Artificial intelligence speech recognition method, device and medium based on deep learning

Abstract

The present invention relates to an artificial intelligence speech recognition method, device and medium based on deep learning. The method recognizes the user's historical speech information and, when the recognition result matches the user's historical text input information and related words of the historical text input information, associates the historical speech information with the user's historical text input information and related words of the historical text input information, generates data samples to train a speech recognition model, so that the recognition result of the speech recognition model is closer to the user's daily work and life environment, and the speech recognition result is more accurate in the user's work and life field.

Description

Technical Field

[0001] The present invention relates to the field of speech recognition technology, and in particular to an artificial intelligence speech recognition method, device and medium based on deep learning. Background Art

[0002] With the development of science and technology, the application of computer devices covers all aspects of people's work and life. At the same time, users have higher and higher requirements for computer devices. In the process of using computer devices, there are many scenarios where text input is required, such as inputting text in a document, inputting text in a search bar, etc.

[0003] In the general consumer field, the current accuracy of speech-to-text is low because there are many homophones in Chinese, and the accuracy of converting speech to text only through voice features needs to be improved. Summary of the invention

[0004] In view of this, an object of the present invention is to provide an artificial intelligence speech recognition method, device and medium based on deep learning, which can improve the accuracy of speech-to-text conversion.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] An artificial intelligence speech recognition method based on deep learning of the present invention comprises the steps of:

[0007] Obtaining the user's historical voice information, historical text input information, and related words of the historical text input information;

[0008] Recognize the historical voice information according to a pre-established voice recognition model to obtain a recognition result;

[0009] Comparing the recognition result with the historical text input information and the related words, and constructing a data sample with the historical voice information, the historical text input information and the related words if the recognition result matches the historical text input information or the related words;

[0010] Training the speech recognition according to the data sample to obtain an updated speech recognition model;

[0011] The speech information is recognized based on the updated speech recognition model.

[0012] In one embodiment of the present invention, the historical voice information is recognized by using a pre-established voice recognition model to obtain a recognition result, including:

[0013] Converting the historical speech information into a pulse modulation waveform, and mapping the pulse modulation waveform into a two-dimensional coordinate system, wherein the two-dimensional coordinate system includes an X-axis and a Y-axis, wherein the X-axis is a time axis and the Y-axis is an amplitude axis;

[0014] Extract the inflection point Q from the pulse modulation waveform i and feature point Q i ', and generate the inflection point Q in chronological order i and feature point Q i 'Feature set S;

[0015] The feature set S is input into the speech recognition model to obtain the recognition result.

[0016] In one embodiment of the present invention, the inflection point Q is extracted from the pulse modulation waveform. i and feature point Q i ',include:

[0017] Calculate any two adjacent data points A of the pulse modulation waveform i and A i-1 The amplitude difference between - -1 ;

[0018] When any data point satisfies: Then extract the data point and use it as the inflection point Q i ;

[0019] When any two data points -1 and meet: |- -1 |>αor|- -1 |<β, the data points -1 and data points as feature points Q i '; where α is the first threshold and β is the second threshold.

[0020] In one embodiment of the present invention, the step of establishing the speech recognition model includes:

[0021] Acquire an initial data set including positive samples and samples, wherein the positive samples are pulse modulation data including a feature set S, and the negative samples are pulse modulation data not including the feature set S;

[0022] The feature set S in the positive sample is amplitude-scaled and overall shifted to obtain a training data set;

[0023] The preset artificial neural network is trained according to the training data set to obtain the speech recognition model.

[0024] In one embodiment of the present invention, the feature set S in the positive sample is amplitude-scaled and overall-shifted to obtain a training data set, including:

[0025] Obtain a random multiple sequence N={n}, and multiply any element in the random multiple sequence N={n} by the amplitude of each data point in the feature set S to obtain a plurality of scaled feature sets n×S;

[0026] Obtain a random displacement sequence M={m}, add any element in the random displacement sequence M={m} to the amplitude of each data point in the feature set S, and obtain a plurality of displacement feature sets m+S;

[0027] The training data set is constructed based on the feature set S, a plurality of scaled feature sets n×S, and a plurality of shifted feature sets m+S.

[0028] In one embodiment of the present invention, obtaining related words of historical text input information includes:

[0029] Segmenting the historical text input information and obtaining nouns therein;

[0030] Extract related words of the noun from a preset word library.

[0031] In one embodiment of the present invention, comparing the recognition result with the historical text input information and the related words includes:

[0032] Pre-establish the mapping relationship between text information and recognition results;

[0033] The text information is obtained according to the recognition result and the mapping relationship, and the text information is compared with the historical text input information and the related words.

[0034] The present invention also provides a speech recognition device based on artificial intelligence, comprising:

[0035] A collection module, used to obtain the user's historical voice information, historical text input information and related words of the historical text input information;

[0036] A recognition module, used to recognize the historical voice information according to a pre-established voice recognition model to obtain a recognition result;

[0037] a sample extraction module, for comparing the recognition result with the historical text input information and the related words, and constructing a data sample with the historical voice information, the historical text input information and the related words when the recognition result matches the historical text input information or the related words;

[0038] A model training module, used to train the speech recognition according to the data sample to obtain an updated speech recognition model;

[0039] The recognition module is used to recognize speech information based on the updated speech recognition model.

[0040] The present invention also provides a storage medium, which stores a computer program. When the computer program is loaded and executed by a processor, it implements the artificial intelligence speech recognition method based on deep learning as described above.

[0041] The present invention also provides a computer device, comprising: a processor and a memory; wherein the memory is used to store a computer program; the processor is used to load and execute the computer program so that the computer device executes the deep learning-based artificial intelligence speech recognition method as described above.

[0042] The beneficial effects of the present invention are as follows: the deep learning-based artificial intelligence speech recognition method, device and medium of the present invention recognize the user's historical speech information, and when the recognition result is matched with the user's historical text input information and the related words of the historical text input information, the historical voice information is associated with the user's historical text input information and the related words of the historical text input information, and data samples are generated to train the speech recognition model, so that the recognition result of the speech recognition model is closer to the user's daily work and life environment, and the speech recognition result is more accurate in the user's work and life field. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The present invention will be further described below in conjunction with the accompanying drawings and embodiments:

[0044] Figure 1 Schematic diagram of a flow chart of an artificial intelligence speech recognition method based on deep learning in one embodiment of the present invention;

[0045] Figure 2 It is a structural diagram of a speech recognition device based on artificial intelligence in one embodiment of the present invention. DETAILED DESCRIPTION

[0046] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0047] It should be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present invention, and thus the drawings only show the layers related to the present invention rather than being drawn according to the number, shape and size of the layers in actual implementation. In actual implementation, the type, quantity and proportion of each layer may be changed arbitrarily, and the layer layout may also be more complicated.

[0048] In the following description, numerous details are discussed to provide a more thorough explanation of the embodiments of the present invention. However, it is apparent to those skilled in the art that the embodiments of the present invention can be implemented without these specific details. Figure 1 As shown: The artificial intelligence speech recognition method based on deep learning of this embodiment includes the following steps:

[0049] S110. Obtain the user's historical voice information, historical text input information, and related words of the historical text input information;

[0050] In this embodiment, the user's historical voice information can be obtained from the mobile phone voice files and communication records, and the historical text input information can be obtained from the mobile phone's chat records and search records, and the computer's work records.

[0051] S120. Recognize historical voice information according to a pre-established voice recognition model to obtain a recognition result;

[0052] S130. Compare the recognition result with the historical text input information and related words, and if the recognition result matches the historical text input information or related words, construct a data sample with the historical voice information, historical text input information and related words;

[0053] The recognition result converts the speech into text, and then compares the text information with the historical text input information and related words. If the text information matches the historical text input information and related words, the historical speech information is associated with the historical text input information and related words, and a data sample is generated;

[0054] S140. Train the speech recognition according to the data sample to obtain an updated speech recognition model;

[0055] S150. Recognize the speech information based on the updated speech recognition model.

[0056] More and more data samples are generated based on information related to the user's work and life, and the speech recognition model is continuously updated, so that the recognition results of the speech recognition model are more in line with the user's work and life environment, and the recognition results are more accurate.

[0057] In some embodiments, the process of obtaining the recognition result by recognizing the historical voice information through the pre-established voice recognition model may include steps 210 to S230, which are specifically described as follows:

[0058] S210. Convert the historical voice information into a pulse modulation waveform, and map the pulse modulation waveform to a two-dimensional coordinate system, the two-dimensional coordinate system includes an X-axis and a Y-axis, wherein the X-axis is a time axis and the Y-axis is an amplitude axis;

[0059] S220. Extract the inflection point Q from the pulse modulation waveform i and feature point Q i ', and generate the inflection point Q in chronological order i and feature point Q i 'Feature set S;

[0060] S230. Input the feature set S into the speech recognition model to obtain the recognition result.

[0061] In this embodiment, the historical voice information is converted into PCM data, and then the PCM data is mapped to a two-dimensional coordinate system to form a PCM waveform. In this embodiment, the inflection point Q of the PCM waveform is extracted. i and feature point Q i ' as the feature of PCM data, and then input the feature data into the speech recognition model for recognition.

[0062] In some embodiments, the inflection point Q is extracted from the pulse modulation waveform. i and feature point Q i The process may include step S310 to step S330, which are described in detail as follows:

[0063] S310. Calculate any two adjacent data points A of the pulse modulation waveform i and A i-1 The amplitude difference between - -1 ;

[0064] S320. When any data point satisfies: Then extract the data point and use it as the inflection point Q i ;

[0065] In step S320, when , indicating the data point -1 , data points +1 are all above the data point, or the data point -1 ; Data points +1 If they are all below the data point, it means that the data point is an inflection point;

[0066] S330. When there are any two data points -1and meet: |- -1 |>αor|- -1 |<β, the data points -1 and data points as feature points Q i '; where α is the first threshold and β is the second threshold;

[0067] In step S330, if: |- -1 |>αor|- -1 |<β means that two adjacent data points, -1 The data segments formed are too steep or too smooth, and the data segments formed by such data points are collected as data features of the PCM waveform.

[0068] In some embodiments, the process of establishing the speech recognition model may include steps S410 to S430, which are described in detail as follows:

[0069] S410. Obtain an initial data set including positive samples and negative samples, where the positive samples are pulse modulation data including the feature set S, and the negative samples are pulse modulation data not including the feature set S;

[0070] S420. Perform amplitude scaling and overall displacement on the feature set S in the positive sample to obtain a training data set;

[0071] By performing amplitude scaling and overall shifting on the feature set S, the diversity of positive samples is increased, thereby improving the recognition accuracy of the trained speech recognition model.

[0072] S430. Train the preset artificial neural network according to the training data set to obtain a speech recognition model.

[0073] In some embodiments, the process of performing amplitude scaling and overall displacement on the feature set S in the positive sample to obtain the training data set may include steps S510 to S530, which are specifically described as follows:

[0074] S510. Obtain a random multiple sequence N={n}, multiply any element in the random multiple sequence N={n} by the amplitude of each data point in the feature set S, and obtain multiple scaled feature sets n×S;

[0075] S520. Obtain a random displacement sequence M={m}, add any element in the random displacement sequence M={m} to the amplitude of each data point in the feature set S, and obtain a plurality of displacement feature sets m+S;

[0076] S530. Construct a training data set based on the feature set S, multiple scaled feature sets n×S, and multiple shifted feature sets m+S.

[0077] In some embodiments, the process of obtaining related words of historical text input information may include steps S610 to S620, which are specifically described as follows:

[0078] S610. Segment the historical text input information and obtain the nouns therein;

[0079] Common word segmentation algorithms include a forward maximum matching algorithm, a reverse maximum matching algorithm, and a bidirectional maximum matching algorithm, etc., which are not limited in this embodiment.

[0080] S620. Extract related words of the noun from a preset vocabulary.

[0081] In step S620, the word library can be obtained from the network, and the existing related word algorithm is used to obtain related words of the noun, and the user's historical text input information is reasonably expanded to increase recognition reliability.

[0082] In some embodiments, the process of comparing the recognition result with the historical text input information and related words may include steps S710 to S720, which are described in detail as follows:

[0083] S710. Pre-establish a mapping relationship between text information and recognition results;

[0084] S720. Obtain text information based on the recognition result and the mapping relationship, and compare the text information with historical text input information and related words.

[0085] The deep learning-based artificial intelligence speech recognition method of the present invention recognizes the user's historical speech information, and when the recognition result matches the user's historical text input information and the related words of the historical text input information, the historical speech information is associated with the user's historical text input information and the related words of the historical text input information, and data samples are generated to train the speech recognition model, so that the recognition result of the speech recognition model is closer to the user's daily work and life environment, and the speech recognition result is more accurate in the user's work and life field.

[0086] The present invention also provides a speech recognition device based on artificial intelligence, comprising:

[0087] A collection module, used to obtain the user's historical voice information, historical text input information and related words of the historical text input information;

[0088] A recognition module, used to recognize the historical voice information according to a pre-established voice recognition model to obtain a recognition result;

[0089] a sample extraction module, for comparing the recognition result with the historical text input information and the related words, and constructing a data sample with the historical voice information, the historical text input information and the related words when the recognition result matches the historical text input information or the related words;

[0090] A model training module, used to train the speech recognition according to the data sample to obtain an updated speech recognition model;

[0091] The recognition module is used to recognize speech information based on the updated speech recognition model.

[0092] The artificial intelligence-based speech recognition device of the present invention recognizes the user's historical speech information, and when the recognition result matches the user's historical text input information and the related words of the historical text input information, associates the historical speech information with the user's historical text input information and the related words of the historical text input information, generates data samples to train the speech recognition model, so that the recognition result of the speech recognition model is closer to the user's daily work and life environment, and the speech recognition result is more accurate in the user's work and life field.

[0093] This embodiment further provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, any one of the methods in this embodiment is implemented.

[0094] This embodiment also provides an electronic terminal, including: a processor and a memory;

[0095] The memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory, so that the terminal executes any one of the methods in this embodiment.

[0096] The computer-readable storage medium in this embodiment can be understood by ordinary technicians in this field: all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to the computer program. The aforementioned computer program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, magnetic disk or optical disk and other media that can store program codes.

[0097] The electronic terminal provided in this embodiment includes a processor, a memory, a transceiver and a communication interface. The memory and the communication interface are connected to the processor and the transceiver and complete communication with each other. The memory is used to store computer programs, the communication interface is used to communicate, and the processor and the transceiver are used to run computer programs so that the electronic terminal executes each step of the above method.

[0098] In this embodiment, the memory may include a random access memory (RAM), and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0099] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0100] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.

Claims

1. An artificial intelligence speech recognition method based on deep learning, characterized in that: Includes steps: Obtain the user's historical voice information, historical text input information, and related words of the historical text input information; Recognize the historical voice information according to a pre-established voice recognition model to obtain a recognition result; Comparing the recognition result with the historical text input information and the related words, and constructing a data sample with the historical voice information, the historical text input information and the related words if the recognition result matches the historical text input information or the related words; Training the speech recognition according to the data sample to obtain an updated speech recognition model; Recognize the speech information based on the updated speech recognition model; Recognizing the historical voice information according to a pre-established voice recognition model to obtain a recognition result, including: Converting the historical speech information into a pulse modulation waveform, and mapping the pulse modulation waveform into a two-dimensional coordinate system, wherein the two-dimensional coordinate system includes an X-axis and a Y-axis, wherein the X-axis is a time axis and the Y-axis is an amplitude axis; Extract the inflection point Q from the pulse modulation waveform i and feature point Q i ', and generate the inflection point Q in chronological order i and feature point Q i 'Feature set S; Inputting the feature set S into the speech recognition model to obtain a recognition result; Extract the inflection point Q from the pulse modulation waveform i and feature point Q i ',include: Calculate any two adjacent data points A of the pulse modulation waveform i and A i-1 The amplitude difference A i -A i-1 ; When any data point A i satisfy: Then extract data point A i And as the inflection point Q i ; When there are any two data points A i-1 and A i Satisfies: |A i -A i-1 |>αor|A i -A i-1 |<β, data point A i-1 and data point A i And as the feature point Q i '; where α is the first threshold and β is the second threshold; The steps of establishing the speech recognition model include: Acquire an initial data set including positive samples and negative samples, wherein the positive samples are pulse modulation data including a feature set S, and the negative samples are pulse modulation data not including the feature set S; The feature set S in the positive sample is amplitude-scaled and overall shifted to obtain a training data set; Training a preset artificial neural network according to the training data set to obtain the speech recognition model; The feature set S in the positive sample is amplitude-scaled and overall shifted to obtain a training data set, including: Obtain a random multiple sequence N={n}, and multiply any element in the random multiple sequence N={n} by the amplitude of each data point in the feature set S to obtain a plurality of scaled feature sets n×S; Obtain a random displacement sequence M={m}, add any element in the random displacement sequence M={m} to the amplitude of each data point in the feature set S, and obtain a plurality of displacement feature sets m+S; The training data set is constructed based on the feature set S, a plurality of scaled feature sets n×S, and a plurality of shifted feature sets m+S.

2. The artificial intelligence speech recognition method based on deep learning according to claim 1, characterized in that: Get the related words of historical text input information, including: Segmenting the historical text input information and obtaining nouns therein; Extract related words of the noun from a preset word library.

3. The artificial intelligence speech recognition method based on deep learning according to claim 1, characterized in that: Comparing the recognition result with the historical text input information and the related words includes: Pre-establish the mapping relationship between text information and recognition results; The text information is obtained according to the recognition result and the mapping relationship, and the text information is compared with the historical text input information and the related words.

4. A speech recognition device based on artificial intelligence, characterized in that: include: A collection module, used to obtain the user's historical voice information, historical text input information and related words of the historical text input information; A recognition module, used to recognize the historical voice information according to a pre-established voice recognition model to obtain a recognition result; a sample extraction module, for comparing the recognition result with the historical text input information and the related words, and constructing a data sample with the historical voice information, the historical text input information and the related words when the recognition result matches the historical text input information or the related words; A model training module, used to train the speech recognition according to the data sample to obtain an updated speech recognition model; The recognition module is used to recognize speech information based on the updated speech recognition model.

5. A storage medium storing a computer program, characterized in that: When the computer program is loaded and executed by the processor, it implements the deep learning-based artificial intelligence speech recognition method as described in any one of claims 1 to 3.

6. A computer device, characterized in that: include: A processor and a memory; wherein the memory is used to store a computer program; the processor is used to load and execute the computer program so that the computer device executes the deep learning-based artificial intelligence speech recognition method as described in any one of claims 1 to 3.

Images