WO2021051572A1 - Voice recognition method and apparatus, and computer device - Google Patents

Abstract

A voice recognition method and apparatus, a computer device and a non-volatile computer-readable storage medium. Said method comprises: acquiring a voice segment, and framing the voice segment to obtain each frame of voice data (S500); sequentially windowing, according to a preset stable windowing algorithm, each frame of voice data of the voice segment, so as to obtain a windowed voice frame of the voice segment (S502); extracting a Mel frequency cepstral coefficient vector (MFCC) of the windowed voice frame of the voice segment (S504); calculating the distance between the MFCC and a voiceprint discrimination vector (S506); and when the distance is smaller than a preset threshold, determining that the recognition result of the voice segment is passed (S508). The voice recognition method can calculate a characteristic vector in a voice segment more accurately, thereby improving the accuracy of voice recognition.

Description

Speech recognition method, device and computer equipment

This application claims the priority of a Chinese patent application filed with the Chinese Patent Office, the application number is 201910871726.5, and the invention title is "speech recognition method, device and computer equipment" on September 16, 2019, the entire content of which is incorporated into this application by reference in.

Technical field

This application relates to the field of speech recognition technology, and in particular to a speech recognition method, device, computer equipment, and non-volatile computer-readable storage medium.

Background technique

Voice recognition is a type of biometric recognition technology, which is a technology that automatically recognizes the user's identity corresponding to the voice based on the voice parameters that reflect the physiological or behavioral characteristics of the voice in the voice waveform. In the prior art, speech recognition generally uses the voiceprint features in the voice signal for recognition. Among them, in the voiceprint feature extraction stage, the existing windowing process, such as the use of Hanning window, Hamming window, and triangular window , Gaussian window, etc. to add windows to voice data. The inventor realizes that the existing windowing processing methods almost always modify the original speech signal, which causes the loss of part of the voiceprint feature information and reduces the accuracy of speech recognition.

Summary of the invention

In view of this, this application proposes a speech recognition method, device, computer equipment, and non-volatile computer-readable storage medium, which can obtain speech fragments and then divide into frames to obtain each frame of speech data, and then smoothly add according to a preset The windowing algorithm windows each frame of speech data to obtain a windowed speech frame; then, extract the Mel frequency cepstrum feature vector MFCC of the windowed speech frame of the speech segment, and calculate the MFCC and voiceprint The distance of the discrimination vector; when the distance is less than a preset threshold, it is determined that the recognition result of the speech segment is passed. Through the above method, the feature vector in the speech segment can be calculated more accurately with a small amount of modification to the speech signal, thereby improving the accuracy of speech recognition.

First of all, in order to achieve the above objective, the present application provides a voice recognition method, which includes:

Acquire a voice segment, divide the voice segment into frames to obtain each frame of voice data; according to a preset smooth windowing algorithm, sequentially window each frame of the voice segment of the voice segment to obtain the voice segment Windowed speech frame; extracting the Mel frequency cepstrum feature vector MFCC of the windowed speech frame of the speech segment; calculating the distance between the MFCC and the voiceprint discrimination vector, wherein the voiceprint discrimination vector is the The sampled voice information of the user is input into the voiceprint feature training model for training; when the distance is less than a preset threshold, it is determined that the recognition result of the voice segment is passed.

In addition, in order to achieve the above objective, the present application also provides a voice recognition device, which includes:

The framing module is used to obtain speech fragments, and the speech fragments are divided into frames to obtain each frame of speech data; the windowing module is used to sequentially calculate each frame of the speech fragments according to a preset smooth windowing algorithm The speech data is windowed to obtain the windowed speech frame of the speech segment; the extraction module is used to extract the Mel frequency cepstrum feature vector MFCC of the windowed speech frame of the speech segment; the calculation module is used to calculate the The distance between the MFCC and the voiceprint identification vector, where the voiceprint identification vector is obtained by pre-inputting sampled voice information of the user into the voiceprint feature training model for training; the identification module is used for when the distance is less than a preset When the threshold is set, it is determined that the recognition result of the speech segment is passed.

Further, this application also proposes a computer device, the computer device includes a memory and a processor, the memory stores computer-readable instructions that can run on the processor, and the computer-readable instructions are The implementation steps when the processor is executed:

Acquire a voice segment, divide the voice segment into frames to obtain each frame of voice data; according to a preset smooth windowing algorithm, sequentially window each frame of the voice segment of the voice segment to obtain the voice segment Windowed speech frame; extracting the Mel frequency cepstrum feature vector MFCC of the windowed speech frame of the speech segment; calculating the distance between the MFCC and the voiceprint discrimination vector, wherein the voiceprint discrimination vector is the The sampled voice information of the user is input into the voiceprint feature training model for training; when the distance is less than a preset threshold, it is determined that the recognition result of the voice segment is passed.

Further, in order to achieve the above object, the present application also provides a non-volatile computer-readable storage medium, the non-volatile computer-readable storage medium stores computer-readable instructions, and the computer-readable instructions can be At least one processor executes, so that the at least one processor executes the steps:

Acquire a voice segment, divide the voice segment into frames to obtain each frame of voice data; according to a preset smooth windowing algorithm, sequentially window each frame of the voice segment of the voice segment to obtain the voice segment Windowed speech frame; extracting the Mel frequency cepstrum feature vector MFCC of the windowed speech frame of the speech fragment; calculating the distance between the MFCC and the voiceprint discrimination vector, wherein the voiceprint discrimination vector is the The sampled voice information of the user is input into the voiceprint feature training model for training; when the distance is less than a preset threshold, it is determined that the recognition result of the voice segment is passed.

The speech recognition method, device, computer equipment, and non-volatile computer-readable storage medium proposed in this application can obtain speech fragments and then divide into frames to obtain each frame of speech data, and then compare all the speech data according to a preset smooth windowing algorithm. Each frame of speech data is windowed to obtain a windowed speech frame; then, the Mel frequency cepstrum feature vector MFCC of the windowed speech frame of the speech segment is extracted, and the distance between the MFCC and the voiceprint discrimination vector is calculated ; When the distance is less than a preset threshold, it is determined that the recognition result of the speech segment is passed. Through the above method, the feature vector in the speech segment can be calculated more accurately with a small amount of modification to the speech signal, thereby improving the accuracy of speech recognition.

Description of the drawings

Fig. 1 is a schematic diagram of an optional hardware architecture of the computer equipment of the present application;

2 is a schematic diagram of program modules of an embodiment of the speech recognition device of the present application;

Fig. 3 is a schematic flowchart of an embodiment of a speech recognition method according to the present application.

detailed description

In order to make the purpose, technical solutions, and advantages of this application clearer, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present application, and are not used to limit the present application. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of this application.

It should be noted that the descriptions related to "first", "second", etc. in this application are only for descriptive purposes, and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of indicated technical features . Therefore, the features defined with "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but it must be based on what can be achieved by a person of ordinary skill in the art. When the combination of technical solutions is contradictory or cannot be achieved, it should be considered that such a combination of technical solutions does not exist. , Is not within the scope of protection required by this application.

Refer to FIG. 1, which is a schematic diagram of an optional hardware architecture of the computer device 1 of the present application.

In this embodiment, the computer device 1 may include, but is not limited to, a memory 11, a processor 12, and a network interface 13 that can communicate with each other through a system bus.

The computer device 1 is connected to the network through the network interface 13 (not shown in FIG. 1), and connected to other terminal devices such as mobile terminals (Mobile Terminal), mobile phones (Mobile Telephone), user equipment (User Equipment, UE), and Mobile phones (handset) and portable equipment (portable equipment), PC terminal, etc. The network may be Intranet, Internet, Global System of Mobile Communication (GSM), Wideband Code Division Multiple Access (WCDMA), 4G network, 5G Network, Bluetooth (Bluetooth), Wi-Fi, call network and other wireless or wired networks.

It should be pointed out that FIG. 1 only shows the computer device 1 with components 11-13, but it should be understood that it is not required to implement all the illustrated components, and more or fewer components may be implemented instead.

Wherein, the memory 11 includes at least one type of non-volatile computer-readable storage medium, and the non-volatile computer-readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX). Memory, etc.), random access memory (RAM), static random access memory (SRAM), read only memory (ROM), electrically erasable programmable read only memory (EEPROM), programmable read only memory (PROM), magnetic memory , Disks, CDs, etc. In some embodiments, the memory 11 may be an internal storage unit of the computer device 1, for example, a hard disk or a memory of the computer device 1. In other embodiments, the memory 11 may also be an external storage device of the computer device 1, such as a plug-in hard disk, a smart media card (SMC), and a secure digital ( Secure Digital, SD card, Flash Card, etc. Of course, the memory 11 may also include both the internal storage unit of the computer device 1 and its external storage device. In this embodiment, the memory 11 is generally used to store an operating system and various application software installed in the computer device 1, such as the program code of the voice recognition apparatus 200. In addition, the memory 11 can also be used to temporarily store various types of data that have been output or will be output.

The memory 11 stores computer readable instructions, and the computer readable instructions can be executed by at least one processor, so that the at least one processor executes the steps:

Acquire a voice segment, divide the voice segment into frames to obtain each frame of voice data; according to a preset smooth windowing algorithm, sequentially window each frame of the voice segment of the voice segment to obtain the voice segment Windowed speech frame; extracting the Mel frequency cepstrum feature vector MFCC of the windowed speech frame of the speech segment; calculating the distance between the MFCC and the voiceprint discrimination vector, wherein the voiceprint discrimination vector is the The sampled voice information of the user is input into the voiceprint feature training model for training; when the distance is less than a preset threshold, it is determined that the recognition result of the voice segment is passed.

In some embodiments, the processor 12 may be a central processing unit (Central Processing Unit, CPU), a controller, a microcontroller, a microprocessor, or other data processing chips. The processor 12 is generally used to control the overall operation of the computer device 1, such as performing data interaction or communication-related control and processing. In this embodiment, the processor 12 is configured to run the program code or process data stored in the memory 11, for example, to run the voice recognition device 200.

The network interface 13 may include a wireless network interface or a wired network interface. The network interface 13 is usually used to connect the computer device 1 with other terminal devices such as mobile terminals, mobile phones, user equipment, mobile phones and portable devices, PC terminals, etc. Establish a communication connection between.

In this embodiment, when a voice recognition device 200 is installed and running in the computer device 1, when the voice recognition device 200 is running, it can obtain voice fragments and then divide into frames to obtain each frame of voice data, and then according to the preset The stationary windowing algorithm of ”performs windowing on each frame of speech data to obtain a windowed speech frame; then, extracts the Mel frequency cepstrum feature vector MFCC of the windowed speech frame of the speech segment, and calculates the MFCC The distance to the voiceprint discrimination vector; when the distance is less than a preset threshold, it is determined that the recognition result of the voice information is passed. Through the above method, the feature vector in the speech segment can be calculated more accurately, thereby improving the accuracy of speech recognition.

So far, the application environment of each embodiment of the present application and the hardware structure and functions of related devices have been introduced in detail. Hereinafter, various embodiments of the present application will be proposed based on the above-mentioned application environment and related equipment.

First, this application proposes a voice recognition device 200.

Refer to FIG. 2, which is a program module diagram of an embodiment of the speech recognition device 200 of the present application.

In this embodiment, the speech recognition device 200 includes a series of computer-readable instructions stored on the memory 11, and when the computer-readable instructions are executed by the processor 12, the speech recognition functions of the various embodiments of the present application can be implemented. . In some embodiments, the speech recognition apparatus 200 may be divided into one or more modules based on specific operations implemented by various parts of the computer-readable instructions. For example, in FIG. 2, the speech recognition device 200 may be divided into a frame module 201, a windowing module 202, an extraction module 203, a calculation module 204 and a recognition module 205. among them:

The framing module 201 is used to obtain a voice segment, and framing the voice segment to obtain each frame of voice data.

In this embodiment, the computer device 1 is connected to a user terminal, such as a mobile phone, a mobile terminal, a PC terminal, etc., and then the user's voice information is obtained through the user terminal. Of course, in other embodiments, the computer device 1 may also directly provide a pickup unit to collect the user's voice data. The voice data includes at least one voice segment. Therefore, the framing module 201 can obtain the voice segment. After the framing module 201 obtains the speech segment, it further divides the speech segment to obtain the speech data of each frame. Of course, due to the physiological characteristics of the human body, the high frequency part of the speech segment is often suppressed. Therefore, in other embodiments, the framing module 201 also performs pre-emphasis processing on the speech segment to compensate for the The high frequency components.

The windowing module 202 is configured to sequentially window each frame of speech data of the speech segment according to a preset smooth windowing algorithm to obtain a windowed speech frame of the speech segment.

Specifically, after the framing module 201 divides the speech segment into frames, the windowing module 202 further performs windowing on each frame of speech data of the speech segment. In this embodiment, the windowing module 202 sequentially windows each frame of speech data of the speech segment according to a preset smooth windowing algorithm, and then obtains the windowed speech frame of the speech segment. Wherein, the stable windowing algorithm is:

Among them, T1 is the time length of the windowed speech frame, w(t) represents the weighted value of the speech signal at time t within the time length range of the speech frame, and K and K′ are constant variables, K< K'and K+K'=1, K is set according to the environmental noise.

In this embodiment, when the computer device 1 performs windowing on each frame of speech data, it first obtains the frequency distribution information of the environmental noise in the speech data, and then automatically adjusts the variable K, and then divides the frame according to the variable K. Segmented windowing includes: adopting cosine waveform-like windowing for the beginning and end of the speech frame to reduce environmental noise interference in the low-frequency part; adopting rectangular-like windowing for the middle part of the speech frame to avoid sudden bursts High frequency noise generated by mutation. Wherein, for the process of automatically adjusting the variable K, the computer device 1 may randomly select two voice sub-frames from the voice sub-frames in the voice fragment in advance, and then convert them to the frequency domain by Fourier transform, and detect the The frequency distribution of environmental noise, and then set the KT1 at a position higher than the maximum frequency of the environmental noise.

The extraction module 203 is configured to extract the Mel frequency cepstrum feature vector MFCC of the windowed speech frame of the speech segment.

Specifically, after the windowing module 202 performs windowing on all the voice sub-frames of the voice segment, the extraction module 203 further processes the windowed voice frames of the voice segment to extract the Mel frequency cepstrum feature Vector MFCC. In this embodiment, the extraction module 203 first performs discrete Fourier transform on the windowed speech frame, from the time domain to the frequency domain; then according to the formula:

Map the linear spectrum domain of the windowed speech frame to the Mel spectrum domain; finally input to a set of Mel triangle filter bank, calculate the logarithmic energy of the signal output by the filter of each frequency band, and obtain a logarithmic energy sequence; Then, the logarithmic energy sequence is subjected to discrete cosine transform, thereby extracting the Mel frequency cepstrum feature vector MFCC of the windowed speech frame.

The calculation module 204 is configured to calculate the distance between the MFCC and a voiceprint identification vector, where the voiceprint identification vector is obtained by pre-inputting sampled voice information of the user into a voiceprint feature training model for training. The recognition module 205 is configured to determine that the recognition result of the speech segment is passed when the distance is less than a preset threshold.

Specifically, the computer device 1 will sample the user's voice information in advance, and then input the adopted voice information into a voiceprint feature training model for training, so as to obtain a voiceprint identification vector corresponding to the user. Therefore, after the extraction module 203 extracts the MFCC of the speech segment, the calculation module 204 further calculates the distance between the MFCC and the voiceprint discrimination vector. The distance is the cosine distance, and the calculation formula corresponding to the distance is:

Among them, x represents the standard voiceprint discrimination vector, and y represents the current voiceprint discrimination vector. In this embodiment, the calculation module 204 uses the cosine distance formula to calculate the distance between the MFCC of the speech segment and the preset voiceprint discrimination vector, and then the recognition module 205 compares the distance with the preset When the distance is less than the threshold, it is determined that the recognition result of the speech segment is passed.

Specifically, the computer device 1 preliminarily trains the voiceprint identification vectors of different users through GMM and calculates the distances of the MFCC respectively, thereby selecting the first voiceprint identification corresponding to the smallest distance which is less than the preset threshold. Vector, the first user corresponding to the first voiceprint discrimination vector is used as the target user corresponding to the voice segment.

Of course, in other embodiments, the computer device 1 will also pre-train a GMM (Gaussian Mixture Model) with higher accuracy, where the GMM serves as a universal background model (UBM, Universal Background Model) , Can be used to extract the voiceprint identification vector in speech, wherein the GMM can be trained through a series of sample data, so as to improve the training accuracy of the voiceprint identification vector. The training process of the GMM is as follows:

B1. Obtain a preset number (for example, 100,000) of voice data samples. Each voice data sample can be collected from the voices of different people in different environments (that is, corresponding to a voiceprint identification vector), such voice data samples It is used to train a general background model that can characterize general speech characteristics.

B2. Process each voice data sample separately to extract the preset type voiceprint feature corresponding to each voice data sample, and construct the voiceprint feature corresponding to each voice data sample based on the preset type voiceprint feature corresponding to each voice data sample vector;

B3. Divide the constructed voiceprint feature vectors of all preset types into a training set of a first percentage and a validation set of a second percentage, and the first percentage and the second percentage will be less than or Equal to 100%;

B4. Use the voiceprint feature vectors in the training set to train the first model, and use the verification set to verify the accuracy of the trained first model after the training is completed;

B5. If the accuracy rate is greater than the preset accuracy rate (for example, 98.5%), the model training ends, otherwise, increase the number of voice data samples, and re-execute the above steps B2, B3, B4, B5 based on the increased voice data samples .

Therefore, the computer device 1 first trains the collected user's voice information according to the trained GMM to obtain the corresponding voiceprint discrimination vector, and then the calculation module 204 uses the voiceprint discrimination vector to calculate the difference between the voice segment and the voice segment. The distance of the corresponding MFCC to improve accuracy.

It can be seen from the above that the computer device 1 can obtain the voice segment and then divide the frame to obtain each frame of voice data, and then perform windowing on each frame of voice data according to a preset smooth windowing algorithm to obtain a windowed voice frame ; Next, extract the Mel frequency cepstrum feature vector MFCC of the windowed speech frame of the speech segment, and calculate the distance between the MFCC and the voiceprint discrimination vector; when the distance is less than a preset threshold, determine the The recognition result of the speech fragment is passed. Through the above method, the feature vector in the speech segment can be calculated more accurately with a small amount of modification to the speech signal, thereby improving the accuracy of speech recognition.

In addition, this application also proposes a voice recognition method, which is applied to computer equipment.

Refer to FIG. 3, which is a schematic flowchart of an embodiment of a speech recognition method according to the present application. In this embodiment, according to different requirements, the execution order of the steps in the flowchart shown in FIG. 3 can be changed, and some steps can be omitted.

Step S500: Acquire a voice segment, divide the voice segment into frames, and obtain each frame of voice data.

In this embodiment, the computer device is connected to a user terminal, such as a mobile phone, a mobile terminal, a PC terminal, and other devices, and then the user's voice information is obtained through the user terminal. Of course, in other embodiments, the computer device can also directly provide a pickup unit to collect the user's voice data, and the voice data includes at least one voice segment. Therefore, the computer device can obtain the voice segment. After acquiring the voice segment, the computer device further divides the voice segment into frames to obtain the voice data of each frame. Of course, due to the physiological characteristics of the human body, the high frequency part of the speech segment is often suppressed. Therefore, in other embodiments, the computer device also performs pre-emphasis processing on the speech segment to compensate for the high frequency in the speech segment. Frequency components.

In step S502, each frame of voice data of the voice segment is sequentially windowed according to a preset smooth windowing algorithm to obtain a windowed voice frame of the voice segment.

Specifically, after the computer device divides the speech segment into frames, it further performs windowing on each frame of speech data of the speech segment. In this embodiment, the computer device sequentially windows each frame of speech data of the speech segment according to a preset smooth windowing algorithm, and then obtains a windowed speech frame of the speech segment. Wherein, the stable windowing algorithm is:

Among them, T1 is the time length of the windowed speech frame, w(t) represents the weighted value of the speech signal at time t within the time length range of the speech frame, and K and K′ are constant variables, K< K'and K+K'=1, K is set according to the environmental noise.

In this embodiment, when the computer device performs windowing on each frame of speech data, it first obtains the frequency distribution information of the environmental noise in the speech data, and then automatically adjusts the variable K, and then divides the frame according to the variable K. Segmented windowing includes: adopting cosine waveform-like windowing for the start and end of the speech frame to reduce environmental noise interference in the low frequency part; adopting rectangular-like windowing for the middle part of the speech frame to avoid sudden mutation High-frequency noise generated. Wherein, for the process of automatically adjusting the variable K, the computer device may randomly select two voice sub-frames from the voice sub-frames in the voice segment in advance, and then convert them to the frequency domain by Fourier transform, and detect the environment therein. The frequency distribution of the noise, and then set the KT1 at a position higher than the maximum frequency of the environmental noise.

Step S504: Extract the Mel frequency cepstrum feature vector MFCC of the windowed speech frame of the speech segment.

Specifically, after the computer device performs windowing on all voice sub-frames of the voice segment, it further processes the windowed voice frames of the voice segment to extract the Mel frequency cepstrum feature vector MFCC. In this embodiment, the computer device first performs discrete Fourier transform on the windowed speech frame, from the time domain to the frequency domain; then according to the formula:

Map the linear spectrum domain of the windowed speech frame to the Mel spectrum domain; finally input to a set of Mel triangle filter bank, calculate the logarithmic energy of the signal output by the filter of each frequency band, and obtain a logarithmic energy sequence; Then, the logarithmic energy sequence is subjected to discrete cosine transform, thereby extracting the Mel frequency cepstrum feature vector MFCC of the windowed speech frame.

Step S506: Calculate the distance between the MFCC and the voiceprint identification vector, where the voiceprint identification vector is obtained by pre-inputting sampled voice information of the user into a voiceprint feature training model for training.

Step S508: When the distance is less than a preset threshold, it is determined that the recognition result of the speech segment is passed.

Specifically, the computer device will sample the user's voice information in advance, and then input the adopted voice information into the voiceprint feature training model for training, so as to obtain the voiceprint identification vector corresponding to the user. Therefore, after the computer device extracts the MFCC of the speech segment, it will further calculate the distance between the MFCC and the voiceprint discrimination vector. The distance is the cosine distance, and the calculation formula corresponding to the distance is:

Among them, x represents the standard voiceprint discrimination vector, and y represents the current voiceprint discrimination vector. In this embodiment, the computer device uses the cosine distance formula to calculate the distance between the MFCC of the speech segment and the preset voiceprint discrimination vector, and then the computer device compares the distance with a preset threshold. Compare; when the distance is less than the threshold, it is determined that the recognition result of the speech segment is passed.

Specifically, the computer device preliminarily trains the voiceprint identification vectors of different users through the GMM and calculates the distances of the MFCC respectively, thereby selecting the first voiceprint identification vector corresponding to the smallest distance that is less than a preset threshold. , Taking the first user corresponding to the first voiceprint discrimination vector as the target user corresponding to the voice segment.

Of course, in other embodiments, the computer device will also pre-train a GMM (Gaussian Mixture Model) with higher accuracy, where the GMM is used as a universal background model (UBM, Universal Background Model), It can be used to extract the voiceprint discrimination vector in speech, where the GMM can be trained through a series of sample data, so as to improve the training accuracy of the voiceprint discrimination vector. The training process of the GMM is as follows:

B1. Obtain a preset number (for example, 100,000) of voice data samples. Each voice data sample can be collected from the voices of different people in different environments (that is, corresponding to a voiceprint identification vector), such voice data samples It is used to train a general background model that can characterize general speech characteristics.

B2. Process each voice data sample separately to extract the preset type voiceprint feature corresponding to each voice data sample, and construct the voiceprint feature corresponding to each voice data sample based on the preset type voiceprint feature corresponding to each voice data sample vector;

B3. Divide the constructed voiceprint feature vectors of all preset types into a training set of a first percentage and a validation set of a second percentage, and the first percentage and the second percentage will be less than or Equal to 100%;

B4. Use the voiceprint feature vectors in the training set to train the first model, and use the verification set to verify the accuracy of the trained first model after the training is completed;

B5. If the accuracy rate is greater than the preset accuracy rate (for example, 98.5%), the model training ends, otherwise, increase the number of voice data samples, and re-execute the above steps B2, B3, B4, B5 based on the increased voice data samples .

Therefore, the computer device first trains the collected user's voice information according to the trained GMM to obtain the corresponding voiceprint identification vector, and then the calculation module 204 uses the voiceprint identification vector to calculate the corresponding voice segment MFCC distance, thereby improving accuracy.

The speech recognition method proposed in this embodiment can obtain each frame of speech data by framing after acquiring a speech fragment, and then windowing each frame of speech data according to a preset smooth windowing algorithm to obtain a windowed speech frame; Then, extract the Mel frequency cepstrum feature vector MFCC of the windowed speech frame of the speech segment, and calculate the distance between the MFCC and the voiceprint discrimination vector; when the distance is less than a preset threshold, determine the speech The recognition result of the fragment is passed. Through the above method, the feature vector in the speech segment can be calculated more accurately with a small amount of modification to the speech signal, thereby improving the accuracy of speech recognition.

The serial numbers of the foregoing embodiments of the present application are only for description, and do not represent the advantages and disadvantages of the embodiments.

Through the description of the above implementation manners, those skilled in the art can clearly understand that the above-mentioned embodiment method can be implemented by means of software plus the necessary general hardware platform, of course, it can also be implemented by hardware, but in many cases the former is better.的实施方式。 Based on this understanding, the technical solution of this application essentially or the part that contributes to the existing technology can be embodied in the form of a software product, and the computer software product is stored in a storage medium (such as ROM/RAM, magnetic disk, The optical disc) includes several instructions to enable a terminal device (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the method described in each embodiment of the present application.

The above are only the preferred embodiments of the application, and do not limit the scope of the patent for this application. Any equivalent structure or equivalent process transformation made using the content of the description and drawings of the application, or directly or indirectly applied to other related technical fields , The same reason is included in the scope of patent protection of this application.

Claims (20)

A speech recognition method, the method includes the steps: Acquiring a voice segment, framing the voice segment to obtain each frame of voice data; Sequentially windowing each frame of speech data of the speech segment according to a preset smooth windowing algorithm to obtain a windowed speech frame of the speech segment; Extracting the Mel frequency cepstrum feature vector MFCC of the windowed speech frame of the speech segment; Calculating the distance between the MFCC and a voiceprint identification vector, where the voiceprint identification vector is obtained by pre-inputting sampled voice information of the user into a voiceprint feature training model for training; When the distance is less than the preset threshold, it is determined that the recognition result of the speech segment is passed. The speech recognition method according to claim 1, wherein the stable windowing algorithm is:

Among them, T1 is the time length of the windowed speech frame, w(t) represents the weighted value of the speech signal at time t within the time length range of the speech frame, and K and K′ are constant variables, K< K'and K+K'=1, K is set according to the environmental noise. The speech recognition method according to claim 2, the method further comprising: When each frame of speech data is windowed, the frequency distribution information of the environmental noise in the speech data is obtained, and then the K is adjusted according to the highest frequency distribution of the noise. The speech recognition method according to claim 2, wherein the smooth windowing algorithm comprises: For the frame start and end of the speech frame frequency distribution, use cosine waveform windowing to reduce the environmental noise interference in the low frequency part; For the middle part of the frequency distribution of the speech frame, a rectangular window is used to avoid high frequency noise caused by sudden mutation. The speech recognition method according to claim 1, wherein the voiceprint feature training model is a Gaussian mixture model GMM, and the method further comprises: The voiceprint identification vectors of different users are trained by GMM and the distance calculation is performed on the MFCC respectively; Selecting the first voiceprint discrimination vector corresponding to the smallest distance smaller than the preset threshold; The first user corresponding to the first voiceprint discrimination vector is used as the target user corresponding to the voice segment. The speech recognition method according to claim 1, wherein the distance is a cosine distance, and the calculation formula corresponding to the distance is:

Among them, x represents the standard voiceprint discrimination vector, and y represents the current voiceprint discrimination vector. 5. The speech recognition method according to claim 1, before said framing said speech segment, said method further comprises: Pre-emphasis is performed on the voice segment to compensate for high frequency components in the voice segment. A speech recognition device, the device includes: The framing module is used to obtain speech fragments, and divide the speech fragments into frames to obtain each frame of speech data; The windowing module is configured to sequentially window each frame of voice data of the voice segment according to a preset smooth windowing algorithm to obtain a windowed voice frame of the voice segment; An extraction module for extracting the Mel frequency cepstrum feature vector MFCC of the windowed speech frame of the speech segment; A calculation module for calculating the distance between the MFCC and a voiceprint identification vector, where the voiceprint identification vector is obtained by pre-inputting sampled voice information of the user into a voiceprint feature training model for training; The recognition module is configured to determine that the recognition result of the speech segment is passed when the distance is less than a preset threshold. 8. The speech recognition device of claim 8, wherein the smooth windowing algorithm is:

Among them, T1 is the time length of the windowed speech frame, w(t) represents the weighted value of the speech signal at time t within the time range of the speech frame, and K and K′ are constant variables, K< K'and K+K'=1, K is set according to the environmental noise. The speech recognition device according to claim 9, wherein the windowing module is further used for: When each frame of speech data is windowed, the frequency distribution information of the environmental noise in the speech data is obtained, and then the K is adjusted according to the highest frequency distribution of the noise. 8. The speech recognition device according to claim 8, wherein the voiceprint feature training model is a Gaussian mixture model GMM, The calculation module is further configured to train the voiceprint discrimination vectors of different users through GMM and calculate the distances of the MFCC respectively; The recognition module is further configured to select a first voiceprint identification vector corresponding to a distance smaller than a preset threshold and the smallest distance; and use the first user corresponding to the first voiceprint identification vector as the voice segment corresponding Target users. 8. The speech recognition device according to claim 8, wherein the distance is a cosine distance, and the calculation formula corresponding to the distance is:

Among them, x represents the standard voiceprint discrimination vector, and y represents the current voiceprint discrimination vector. A computer device, the computer device includes a memory and a processor, the memory stores computer-readable instructions that can run on the processor, and the computer-readable instructions implement steps when executed by the processor : Acquiring a voice segment, framing the voice segment to obtain each frame of voice data; Sequentially windowing each frame of speech data of the speech segment according to a preset smooth windowing algorithm to obtain a windowed speech frame of the speech segment; Extracting the Mel frequency cepstrum feature vector MFCC of the windowed speech frame of the speech segment; Calculating the distance between the MFCC and a voiceprint identification vector, where the voiceprint identification vector is obtained by pre-inputting sampled voice information of the user into a voiceprint feature training model for training; When the distance is less than the preset threshold, it is determined that the recognition result of the speech segment is passed. The computer device according to claim 13, wherein the smooth windowing algorithm is:

Among them, T1 is the time length of the windowed speech frame, w(t) represents the weighted value of the speech signal at time t within the time range of the speech frame, and K and K′ are constant variables, K< K'and K+K'=1, K is set according to the environmental noise. The computer device according to claim 14, wherein the computer-readable instructions when executed by the processor further implement the steps: When each frame of speech data is windowed, the frequency distribution information of the environmental noise in the speech data is obtained, and then the K is adjusted according to the highest frequency distribution of the noise. The computer device according to claim 13, wherein the voiceprint feature training model is a Gaussian mixture model GMM, and the computer-readable instructions further implement the steps when executed by the processor: The voiceprint identification vectors of different users are trained by GMM and the distance calculation is performed on the MFCC respectively; Selecting the first voiceprint discrimination vector corresponding to the smallest distance smaller than the preset threshold; The first user corresponding to the first voiceprint discrimination vector is used as the target user corresponding to the voice segment. A non-volatile computer-readable storage medium, the non-volatile computer-readable storage medium stores computer-readable instructions, the computer-readable instructions can be executed by at least one processor, so that the at least one Processor execution steps: Acquiring a voice segment, framing the voice segment to obtain each frame of voice data; Sequentially windowing each frame of speech data of the speech segment according to a preset smooth windowing algorithm to obtain a windowed speech frame of the speech segment; Extracting the Mel frequency cepstrum feature vector MFCC of the windowed speech frame of the speech segment; Calculating the distance between the MFCC and a voiceprint identification vector, where the voiceprint identification vector is obtained by pre-inputting sampled voice information of the user into a voiceprint feature training model for training; When the distance is less than the preset threshold, it is determined that the recognition result of the speech segment is passed. The non-volatile computer-readable storage medium of claim 17, wherein the smooth windowing algorithm is:

Among them, T1 is the time length of the windowed speech frame, w(t) represents the weighted value of the speech signal at time t within the time length range of the speech frame, and K and K′ are constant variables, K< K'and K+K'=1, K is set according to the environmental noise. The non-volatile computer-readable storage medium of claim 18, wherein the computer-readable instructions further implement the steps when executed by the processor: When each frame of speech data is windowed, the frequency distribution information of the environmental noise in the speech data is obtained, and then the K is adjusted according to the highest frequency distribution of the noise. The non-volatile computer-readable storage medium according to claim 17, wherein the voiceprint feature training model is a Gaussian Mixture Model GMM, and the computer-readable instructions further implement the steps when executed by the processor: The voiceprint identification vectors of different users are trained by GMM and the distance calculation is performed on the MFCC respectively; Selecting the first voiceprint discrimination vector corresponding to the smallest distance smaller than the preset threshold; The first user corresponding to the first voiceprint discrimination vector is used as the target user corresponding to the voice segment.

Images