing model to obtain second time-frequency signal characteristic data, so that the enhanced time-frequency signal characteristic data can be obtained, and the effect of data processing is improved.
And step 305, performing voice synthesis on the second time-frequency signal characteristic data to obtain a second voice signal.
Wherein the second voice signal is the voice signal after the first voice signal is enhanced.
The second time-frequency signal characteristic data can be synthesized into a second voice signal through a splicing and adding algorithm, inverse Fourier transform and other processes, so that voice enhancement processing is completed.
Therefore, through the voice processing method, the voice signal to be processed can be extracted and inferred through the characteristics of the time dimension and the frequency dimension, a complex signal mask is output, and then the complex signal mask is multiplied by the complex frequency spectrum of the voice signal to be processed to obtain an enhanced signal, so that the voice processing effect is improved.
The following describes another speech processing method in the embodiment of the present application with reference to fig. 4, and fig. 4 is a schematic flow chart of another speech processing method provided in the embodiment of the present application, specifically including the following steps:
step 401, acquiring a training speech signal and a clean speech signal.
The clean voice signal is a voice signal which meets the quality requirement after the training voice signal is enhanced, and it can be understood that the training voice signal is a voice signal with noise, and the clean voice signal is a clean voice signal corresponding to the training voice signal.
Step 402, preprocessing the training voice signal to obtain training first time-frequency signal characteristic data.
The training first time-frequency signal characteristic data may be complex spectrum, amplitude spectrum, and the like, which is not particularly limited herein.
In one possible embodiment, the training speech signal may be framed, windowed, and fourier transformed to obtain training first time-frequency signal characteristic data.
Step 403, inputting the training first time-frequency signal characteristic data into a preset voice processing model, and obtaining a result voice signal according to the output of the preset voice processing model.
The first time-frequency signal feature data for training can be subjected to feature extraction through the preset voice processing model to obtain first feature data for training, the first feature data for training can be subjected to dimension conversion through the preset voice processing model to obtain second feature data for training, the second feature data for training can be subjected to decoding processing through the preset voice processing model to obtain second time-frequency signal feature data for training, and finally, the second time-frequency signal feature data for training can be subjected to voice synthesis to obtain the result voice signal.
Specifically, the preset speech processing model may include an encoder, a dual-path modeling module, and a decoder, and the specific model input/output process may refer to the input/output process of the trained speech processing model described in fig. 3, which is not described herein.
Step 404, pre-processing the result voice signal for a preset number of times to obtain the result time-frequency signal characteristic data of a preset number, and pre-processing the clean voice signal for the preset number of times to obtain the reference time-frequency signal characteristic data of the preset number.
The value of the preset number is the same as the value of the preset times, the dimensionality between the characteristic data of each result time-frequency signal is different, the dimensionality between the characteristic data of each reference time-frequency signal is different, the characteristic data of the result time-frequency signal can be complex spectrum, amplitude spectrum and the like, the characteristic data of the reference time-frequency signal can correspond to the characteristic data of the result time-frequency signal and is complex spectrum, amplitude spectrum and the like, and the characteristic data of the reference time-frequency signal is not particularly limited herein.
The method comprises the steps of carrying out framing, windowing and fourier transformation on the result voice signal for the preset times to obtain the result time-frequency signal characteristic data of the preset number, and carrying out framing, windowing and fourier transformation on the clean voice signal for the preset times to obtain the reference time-frequency signal characteristic data of the preset number, wherein the coefficient groups of framing, windowing and fourier transformation corresponding to the result voice signal at any time are the same as the coefficient groups of framing, windowing and fourier transformation corresponding to the clean voice signal at the time. For example, the coefficient sets required for framing, windowing and fourier transforming the resultant speech signal for the first time are identical to the coefficient sets required for framing, windowing and fourier transforming the clean speech signal for the first time, and the coefficient sets required for framing, windowing and fourier transforming the resultant speech signal for the second time are identical to the coefficient sets required for framing, windowing and fourier transforming the clean speech signal for the second time, and so on, which will not be described in detail herein.
It should be noted that the time domain length of the resulting speech signal needs to be consistent with the time domain length of the clean speech signal, and meanwhile, when the short-time fourier transform of different dimensions is performed, the short frame and the long frame can be covered when parameters are set, and because the time length of the short frame is small, rapid amplitude change and clear speech information can be captured, and the time length of the long frame is large, so that the method has better frequency spectrum resolution and richer harmonic information. Therefore, when the parameter setting is proper, the multi-dimensional loss data can supplement information to the network, and the complementary effect is achieved.
Step 405, determining a target loss fraction according to the result time-frequency signal characteristic data in each dimension and the reference time-frequency signal characteristic data in each dimension.
The loss score in each dimension may be determined, where the loss score in each dimension is used to reflect a gap between the result time-frequency signal feature data in each dimension and the reference time-frequency signal feature data in the dimension, and then the loss score in each dimension is combined to obtain the target loss score.
In one possible embodiment, the merging process may be to weight sum the loss scores in each dimension to obtain the target loss score.
In one possible embodiment, the merging process may be to average the loss scores in each dimension to obtain a target loss score.
It is understood that the target loss fraction may be a mean square error or an average absolute error of the magnitude spectrum, or may be a dynamic compression loss combining magnitude and phase, so long as the difference between the resulting time-frequency signal characteristic data in each dimension and the reference time-frequency signal characteristic data in the dimension can be reflected, and may be a multi-dimensional frequency domain loss and/or a multi-dimensional time domain loss.
Step 406, adjusting the preset speech processing model according to the target loss score until the target loss score is lower than a preset score threshold value, so as to obtain a trained speech processing model.
The target loss score may be fed back to the preset speech processing model, parameters of the preset speech processing model are adjusted, and steps 401-406 are repeatedly performed until the target loss score is lower than a preset score threshold or converges, so as to obtain a trained speech processing model, which is not described herein.
Therefore, the multi-dimensional loss is calculated, so that the parameter adjustment can be performed on the preset voice processing model by utilizing the multi-dimensional loss, and the voice processing effect of the model is improved.
Step 407, pre-processing the first voice signal to obtain first time-frequency signal characteristic data.
And step 408, performing feature extraction on the first time-frequency signal feature data through the trained voice processing model to obtain first feature data.
And 409, performing dimension conversion on the first feature data through the trained voice processing model to obtain second feature data.
And 410, performing decoding processing on the second characteristic data through the trained voice processing model to obtain second time-frequency signal characteristic data.
In step 411, the second time-frequency signal feature data is synthesized to obtain a second speech signal.
Therefore, through the voice processing method, a voice processing model with better multi-dimensional loss component effect can be utilized, the characteristics of voice signals are extracted from time dimension and frequency dimension, the voice signals are enhanced, and the voice processing effect is improved.
The above steps not described in detail may be referred to the description of the steps of the method in fig. 3, and are not described in detail herein.
For easy understanding, an exemplary speech processing method in the embodiment of the present application is described below with reference to fig. 5, and fig. 5 is a schematic flow chart of a speech processing method with a combined architecture according to the embodiment of the present application, which specifically includes the following steps:
Firstly, performing signal processing on a voice signal 1, then inputting the voice signal 1 subjected to the signal processing into a voice processing model, wherein the voice processing model comprises an encoder, a dual-path modeling module and a decoder, the voice signal 1 subjected to the signal processing sequentially passes through the encoder, the dual-path modeling module and the decoder, finally, voice synthesis is performed to obtain a voice signal 2, the voice signal 2 is a voice signal after the voice signal 1 is enhanced, then n times of signal processing are performed on the voice signal 2, meanwhile, n times of signal processing are performed on a voice signal 3, the voice signal 3 is a clean voice signal of the voice signal 1, the dimensionality of each time of signal processing is different, but the coefficient groups required by n times of signal processing are the voice signal 2 and the voice signal 3 are the same, then multi-dimensional loss between the voice signal 2 and the voice signal 3 can be calculated, and the multi-dimensional loss is fed back to the voice processing model to perform parameter adjustment.
Therefore, through the voice processing method, a voice processing model with better multi-dimensional loss component effect can be utilized, the characteristics of voice signals are extracted from time dimension and frequency dimension, the voice signals are enhanced, and the voice processing effect is improved.
The steps not described in detail above may refer to the descriptions of the steps of the methods in fig. 3 and fig. 4, and are not described herein.
The foregoing description of the embodiments of the present application has been presented primarily in terms of a method-side implementation. It will be appreciated that the electronic device, in order to achieve the above-described functions, includes corresponding hardware structures and/or software modules that perform the respective functions. Those of skill in the art will readily appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as hardware or combinations of hardware and computer software. Whether a function is implemented as hardware or computer software driven hardware depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.
The embodiment of the application can divide the functional units of the electronic device according to the method example, for example, each functional unit can be divided corresponding to each function, and two or more functions can be integrated in one processing unit. The integrated units may be implemented in hardware or in software functional units. It should be noted that, in the embodiment of the present application, the division of the units is schematic, which is merely a logic function division, and other division manners may be implemented in actual practice.
In the case of dividing each functional module by adopting the corresponding function, fig. 6 is a block diagram of functional units of a speech processing device according to an embodiment of the present application, where the speech processing device 600 includes:
a preprocessing unit 610, configured to preprocess the first speech signal to obtain first time-frequency signal feature data;
a voice model unit 620, configured to perform feature extraction on the first time-frequency signal feature data through a trained voice processing model to obtain first feature data, where the first feature data includes features of a frequency dimension and features of a time dimension; performing dimension conversion on the first characteristic data through the trained voice processing model to obtain second characteristic data; decoding the second characteristic data through the trained voice processing model to obtain second time-frequency signal characteristic data, wherein the dimension of the second time-frequency signal characteristic data is the same as that of the first time-frequency signal characteristic data;
the speech synthesis unit 630 is configured to perform speech synthesis on the second time-frequency signal feature data to obtain a second speech signal, where the second speech signal is the speech signal after the first speech signal is enhanced.
It can be seen that, through the above-mentioned voice processing method and related apparatus, first, a first voice signal is preprocessed to obtain first time-frequency signal feature data; then, carrying out feature extraction on the first time-frequency signal feature data through a trained voice processing model to obtain first feature data, wherein the first feature data comprises features of frequency dimension and features of time dimension; performing dimension conversion on the first characteristic data through the trained voice processing model to obtain second characteristic data; decoding the second characteristic data through the trained voice processing model to obtain second time-frequency signal characteristic data, wherein the dimension of the second time-frequency signal characteristic data is the same as that of the first time-frequency signal characteristic data; finally, the second time-frequency signal characteristic data is subjected to voice synthesis to obtain a second voice signal, wherein the second voice signal is the voice signal after the first voice signal is enhanced. The information of the voice signal can not be omitted, and the accuracy of voice processing is improved.
It should be noted that, the specific implementation of each operation may be described in the above-illustrated method embodiment, and the speech processing device 600 may be used to execute the above-described method embodiment of the present application, which is not described herein.
In the case of using integrated units, another speech processing apparatus 700 in the embodiment of the present application will be described in detail below with reference to fig. 7, where the speech processing apparatus 700 includes a processing unit 701 and a communication unit 702, where the processing unit 701 is configured to perform any step in the foregoing method embodiment, and when performing data transmission such as sending, the communication unit 702 is selectively invoked to complete a corresponding operation.
The speech processing device 700 may further comprise a storage unit 703 for storing program code and data. The processing unit 701 may be a processor, the communication unit 702 may be a wireless communication module, the storage unit 703 may be a memory, and the processing unit 701 is specifically configured to:
preprocessing the first voice signal to obtain first time-frequency signal characteristic data;
performing feature extraction on the first time-frequency signal feature data through a trained voice processing model to obtain first feature data, wherein the first feature data comprises features of frequency dimension and features of time dimension;
performing dimension conversion on the first characteristic data through the trained voice processing model to obtain second characteristic data;
Decoding the second characteristic data through the trained voice processing model to obtain second time-frequency signal characteristic data, wherein the dimension of the second time-frequency signal characteristic data is the same as that of the first time-frequency signal characteristic data;
and performing voice synthesis on the second time-frequency signal characteristic data to obtain a second voice signal, wherein the second voice signal is the voice signal after the first voice signal is enhanced.
It can be seen that, through the above-mentioned voice processing method and related apparatus, first, a first voice signal is preprocessed to obtain first time-frequency signal feature data; then, carrying out feature extraction on the first time-frequency signal feature data through a trained voice processing model to obtain first feature data, wherein the first feature data comprises features of frequency dimension and features of time dimension; performing dimension conversion on the first characteristic data through the trained voice processing model to obtain second characteristic data; decoding the second characteristic data through the trained voice processing model to obtain second time-frequency signal characteristic data, wherein the dimension of the second time-frequency signal characteristic data is the same as that of the first time-frequency signal characteristic data; finally, the second time-frequency signal characteristic data is subjected to voice synthesis to obtain a second voice signal, wherein the second voice signal is the voice signal after the first voice signal is enhanced. The information of the voice signal can not be omitted, and the accuracy of voice processing is improved.
It should be noted that, the specific implementation of each operation may be described in the above-illustrated method embodiment, and the speech processing device 700 may be used to execute the above-illustrated method embodiment of the present application, which is not described herein.
The embodiment of the application also provides a chip which comprises a processor, a memory and a computer program or instructions stored on the memory, wherein the processor executes the computer program or instructions to realize the steps described in the embodiment of the method.
The embodiment of the application also provides a chip module, which comprises a receiving and transmitting assembly and a chip, wherein the chip comprises a processor, a memory and a computer program or instructions stored on the memory, and the processor executes the computer program or instructions to realize the steps described in the embodiment of the method.
The embodiment of the application also provides a computer storage medium, wherein the computer storage medium stores a computer program for electronic data exchange, and the computer program makes a computer execute part or all of the steps of any one of the method embodiments, and the computer includes an electronic device.
Embodiments of the present application also provide a computer program product comprising a non-transitory computer readable storage medium storing a computer program operable to cause a computer to perform part or all of the steps of any one of the methods described in the method embodiments above. The computer program product may be a software installation package, said computer comprising an electronic device.
For the above embodiments, for simplicity of description, the same is denoted as a series of combinations of actions. It will be appreciated by persons skilled in the art that the application is not limited by the order of acts described, as some steps in embodiments of the application may be performed in other orders or concurrently. In addition, those skilled in the art will appreciate that the embodiments described in the specification are presently preferred embodiments, and that the acts, steps, modules, or units, etc. that are described are not necessarily required by the embodiments of the application.
In the foregoing embodiments, the descriptions of the embodiments of the present application are emphasized, and in part, not described in detail in one embodiment, reference may be made to related descriptions of other embodiments.
The steps of a method or algorithm described in connection with the embodiments disclosed herein may be embodied in hardware, or may be embodied in software instructions executed by a processor. The software instructions may be comprised of corresponding software modules that may be stored in RAM, flash memory, ROM, EPROM, electrically Erasable EPROM (EEPROM), registers, hard disk, a removable disk, a compact disk read-only (CD-ROM), or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor such the processor can read information from, and write information to, the storage medium. In the alternative, the storage medium may be integral to the processor. The processor and the storage medium may reside in an ASIC. In addition, the ASIC may be located in a terminal device or a management device. The processor and the storage medium may reside as discrete components in a terminal device or management device.
Those skilled in the art will appreciate that in one or more of the examples described above, the functions described in the embodiments of the present application may be implemented, in whole or in part, in software, hardware, firmware, or any combination thereof. When implemented in software, may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When loaded and executed on a computer, produces a flow or function in accordance with embodiments of the present application, in whole or in part. The computer may be a general purpose computer, a special purpose computer, a computer network, or other programmable apparatus. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by a wired (e.g., coaxial cable, fiber optic, digital subscriber line (digital subscriber line, DSL)), or wireless (e.g., infrared, wireless, microwave, etc.). The computer readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. that contains an integration of one or more available media. The usable medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a magnetic tape), an optical medium (e.g., a digital video disc (digital video disc, DVD)), or a semiconductor medium (e.g., a Solid State Disk (SSD)), or the like.
The respective apparatuses and the respective modules/units included in the products described in the above embodiments may be software modules/units, may be hardware modules/units, or may be partly software modules/units, and partly hardware modules/units. For example, for each device or product applied to or integrated on a chip, each module/unit included in the device or product may be implemented in hardware such as a circuit, or at least part of the modules/units may be implemented in software program, where the software program runs on a processor integrated inside the chip, and the rest (if any) of the modules/units may be implemented in hardware such as a circuit; for each device and product applied to or integrated in the chip module, each module/unit contained in the device and product can be realized in a hardware manner such as a circuit, different modules/units can be located in the same component (such as a chip, a circuit module and the like) or different components of the chip module, or at least part of the modules/units can be realized in a software program, the software program runs on a processor integrated in the chip module, and the rest (if any) of the modules/units can be realized in a hardware manner such as a circuit; for each device, product, or application to or integrated with the terminal device, each module/unit included in the device may be implemented in hardware such as a circuit, and different modules/units may be located in the same component (e.g., a chip, a circuit module, etc.) or different components in the terminal device, or at least some modules/units may be implemented in a software program, where the software program runs on a processor integrated within the terminal device, and the remaining (if any) part of the modules/units may be implemented in hardware such as a circuit.
The foregoing detailed description of the embodiments of the present application further illustrates the purposes, technical solutions and advantageous effects of the embodiments of the present application, and it should be understood that the foregoing description is only a specific implementation of the embodiments of the present application, and is not intended to limit the scope of the embodiments of the present application, and any modifications, equivalent substitutions, improvements, etc. made on the basis of the technical solutions of the embodiments of the present application should be included in the scope of the embodiments of the present application.
Claims (12)
1. A method of speech processing, the method comprising:
preprocessing the first voice signal to obtain first time-frequency signal characteristic data;
performing feature extraction on the first time-frequency signal feature data through a trained voice processing model to obtain first feature data, wherein the first feature data comprises features of frequency dimension and features of time dimension;
performing dimension conversion on the first characteristic data through the trained voice processing model to obtain second characteristic data;
decoding the second characteristic data through the trained voice processing model to obtain second time-frequency signal characteristic data, wherein the dimension of the second time-frequency signal characteristic data is the same as that of the first time-frequency signal characteristic data;
And performing voice synthesis on the second time-frequency signal characteristic data to obtain a second voice signal, wherein the second voice signal is the voice signal after the first voice signal is enhanced.
2. The method of claim 1, wherein prior to preprocessing the first speech signal to obtain the first time-frequency signal characteristic data, the method further comprises:
acquiring a training voice signal and a clean voice signal, wherein the clean voice signal is a voice signal which meets the quality requirement after the training voice signal is enhanced;
preprocessing the training voice signal to obtain training first time-frequency signal characteristic data;
inputting the training first time-frequency signal characteristic data into a preset voice processing model, and obtaining a result voice signal according to the output of the preset voice processing model;
preprocessing the result voice signal for a preset number of times to obtain result time-frequency signal characteristic data of a preset number, preprocessing the clean voice signal for the preset number of times to obtain reference time-frequency signal characteristic data of the preset number, wherein the value of the preset number is the same as that of the preset number of times, the dimensionality between each result time-frequency signal characteristic data is different, and the dimensionality between each reference time-frequency signal characteristic data is different;
Determining a target loss fraction according to the result time-frequency signal characteristic data in each dimension and the reference time-frequency signal characteristic data in each dimension;
and adjusting the preset voice processing model according to the target loss fraction until the target loss fraction is lower than a preset fraction threshold value so as to obtain the trained voice processing model.
3. The method according to claim 2, wherein inputting the training first time-frequency signal characteristic data into a preset speech processing model, and obtaining a resultant speech signal according to an output of the preset speech processing model, comprises:
performing feature extraction on the training first time-frequency signal feature data through the preset voice processing model to obtain training first feature data;
performing dimension conversion on the first feature data for training through the preset voice processing model to obtain second feature data for training;
decoding the training second characteristic data through the preset voice processing model to obtain training second time-frequency signal characteristic data;
and performing voice synthesis on the training second time-frequency signal characteristic data to obtain the result voice signal.
4. The method according to claim 2, wherein the preprocessing the resultant speech signal for a preset number of times to obtain a preset number of resultant time-frequency signal characteristic data, and the preprocessing the clean speech signal for the preset number of times to obtain the preset number of reference time-frequency signal characteristic data, comprises:
and carrying out framing, windowing and Fourier transformation on the result voice signals for the preset times to obtain the result time-frequency signal characteristic data of the preset number, and carrying out framing, windowing and Fourier transformation on the clean voice signals for the preset times to obtain the reference time-frequency signal characteristic data of the preset number, wherein the coefficient groups of framing, windowing and Fourier transformation corresponding to the result voice signals at any time are the same as the coefficient groups of framing, windowing and Fourier transformation corresponding to the clean voice signals at the same time.
5. The method of claim 2, wherein said determining a target loss fraction from the resulting time-frequency signal characteristic data in each dimension and the reference time-frequency signal characteristic data in each dimension comprises:
Determining a loss fraction in each dimension, wherein the loss fraction in each dimension is used for reflecting the difference between the result time-frequency signal characteristic data in each dimension and the reference time-frequency signal characteristic data in the dimension;
and combining the loss fractions in each dimension to obtain the target loss fraction.
6. The method of claim 1, wherein dimension converting the first feature data by the trained speech processing model to obtain second feature data comprises:
performing first dimension conversion on the first feature data through the trained voice processing model to obtain first conversion features;
and performing second dimension conversion on the first conversion feature through the trained voice processing model to obtain second feature data, wherein the first dimension conversion is used for indicating frequency dimension conversion or time dimension conversion, the second dimension conversion is used for indicating frequency dimension conversion or time dimension conversion, and the first dimension conversion is different from the second dimension conversion.
7. The method of claim 1, wherein decoding the second feature data with the trained speech processing model to obtain second time-frequency signal feature data comprises:
Performing dimension recovery on the second characteristic data through the trained voice processing model to obtain first mask data;
and multiplying the first mask data and the first time-frequency signal characteristic data by the trained voice processing model to obtain the second time-frequency signal characteristic data.
8. A speech processing apparatus, the apparatus comprising:
the preprocessing unit is used for preprocessing the first voice signal to obtain first time-frequency signal characteristic data;
the voice model unit is used for carrying out feature extraction on the first time-frequency signal feature data through a trained voice processing model to obtain first feature data, wherein the first feature data comprises features of frequency dimension and features of time dimension; performing dimension conversion on the first characteristic data through the trained voice processing model to obtain second characteristic data; decoding the second characteristic data through the trained voice processing model to obtain second time-frequency signal characteristic data, wherein the dimension of the second time-frequency signal characteristic data is the same as that of the first time-frequency signal characteristic data;
And the voice synthesis unit is used for performing voice synthesis on the second time-frequency signal characteristic data to obtain a second voice signal, wherein the second voice signal is the voice signal after the first voice signal is enhanced.
9. An electronic device, comprising: a processor, a memory, and one or more programs; the one or more programs are stored in the memory and configured to be executed by the processor, the programs comprising instructions for performing the steps in the method of any of claims 1-7.
10. A chip comprising a processor, a memory and a computer program or instructions stored on the memory, wherein the processor executes the computer program or instructions to carry out the steps of the method of any one of claims 1-7.
11. A chip module comprising a transceiver component and a chip comprising a processor, a memory and a computer program or instructions stored on the memory, wherein the processor executes the computer program or instructions to carry out the steps of the method of any one of claims 1-7.
12. A computer storage medium storing a computer program comprising program instructions which, when executed by a processor, cause the processor to perform the method of any of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311086762.3A CN117059114A (en) | 2023-08-25 | 2023-08-25 | Voice processing method and related device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311086762.3A CN117059114A (en) | 2023-08-25 | 2023-08-25 | Voice processing method and related device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117059114A true CN117059114A (en) | 2023-11-14 |
Family
ID=88653352
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311086762.3A Pending CN117059114A (en) | 2023-08-25 | 2023-08-25 | Voice processing method and related device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117059114A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN119743334A (en) * | 2025-03-06 | 2025-04-01 | 北京国都互联科技有限公司 | Data encryption transmission method and system |
-
2023
- 2023-08-25 CN CN202311086762.3A patent/CN117059114A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN119743334A (en) * | 2025-03-06 | 2025-04-01 | 北京国都互联科技有限公司 | Data encryption transmission method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112562691B (en) | Voiceprint recognition method, voiceprint recognition device, computer equipment and storage medium | |
| WO2022126924A1 (en) | Training method and apparatus for speech conversion model based on domain separation | |
| KR20190110939A (en) | Environment sound recognition method based on convolutional neural networks, and system thereof | |
| CN111247585A (en) | Voice conversion method, device, equipment and storage medium | |
| US20210193149A1 (en) | Method, apparatus and device for voiceprint recognition, and medium | |
| CN112767927A (en) | Method, device, terminal and storage medium for extracting voice features | |
| WO2024055752A9 (en) | Speech synthesis model training method, speech synthesis method, and related apparatuses | |
| CN115602165B (en) | Digital employee intelligent system based on financial system | |
| CN113421584B (en) | Audio noise reduction method, device, computer equipment and storage medium | |
| CN116030792A (en) | Method, device, electronic device and readable medium for converting speech timbre | |
| CN106653056A (en) | Fundamental frequency extraction model based on LSTM recurrent neural network and training method thereof | |
| CN119314499B (en) | Audio signal processing method, device, computer equipment and storage medium | |
| CN117672254A (en) | Voice conversion method, device, computer equipment and storage medium | |
| CN117059114A (en) | Voice processing method and related device | |
| CN117238277A (en) | Intent recognition method, device, storage medium and computer equipment | |
| CN116913304A (en) | Real-time voice stream noise reduction method, device, computer equipment and storage medium | |
| CN119229889B (en) | A method and device for speech denoising under low signal-to-noise ratio | |
| CN111933172B (en) | Human voice separation and extraction method and device, computer equipment and storage medium | |
| CN119580749A (en) | Speech signal reconstruction method, device, equipment and storage medium | |
| CN120220693A (en) | Voiceprint recognition method, device, electronic device and storage medium | |
| Zheng et al. | Bandwidth extension WaveNet for bone-conducted speech enhancement | |
| CN119207381A (en) | Training method, device, equipment and medium for speech conversion model | |
| CN118447820A (en) | Voice conversion method, device, equipment and medium based on style | |
| CN110875037A (en) | Voice data processing method and device and electronic equipment | |
| JP2019132948A (en) | Voice conversion model learning device, voice conversion device, method, and program |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |