A feature extraction method, device, electronic device and storage medium

Abstract

The present application relates to a feature extraction method, device, electronic device and storage medium, which are applied to speech with a sampling rate of 8K to 16K. The method includes: in the feature extraction process, obtaining the multidimensional MFCC feature energy value of the high-frequency part; judging whether the multidimensional MFCC feature energy value of the high-frequency part satisfies the feature of the zeroing condition; if the judgment result is that the multidimensional MFCC feature energy value of the high-frequency part satisfies the feature of the zeroing condition, then setting the dimension corresponding to the feature energy value to a marked position that may be set to zero. The present application compares the feature differences of different sampling rates, sets the last 7 bits of the 40-dimensional part representing the high-frequency part to zero, thereby optimizing the feature extraction method to compensate for the loss of performance. By optimizing the feature extraction method of the upsampled data, the converted feature expression is closer to the trained speech data, the model performance is improved, the feature compatibility and generalization are reduced, the requirements for data are relatively reduced, and the cost of data collection and processing is reduced.

Description

Technical Field

[0001] The present application relates to the technical field of speech feature extraction, and in particular to a feature extraction method, device, electronic device and storage medium. Background Art

[0002] At present, mainstream speech recognition systems theoretically support speech signals with multiple different sampling rates, but most of the acoustic models of mainstream speech systems only support signals with sampling rates of 8k (8000) and 16k (16000). For externally transmitted speech information that is not these two, the speech is often upsampled or downsampled through some audio conversion methods. For example, 48k speech is downsampled to 16k, and 8k speech is upsampled to 16k. The advantage of this is that it can save system overhead. If the conversion method is not adopted, theoretically, models with different sampling rates are required to recognize speech with different sampling rates. On the other hand, the method of selecting only one or two speech for acoustic model training reduces the cost of data collection. There is no need to collect speech with various sampling rates, but only needs to collect speech with a certain sampling rate. However, the problem with this method is that when the test speech (commonly 16k or 8k) matches the acoustic model training data, the recognition performance is good, but when the speech does not match the training data, the performance is relatively poor. For example, using speech upsampled from 8k to 16k to test the 16k model is worse than the 8k model. Summary of the invention

[0003] Based on the above problems, the present application provides a feature extraction method, device, electronic device and storage medium.

[0004] In a first aspect, an embodiment of the present application provides a feature extraction method, which is applied to speech with a sampling rate of 8K to 16K, including:

[0005] In the feature extraction process, the multi-dimensional MFCC feature energy value of the high-frequency part is obtained;

[0006] A feature for judging whether the multi-dimensional MFCC feature energy value of the high-frequency part meets the zeroing condition;

[0007] If the judgment result is that the multi-dimensional MFCC feature energy value of the high-frequency part satisfies the zeroing condition, the dimension corresponding to the feature energy value is set to the mark position that may be set to zero.

[0008] Furthermore, in the above-mentioned feature extraction method, the feature of judging whether the multi-dimensional MFCC feature energy value of the high-frequency part satisfies the zeroing condition includes:

[0009] Compare the multi-dimensional MFCC feature energy value of the high-frequency part with the multi-dimensional preset threshold;

[0010] If the feature energy value is less than the preset threshold, the feature meets the zeroing condition;

[0011] If the feature energy value is greater than or equal to the preset threshold, the feature does not meet the zeroing condition.

[0012] Furthermore, in the above-mentioned feature extraction method, the dimension of the high-frequency part is 7 dimensions.

[0013] Furthermore, the above feature extraction method further includes:

[0014] If the characteristic energy value of the current dimension is less than the preset threshold of the current dimension, and the characteristic energy value of the 6th dimension is less than the preset threshold of the 6th dimension, and the characteristic energy value of the 7th dimension is less than the preset threshold of the 7th dimension, then the marked position of the current dimension is set to zero.

[0015] Furthermore, the above feature extraction method further includes:

[0016] If the characteristic energy value of 1 dimension is less than the preset threshold of 1 dimension, and the characteristic energy value of the 6th dimension is less than the preset threshold of the 6th dimension, and the characteristic energy value of the 7th dimension is less than the preset threshold of the 7th dimension, then the marked position of 1 dimension is set to zero;

[0017] If the characteristic energy value of the 2nd dimension is less than the preset threshold of the 2nd dimension, and the characteristic energy value of the 6th dimension is less than the preset threshold of the 6th dimension, and the characteristic energy value of the 7th dimension is less than the preset threshold of the 7th dimension, then the marked position of the 2nd dimension is set to zero;

[0018] If the characteristic energy value of the 3rd dimension is less than the preset threshold of the 3rd dimension, and the characteristic energy value of the 6th dimension is less than the preset threshold of the 6th dimension, and the characteristic energy value of the 7th dimension is less than the preset threshold of the 7th dimension, then the marked position of the 3rd dimension is set to zero;

[0019] If the characteristic energy value of the 4th dimension is less than the preset threshold of the 4th dimension, and the characteristic energy value of the 6th dimension is less than the preset threshold of the 6th dimension, and the characteristic energy value of the 7th dimension is less than the preset threshold of the 7th dimension, then the marked position of the 4th dimension is set to zero;

[0020] If the characteristic energy value of the 5th dimension is less than the preset threshold of the 5th dimension, and the characteristic energy value of the 6th dimension is less than the preset threshold of the 6th dimension, and the characteristic energy value of the 7th dimension is less than the preset threshold of the 7th dimension, then the marked position of the 5th dimension is set to zero;

[0021] Furthermore, the above feature extraction method further includes:

[0022] A feature for determining whether the multi-dimensional MFCC feature energy value of the high frequency part of the current frame satisfies the zeroing condition;

[0023] If the judgment result is that the multidimensional MFCC feature energy value of the high-frequency part satisfies the zeroing condition, the cumulative number of frames is increased by 1; if the judgment result is that the multidimensional MFCC feature energy value of the high-frequency part does not satisfy the zeroing condition, the cumulative number of frames is reduced by 1;

[0024] Determine whether the accumulated number of frames meets the frame number condition;

[0025] If the accumulated number of frames meets the frame number condition and the multi-dimensional MFCC feature energy value of the high-frequency part of the current frame meets the feature of the zeroing condition, the dimension corresponding to the feature of the current frame that meets the zeroing condition is set to zero.

[0026] Furthermore, in the above-mentioned feature extraction method, judging whether the accumulated number of frames meets the frame number condition includes:

[0027] Determine the size of the accumulated frame number and the preset frame number;

[0028] If the accumulated number of frames is greater than or equal to the preset number of frames, the frame number condition is met;

[0029] If the accumulated frame number is less than the preset frame number, the frame number condition is not met.

[0030] In a second aspect, an embodiment of the present application further provides a feature extraction device, which is applied to speech with a sampling rate of 8K to 16K, including:

[0031] Acquisition module: used to obtain the multi-dimensional MFCC feature energy value of the high-frequency part during the feature extraction process;

[0032] Judgment module: used to judge whether the multi-dimensional MFCC feature energy value of the high-frequency part meets the zeroing condition;

[0033] Setting module: used for setting the dimension corresponding to the feature energy value to a mark position that may be set to zero if the judgment result is that the feature energy value of the multi-dimensional MFCC feature of the high-frequency part meets the zeroing condition.

[0034] In a third aspect, an embodiment of the present application further provides an electronic device, characterized in that it includes: a processor and a memory;

[0035] The processor is used to execute the above-mentioned feature extraction method by calling the program or instruction stored in the memory.

[0036] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, characterized in that the computer-readable storage medium stores a program or instruction, and the program or instruction enables the computer to perform the above-mentioned feature extraction method.

[0037] The advantages of the embodiments of the present application are: the present application relates to a feature extraction method, device, electronic device and storage medium, which are applied to speech with a sampling rate of 8K to 16K, and the method includes: in the feature extraction process, obtaining the multidimensional MFCC feature energy value of the high-frequency part; judging whether the multidimensional MFCC feature energy value of the high-frequency part satisfies the feature of the zeroing condition; if the judgment result is that the multidimensional MFCC feature energy value of the high-frequency part satisfies the feature of the zeroing condition, then the dimension corresponding to the feature energy value is set to a marked position that may be set to zero. The present application compares the feature differences of different sampling rates, sets the partial values ​​of the last 7 bits representing the high-frequency part in 40 dimensions to zero, thereby optimizing the feature extraction method to compensate for the loss of performance, and by optimizing the feature extraction method of the up-sampled data, the converted feature expression is closer to the trained speech data, the model performance is improved, the compatibility and generalization of the features are relatively reduced, and the requirements for data are reduced, reducing the cost of data collection and processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the conventional technology, the drawings required for use in the embodiments or the conventional technology descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0039] Figure 1 A feature extraction method provided in the embodiment of the present application is shown in FIG. Figure 1 ;

[0040] Figure 2 A feature extraction method provided in the embodiment of the present application is shown in FIG. Figure 2 ;

[0041] Figure 3 A feature extraction method provided in the embodiment of the present application is shown in FIG. Figure 3 ;

[0042] Figure 4 A schematic diagram of a feature extraction device provided in an embodiment of the present application;

[0043] Figure 5 It is a schematic block diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0044] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present application. However, the present application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present application, so the present application is not limited by the specific implementation disclosed below.

[0045] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application. The term "and / or" used herein includes any and all combinations of one or more of the related listed items.

[0046] Figure 1 A feature extraction method provided in the embodiment of the present application is shown in FIG. Figure 1 .

[0047] In the first aspect, the present application embodiment provides a feature extraction method, which is applied to speech with a sampling rate of 8K to 16K, combined with Figure 1 , including two steps from S101 to S103:

[0048] S101: In the feature extraction process, a multi-dimensional MFCC feature energy value of a high-frequency part is obtained.

[0049] Specifically, in the embodiment of the present application, the speech applied to the 8K to 16K sampling rate is optimized based on 40-dimensional features. In the feature extraction process, the dimension of the high-frequency part is 7 dimensions, and the multi-dimensional MFCC feature energy value of the high-frequency part is obtained to obtain 7 different feature energy values.

[0050] S102: Determine whether the multi-dimensional MFCC feature energy value of the high-frequency part satisfies the characteristics of the zeroing condition.

[0051] Specifically, in the embodiment of the present application, by comparing 7 different feature energy values ​​with 7 preset thresholds, it is determined whether the 7-dimensional MFCC feature energy value meets the zeroing condition. The detailed steps for determining whether the feature meets the zeroing condition are described in detail below in combination with specific steps.

[0052] S103: If the judgment result is that the multi-dimensional MFCC feature energy value of the high-frequency part satisfies the zeroing condition, the dimension corresponding to the feature energy value is set as a mark position that may be set to zero.

[0053] Specifically, in the embodiment of the present application, by comparing the feature differences at different sampling rates, the values ​​of the last 7 bits representing the high frequency part in the 40 dimensions are set to zero, thereby optimizing the feature extraction method to compensate for the performance loss. By optimizing the feature extraction method of upsampled data, the converted feature expression is made closer to the trained speech data, thereby improving the model performance, and the compatibility and generalization of features. The requirements for data will be relatively reduced, thereby reducing the cost of data collection and processing.

[0054] Figure 2 A feature extraction method provided in the embodiment of the present application is shown in FIG. Figure 2 .

[0055] Furthermore, in the above-mentioned feature extraction method, the feature of judging whether the multi-dimensional MFCC feature energy value of the high-frequency part satisfies the zeroing condition is combined with Figure 2 , including two steps S201 to S202:

[0056] S201: Compare the multi-dimensional MFCC feature energy value of the high-frequency part with the multi-dimensional preset threshold;

[0057] S202: If the feature energy value is less than a preset threshold, the feature meets the zeroing condition; if the feature energy value is greater than or equal to the preset threshold, the feature does not meet the zeroing condition.

[0058] Specifically, in an embodiment of the present application, by comparing 7 different feature energy values ​​with 7 preset thresholds, it is determined whether the 7-dimensional MFCC feature energy value meets the zeroing condition. For example, if the feature energy value of the first dimension is less than the preset threshold of the first dimension, the feature that meets the zeroing condition will have the first dimension marked at zero. If the feature energy value of the first dimension is greater than or equal to the preset threshold of the first dimension, the feature that does not meet the zeroing condition will not have the first dimension marked at zero. The same is true for 2-dimensional, 3-dimensional, etc., and no examples are given here one by one.

[0059] Furthermore, in the above-mentioned feature extraction method, the dimension of the high-frequency part is 7 dimensions.

[0060] Specifically, the present application is for the high-frequency part represented in 40 dimensions, that is, the partial value of the last 7 digits, that is, the dimension of the high-frequency part is 7 dimensions.

[0061] Furthermore, the above feature extraction method further includes:

[0062] If the characteristic energy value of the current dimension is less than the preset threshold of the current dimension, and the characteristic energy value of the 6th dimension is less than the preset threshold of the 6th dimension, and the characteristic energy value of the 7th dimension is less than the preset threshold of the 7th dimension, then the marked position of the current dimension is set to zero.

[0063] Specifically, in an embodiment of the present application, by comparing the characteristic energy value of the current dimension with the size of the preset threshold of the current dimension and combining the characteristic energy values ​​of the last two dimensions with the sizes of the preset thresholds, it is further determined whether the marked position of the current dimension is set to zero, thereby further improving the accuracy of zeroing.

[0064] The detailed steps of further determining whether the label position of the current dimension is set to zero by comparing the feature energy value of the current dimension with the preset threshold of the current dimension and combining the feature energy values ​​of the last two dimensions with the preset threshold are as follows:

[0065] If the characteristic energy value of 1 dimension is less than the preset threshold of 1 dimension, the characteristic energy value of the 6th dimension is less than the preset threshold of the 6th dimension, and the characteristic energy value of the 7th dimension is less than the preset threshold of the 7th dimension, then the marked position of 1 dimension is set to zero.

[0066] If the characteristic energy value of the 2nd dimension is less than the preset threshold of the 2nd dimension, the characteristic energy value of the 6th dimension is less than the preset threshold of the 6th dimension, and the characteristic energy value of the 7th dimension is less than the preset threshold of the 7th dimension, then the marking position of the 2nd dimension is set to zero.

[0067] If the characteristic energy value of the 3rd dimension is less than the preset threshold of the 3rd dimension, and the characteristic energy value of the 6th dimension is less than the preset threshold of the 6th dimension, and the characteristic energy value of the 7th dimension is less than the preset threshold of the 7th dimension, then the marking position of the 3rd dimension is set to zero.

[0068] If the characteristic energy value of the 4th dimension is less than the preset threshold of the 4th dimension, the characteristic energy value of the 6th dimension is less than the preset threshold of the 6th dimension, and the characteristic energy value of the 7th dimension is less than the preset threshold of the 7th dimension, then the marked position of the 4th dimension is set to zero.

[0069] If the characteristic energy value of the 5th dimension is less than the preset threshold of the 5th dimension, the characteristic energy value of the 6th dimension is less than the preset threshold of the 6th dimension, and the characteristic energy value of the 7th dimension is less than the preset threshold of the 7th dimension, then the marked position of the 5th dimension is set to zero.

[0070] Figure 3 A feature extraction method provided in the embodiment of the present application is shown in FIG. Figure 3 .

[0071] Furthermore, the above feature extraction method is combined with Figure 3 , and also includes four steps from S301 to S304:

[0072] S301: Determine whether the multi-dimensional MFCC feature energy value of the high frequency part of the current frame satisfies the zeroing condition.

[0073] S302: If the judgment result is that the energy value of the multi-dimensional MFCC feature of the high-frequency part satisfies the zeroing condition, the accumulated number of frames is increased by 1; if the judgment result is that the energy value of the multi-dimensional MFCC feature of the high-frequency part does not satisfy the zeroing condition, the accumulated number of frames is decreased by 1.

[0074] S303: Determine whether the accumulated number of frames meets the frame number condition.

[0075] S304: If the accumulated number of frames meets the frame number condition and the multi-dimensional MFCC feature energy value of the high-frequency part of the current frame meets the feature of the zeroing condition, the dimension corresponding to the feature of the current frame that meets the zeroing condition is set to zero.

[0076] Specifically, in the embodiment of the present application, the zeroing does not start from the first frame. When the judgment result is that the multi-dimensional MFCC feature energy value of the high-frequency part satisfies the zeroing condition, the cumulative number of frames is added by 1, and the dimension corresponding to the feature that satisfies the zeroing condition of the current frame is set to zero until the cumulative number of frames satisfies the frame number condition, thereby further improving the accuracy of zeroing.

[0077] Furthermore, in the above-mentioned feature extraction method, judging whether the accumulated number of frames meets the frame number condition includes:

[0078] Determine the size of the accumulated frame number and the preset frame number;

[0079] If the accumulated number of frames is greater than or equal to the preset number of frames, the frame number condition is met;

[0080] If the accumulated frame number is less than the preset frame number, the frame number condition is not met.

[0081] Specifically, the preset number of frames in the embodiment of the present application can be 20 frames or 25 frames, which can be flexibly determined according to actual conditions. If it is 20 frames, the cumulative number of frames is also 20 frames. It is further judged whether the multi-dimensional MFCC feature energy value of the high-frequency part of the 21 frames meets the zeroing condition. If the zeroing condition is met, the high-frequency part of the 21 frames is set to zero.

[0082] Figure 4 A schematic diagram of a feature extraction device provided in an embodiment of the present application.

[0083] In a second aspect, the present application also provides a feature extraction device, combined with Figure 4 , applied to speech with a sampling rate of 8K to 16K, including:

[0084] Acquisition module 401: used to acquire the multi-dimensional MFCC feature energy value of the high-frequency part during the feature extraction process.

[0085] Specifically, in the embodiment of the present application, the speech applied to the 8K to 16K sampling rate is optimized based on 40-dimensional features. In the feature extraction process, the dimension of the high-frequency part is 7 dimensions, and the acquisition module 401 obtains the multi-dimensional MFCC feature energy value of the high-frequency part to obtain 7 different feature energy values.

[0086] Judgment module 402: used to judge whether the multi-dimensional MFCC feature energy value of the high-frequency part meets the feature setting condition.

[0087] Specifically, in the embodiment of the present application, the judgment module 402 determines whether the 7-dimensional MFCC feature energy value satisfies the characteristics of the zeroing condition by comparing the sizes of 7 different feature energy values ​​with 7 preset thresholds. The detailed steps for determining whether the characteristics meet the zeroing condition are described in detail below in combination with specific steps.

[0088] Setting module 403 is used for setting the dimension corresponding to the feature energy value as a possible zero-setting mark position if the judgment result is that the feature energy value of the multi-dimensional MFCC feature of the high-frequency part meets the zero setting condition.

[0089] Specifically, in the embodiment of the present application, by comparing the feature differences of different sampling rates, some values ​​representing the high-frequency part in the 40 dimensions are set to zero, that is, the setting module 403 sets some values ​​of the last 7 bits to zero, thereby optimizing the feature extraction method to compensate for the loss performance, and by optimizing the up-sampled data feature extraction method, the converted feature expression is made closer to the trained speech data, thereby improving the model performance, and the compatibility and generalization of the features, and the requirements for data will be relatively reduced, thereby reducing the cost of data collection and processing.

[0090] In a third aspect, an embodiment of the present application further provides an electronic device, characterized in that it includes: a processor and a memory;

[0091] The processor is used to execute the above-mentioned feature extraction method by calling the program or instruction stored in the memory.

[0092] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, characterized in that the computer-readable storage medium stores a program or instruction, and the program or instruction enables the computer to perform the above-mentioned feature extraction method.

[0093] Figure 5 It is a schematic block diagram of an electronic device provided by an embodiment of the present disclosure.

[0094] like Figure 5As shown, the electronic device includes: at least one processor 501, at least one memory 502 and at least one communication interface 503. The various components in the electronic device are coupled together through a bus system 504. The communication interface 503 is used to transmit information between external devices. It can be understood that the bus system 504 is used to achieve connection and communication between these components. In addition to the data bus, the bus system 504 also includes a power bus, a control bus and a status signal bus. However, for the sake of clarity, Figure 5 Various buses are labeled as bus system 504 .

[0095] It can be understood that the memory 502 in this embodiment can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memories.

[0096] In some implementations, the memory 502 stores the following elements, executable units or data structures, or a subset thereof, or an extended set thereof: an operating system and application programs.

[0097] Among them, the operating system includes various system programs, such as the framework layer, the core library layer, the driver layer, etc., which are used to implement various basic services and process hardware-based tasks. The application program includes various application programs, such as a media player (Media Player), a browser (Browser), etc., which are used to implement various application services. A program that implements any of the feature extraction methods provided in the embodiments of the present application can be included in the application program.

[0098] In an embodiment of the present application, the processor 501 calls a program or instruction stored in the memory 502, specifically, a program or instruction stored in an application, and the processor 501 is used to execute the steps of each embodiment of a feature extraction method provided in an embodiment of the present application.

[0099] In the feature extraction process, the multi-dimensional MFCC feature energy value of the high-frequency part is obtained;

[0100] A feature for judging whether the multi-dimensional MFCC feature energy value of the high-frequency part meets the zeroing condition;

[0101] If the judgment result is that the multi-dimensional MFCC feature energy value of the high-frequency part satisfies the zeroing condition, the dimension corresponding to the feature energy value is set to the mark position that may be set to zero.

[0102] Any of the feature extraction methods provided in the embodiment of the present application can be applied to the processor 501, or implemented by the processor 501. The processor 501 can be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in the processor 501 or an instruction in the form of software. The above-mentioned processor 501 can be a general-purpose processor, a digital signal processor (Digital Signal Processor, DSP), an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field programmable gate array (Field Programmable Gate Array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.

[0103] The steps of any method in a feature extraction method provided in an embodiment of the present application can be directly embodied as being executed by a hardware decoding processor, or being executed by a combination of hardware and software units in a decoding processor. The software unit can be located in a storage medium mature in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory 502, and the processor 501 reads the information in the memory 502, and completes the steps of a feature extraction method in combination with its hardware.

[0104] Those skilled in the art will appreciate that although some embodiments described herein include certain features included in other embodiments but not other features, the combination of features from different embodiments is meant to be within the scope of the present application and to form different embodiments.

[0105] Those skilled in the art will appreciate that the description of each embodiment has its own emphasis, and for parts not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0106] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A feature extraction method, characterized in that: Applicable to speech with 8K to 16K sampling rate, including: In the feature extraction process, the multi-dimensional MFCC feature energy value of the high-frequency part is obtained; Determining whether the multidimensional MFCC feature energy value of the high-frequency part satisfies the feature of the zeroing condition, including: comparing the multidimensional MFCC feature energy value of the high-frequency part with the multidimensional preset threshold; if the feature energy value is less than the preset threshold, the feature satisfies the zeroing condition; if the feature energy value is greater than or equal to the preset threshold, the feature does not satisfy the zeroing condition; If the judgment result is that the multi-dimensional MFCC feature energy value of the high-frequency part satisfies the feature of the zeroing condition, the dimension corresponding to the feature energy value is set to a marked position that may be set to zero.

2. A feature extraction method according to claim 1, characterized in that: The dimension of the high frequency part is 7 dimensions.

3. A feature extraction method according to claim 2, characterized in that: The method further comprises: If the characteristic energy value of the current dimension is less than the preset threshold of the current dimension, and the characteristic energy value of the 6th dimension is less than the preset threshold of the 6th dimension, and the characteristic energy value of the 7th dimension is less than the preset threshold of the 7th dimension, then the marked position of the current dimension is set to zero.

4. A feature extraction method according to claim 3, characterized in that: The method comprises: If the characteristic energy value of 1 dimension is less than the preset threshold of 1 dimension, and the characteristic energy value of the 6th dimension is less than the preset threshold of the 6th dimension, and the characteristic energy value of the 7th dimension is less than the preset threshold of the 7th dimension, then the marked position of 1 dimension is set to zero; If the characteristic energy value of the 2nd dimension is less than the preset threshold of the 2nd dimension, and the characteristic energy value of the 6th dimension is less than the preset threshold of the 6th dimension, and the characteristic energy value of the 7th dimension is less than the preset threshold of the 7th dimension, then the marked position of the 2nd dimension is set to zero; If the characteristic energy value of the 3rd dimension is less than the preset threshold of the 3rd dimension, and the characteristic energy value of the 6th dimension is less than the preset threshold of the 6th dimension, and the characteristic energy value of the 7th dimension is less than the preset threshold of the 7th dimension, then the marked position of the 3rd dimension is set to zero; If the characteristic energy value of the 4th dimension is less than the preset threshold of the 4th dimension, and the characteristic energy value of the 6th dimension is less than the preset threshold of the 6th dimension, and the characteristic energy value of the 7th dimension is less than the preset threshold of the 7th dimension, then the marked position of the 4th dimension is set to zero; If the characteristic energy value of the 5th dimension is less than the preset threshold of the 5th dimension, the characteristic energy value of the 6th dimension is less than the preset threshold of the 6th dimension, and the characteristic energy value of the 7th dimension is less than the preset threshold of the 7th dimension, then the marked position of the 5th dimension is set to zero.

5. A feature extraction method according to claim 1, characterized in that: The method further comprises: A feature for determining whether the multi-dimensional MFCC feature energy value of the high frequency part of the current frame satisfies the zeroing condition; If the judgment result is that the multidimensional MFCC feature energy value of the high-frequency part satisfies the zeroing condition, the cumulative number of frames is increased by 1; if the judgment result is that the multidimensional MFCC feature energy value of the high-frequency part does not satisfy the zeroing condition, the cumulative number of frames is reduced by 1; Determining whether the accumulated number of frames meets a frame number condition; If the accumulated number of frames meets the frame number condition and the multi-dimensional MFCC feature energy value of the high-frequency part of the current frame meets the feature of the zeroing condition, the dimension corresponding to the feature of the current frame that meets the zeroing condition is set to zero.

6. A feature extraction method according to claim 5, characterized in that: The determining whether the accumulated number of frames satisfies a frame number condition includes: Determine the size of the accumulated frame number and the preset frame number; If the accumulated frame number is greater than or equal to the preset frame number, the frame number condition is met; If the accumulated frame number is less than the preset frame number, the frame number condition is not met.

7. A feature extraction device, characterized in that: Applicable to speech with 8K to 16K sampling rate, including: Acquisition module: used to obtain the multi-dimensional MFCC feature energy value of the high-frequency part during the feature extraction process; A judgment module is used to judge whether the multidimensional MFCC feature energy value of the high-frequency part satisfies the feature of the zeroing condition, including: comparing the multidimensional MFCC feature energy value of the high-frequency part with the multidimensional preset threshold; if the feature energy value is less than the preset threshold, the feature satisfies the zeroing condition; if the feature energy value is greater than or equal to the preset threshold, the feature does not satisfy the zeroing condition; Setting module: used for setting the dimension corresponding to the feature energy value to a mark position that may be set to zero if the judgment result is that the multi-dimensional MFCC feature energy value of the high-frequency part satisfies the zeroing condition.

8. An electronic device, characterized in that: include: Processor and memory; The processor is used to execute a feature extraction method as described in any one of claims 1 to 6 by calling the program or instruction stored in the memory.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a program or instruction, and the program or instruction enables a computer to execute a feature extraction method according to any one of claims 1 to 6.

Images