A computer speech recognition method

Abstract

The invention discloses a computer speech recognition method, which relates to the field of speech recognition. The technical scheme comprises the following steps: Step 1: establishing an audio database: recording audio through the Internet and manually, establishing an audio database, and translating corresponding text based on the audio; Step 2: learning text groups: loading an embedded processing chip in a computer, splitting the recorded audio through the embedded processing chip, splitting it into "single word", "double word", "triple word"... until "all the text of the sentence", recording the split text as "text group", matching the split "text group" with the database, and judging the actual meaning of the "text group"; and then performing a second correction on the corrected "primary speech recognition model" through deep learning of the facial expression of the recorder, and finally obtaining a "speech recognition model", and realizing rapid recognition of speech through the "speech recognition model", and greatly improving the recognition accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of speech recognition, and in particular to a computer speech recognition method. Background Art

[0002] The fields involved in speech recognition technology include: signal processing, pattern recognition, probability theory and information theory, vocal mechanism and auditory mechanism, artificial intelligence, etc.

[0003] By recording the sound and then recognizing the recorded sound, the sound is converted into information text.

[0004] After searching, the invention patent with Chinese patent number CN113851111A discloses a speech recognition method and speech recognition device, which includes: performing windowing processing on the speech data stream to determine the speech data in the window; performing object recognition processing on the speech data in the window, and adjusting the length of the window according to the object recognition processing result, and determining the speech data in the adjusted window as the target speech segment; performing speech recognition processing on the target speech segment based on the recognition model to obtain the target recognition result. In this way, when performing windowing processing on the speech data stream, the window length is flexibly adjusted according to the result of the object recognition processing to obtain target speech segments of different sizes, which can take into account both the recognition speed and the recognition effect, thereby comprehensively improving the speech recognition performance of the end-to-end speech recognition scenario.

[0005] However, the above-mentioned speech recognition method can only perform object recognition processing on the speech data in the window during use, and adjust the length of the window according to the object recognition processing result, so as to improve the speech recognition performance of the end-to-end speech recognition scenario. In actual use, due to the lack of effective recognition of the recorded speech, it is impossible to accurately translate the meanings of each phrase. At the same time, external noise greatly increases the difficulty of speech recognition, thereby reducing the accuracy of speech recognition. Therefore, a computer speech recognition method is needed. Summary of the invention

[0006] The purpose of the present invention is to solve the shortcomings of the prior art that there is a lack of effective recognition of recorded speech, so that accurate translation cannot be performed according to the meaning of each phrase, and external noise greatly increases the difficulty of speech recognition, and proposes a computer speech recognition method.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] A computer speech recognition method comprises the following steps:

[0009] Step 1: Establishing an audio database: Establish an audio database by inputting audio through the Internet and manually, and translate the corresponding text based on the audio;

[0010] Step 2: Character group learning: The computer is equipped with an embedded processing chip, which is used to split the recorded audio into "single words", "double words", "triple words"... until "all the words in the sentence", and the split words are recorded as "character groups". The actual meaning of the "character groups" is determined by matching the split "character groups" with the database;

[0011] Step 3: Model building: Combine the results of learning a large number of text groups with the audio during recording, and perform deep learning based on actual text translation to generate a "primary speech recognition model";

[0012] Step 4: Model correction: Manually input voice, then use the "primary voice recognition model" for voice recognition, select different environments for input, compare the text recognized by the "primary voice recognition model" with the actual input text, find out the differences between the texts, and correct the "primary voice recognition model" based on the differences;

[0013] Step 5: Image acquisition: The facial expression of the person being recorded is acquired during the speech process, and the acquired facial expression is associated with the corresponding speech;

[0014] Step 6: Associating the collected facial expression data with the text corresponding to the voice translation, so as to further revise the revised "primary voice recognition model" and obtain the "voice recognition model";

[0015] Step 7: Real-time speech recognition: During the speech recognition process, the acquired audio is imported into the "speech recognition model" in real time, and real-time speech recognition is performed through the "speech recognition model".

[0016] The above technical solution further includes:

[0017] When the audio database is being built, the phrase database is also being built.

[0018] When comparing the "text group" with the phrase database, the phrases in the phrase database are sorted according to the number of characters and the tone of voice.

[0019] A noise reduction module is installed in the embedded processing chip to reduce the noise in the received voice. At the same time, the noise reduction module cooperates with the "speech recognition model" to fully reduce the noise of the recorded voice.

[0020] The image acquisition device inside the computer acquires the facial image of the audio recording person, processes the acquired image, reduces the noise in the image, enhances the texture in the image, and then recognizes the processed image.

[0021] Introduce sentiment analysis mechanisms in the process of combining facial expression data with text during translation.

[0022] The sentiment analysis mechanism is used to identify the tone and attitude of the voice entrant, thereby increasing the success rate of voice recognition.

[0023] Different databases and "speech recognition models" are established for different accents.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] In the present invention, the results of learning a large number of "text groups" are combined with the audio during recording to generate a "primary speech recognition model", and the "primary speech recognition model" is trained through manual speech recording. According to the influence of the external environment, the "primary speech recognition model" is corrected, and finally, through deep learning of the facial expressions of the recorder, the corrected "primary speech recognition model" is corrected for the second time, and finally a "speech recognition model" is obtained. The "speech recognition model" can realize rapid recognition of speech and greatly improve the recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 The present invention provides a flowchart of a computer speech recognition method. DETAILED DESCRIPTION

[0027] Embodiment 1

[0028] The first step is to establish an audio database through the Internet and manual audio recording, and translate the corresponding text based on the audio. During the establishment of the audio database, a phrase database is also established, and an embedded processing chip is loaded in the computer;

[0029] The recorded audio is split by an embedded processing chip, and the number of splits is "single word", "double word", "triple word"... until "all words in a sentence". When the "word group" is compared with the phrase database, the phrases in the phrase database are sorted according to the number of words and voice intonation;

[0030] The split characters are recorded as "character groups", and the actual meaning of the "character groups" is determined by matching the split "character groups" with the database;

[0031] The second step is to combine the results of learning a large number of text groups with the audio at the time of recording, and conduct deep learning based on the actual text translation to generate a "primary speech recognition model";

[0032] The third step is to manually input the voice, and then use the "primary voice recognition model" to perform voice recognition, and choose to input in different environments;

[0033] Compare the text recognized by the "primary speech recognition model" with the actual input text to find out the differences between the texts, and modify the "primary speech recognition model" based on the differences;

[0034] The fourth step is to obtain the facial expression of the person recording during the speech process, and to associate the obtained facial expression with the corresponding speech;

[0035] The image acquisition device inside the computer acquires the facial image of the audio recording person, processes the acquired image, reduces the noise in the image, enhances the texture in the image, and then recognizes the processed image;

[0036] The fifth step is to associate the collected facial expression data with the text corresponding to the voice translation. In the process of combining the facial expression data with the text during translation, the sentiment analysis mechanism is introduced to identify the tone and attitude of the voice inputter through the sentiment analysis mechanism to increase the success rate of voice recognition, thereby further correcting the corrected "primary voice recognition model" to obtain the "voice recognition model";

[0037] A noise reduction module is installed in the embedded processing chip to reduce the noise in the received speech. At the same time, the noise reduction module cooperates with the "speech recognition model" to fully reduce the noise of the recorded speech. Different databases and "speech recognition models" are established for different accents.

[0038] The sixth step is to import the acquired audio into the "speech recognition model" in real time during the speech recognition process, and perform real-time speech recognition through the "speech recognition model".

[0039] A computer speech recognition method based on the first embodiment performs speech recognition input, and the results of learning a large number of "text groups" are combined with the audio during input to generate a "primary speech recognition model", and the "primary speech recognition model" is trained by manual speech input, and the "primary speech recognition model" is corrected according to the influence of the external environment, and finally the corrected "primary speech recognition model" is corrected again through deep learning of the facial expressions of the inputter, and finally a "speech recognition model" is obtained, and the "speech recognition model" is used to realize rapid recognition of speech and greatly improve the recognition accuracy. The test sentences are shown in the following table:

[0040] Number of input statements Number of sentences Total Word Count Single sentence recognition accuracy Single word recognition accuracy 5 26 497 94.3％ 98.3％ .

[0041] Embodiment 2

[0042] The first step is to establish an audio database through the Internet and manual audio recording, and translate the corresponding text based on the audio. During the establishment of the audio database, a phrase database is also established, and an embedded processing chip is loaded in the computer;

[0043] The recorded audio is split by an embedded processing chip, and the number of splits is "single word", "double word", "triple word"... until "all words in a sentence". When the "word group" is compared with the phrase database, the phrases in the phrase database are sorted according to the number of words and voice intonation;

[0044] The split characters are recorded as "character groups", and the actual meaning of the "character groups" is determined by matching the split "character groups" with the database;

[0045] The second step is to combine the results of learning a large number of text groups with the audio at the time of recording, and conduct deep learning based on the actual text translation to generate a "primary speech recognition model";

[0046] The third step is to manually input the voice, and then use the "primary voice recognition model" to perform voice recognition, and choose to input in different environments;

[0047] Compare the text recognized by the "primary speech recognition model" with the actual input text to find out the differences between the texts, and modify the "primary speech recognition model" based on the differences;

[0048] A noise reduction module is installed in the embedded processing chip to reduce the noise in the received speech. At the same time, the noise reduction module cooperates with the revised "speech recognition model" to fully reduce the noise of the recorded speech. Different databases and revised "speech recognition models" are established for different accents.

[0049] The fourth step is to import the acquired audio into the revised "speech recognition model" in real time during the speech recognition process, and perform real-time speech recognition through the revised "speech recognition model".

[0050] A computer speech recognition method based on the first embodiment performs speech recognition input, and the results of learning a large number of "text groups" are combined with the audio during input to generate a "primary speech recognition model", and the "primary speech recognition model" is trained by manual speech input. According to the influence of the external environment, the "primary speech recognition model" is corrected, and the corrected "speech recognition model" is used to realize rapid recognition of speech, and the recognition accuracy is greatly improved. The test sentences are shown in the following table:

[0051] Number of input statements Number of sentences Total Word Count Single sentence recognition accuracy Single word recognition accuracy 5 24 426 88.3％ 94.1％ .

[0052] Embodiment 3

[0053] The first step is to establish an audio database through the Internet and manual audio recording, and translate the corresponding text based on the audio. During the establishment of the audio database, a phrase database is also established, and an embedded processing chip is loaded in the computer;

[0054] The recorded audio is split by an embedded processing chip, and the number of splits is "single word", "double word", "triple word"... until "all words in a sentence". When the "word group" is compared with the phrase database, the phrases in the phrase database are sorted according to the number of words and voice intonation;

[0055] The split characters are recorded as "character groups", and the actual meaning of the "character groups" is determined by matching the split "character groups" with the database;

[0056] The second step is to combine the results of learning a large number of text groups with the audio at the time of recording, and conduct deep learning based on the actual text translation to generate a "primary speech recognition model";

[0057] The third step is to obtain the facial expression of the user during the speech process, and associate the obtained facial expression with the corresponding speech;

[0058] The image acquisition device inside the computer acquires the facial image of the audio recording person, processes the acquired image, reduces the noise in the image, enhances the texture in the image, and then recognizes the processed image;

[0059] The fourth step is to associate the collected facial expression data with the text corresponding to the voice translation. In the process of combining the facial expression data with the text during translation, the sentiment analysis mechanism is introduced to identify the tone and attitude of the voice inputter through the sentiment analysis mechanism, thereby increasing the success rate of voice recognition, thereby correcting the "primary voice recognition model" and obtaining the "voice recognition model";

[0060] Create different databases and "speech recognition models" for different accents;

[0061] The fifth step is to import the acquired audio into the "speech recognition model" in real time during the speech recognition process, and perform real-time speech recognition through the "speech recognition model".

[0062] A computer speech recognition method based on the first embodiment performs speech recognition input, and the results of learning a large number of "text groups" are combined with the audio during input to generate a "primary speech recognition model". The "primary speech recognition model" is corrected through deep learning of the facial expressions of the inputter, and finally a "speech recognition model" is obtained. The "speech recognition model" is used to realize rapid recognition of speech and greatly improve the recognition accuracy. The test sentences are shown in the following table:

[0063] Number of input statements Number of sentences Total Word Count Single sentence recognition accuracy Single word recognition accuracy 5 29 539 83.7％ 89.5％ .

[0064] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A computer speech recognition method, characterized in that: The following steps are involved: Step 1: Establishing an audio database: Establish an audio database by inputting audio through the Internet and manually, and translate the corresponding text based on the audio; Step 2: Character group learning: The computer is equipped with an embedded processing chip, which is used to split the recorded audio into "single words", "double words", "triple words"... until "the entire text of the sentence", and the split text is recorded as a "character group". The actual meaning of the "character group" is determined by matching the split "character group" with the database; Step 3: Model building: Combine the results of learning a large number of text groups with the audio during recording, and conduct deep learning based on actual text translation to generate a "primary speech recognition model"; Step 4: Model correction: Manually input voice, then use the "primary voice recognition model" for voice recognition, select different environments for input, compare the text recognized by the "primary voice recognition model" with the actual input text, find out the differences between the text, and correct the "primary voice recognition model" based on the differences; Step 5: Image acquisition: The facial expression of the person being recorded is acquired during the speech process, and the acquired facial expression is associated with the corresponding speech; Step 6: Associating the collected facial expression data with the text corresponding to the voice translation, so as to further revise the revised "primary voice recognition model" and obtain the "voice recognition model"; Step 7: Real-time speech recognition: During the speech recognition process, the acquired audio is imported into the "speech recognition model" in real time, and real-time speech recognition is performed through the "speech recognition model".

2. A computer speech recognition method according to claim 1, characterized in that: When the audio database is being built, the phrase database is also being built.

3. A computer speech recognition method according to claim 2, characterized in that: When comparing the "text group" with the phrase database, the phrases in the phrase database are sorted according to the number of characters and the tone of voice.

4. A computer speech recognition method according to claim 3, characterized in that: A noise reduction module is installed in the embedded processing chip to reduce the noise in the received voice. At the same time, the noise reduction module cooperates with the "speech recognition model" to fully reduce the noise of the recorded voice.

5. A computer speech recognition method according to claim 4, characterized in that: The image acquisition device inside the computer acquires the facial image of the audio recording person, processes the acquired image, reduces the noise in the image, enhances the texture in the image, and then recognizes the processed image.

6. A computer speech recognition method according to claim 5, characterized in that: Introduce sentiment analysis mechanisms in the process of combining facial expression data with text during translation.

7. A computer speech recognition method according to claim 6, characterized in that: The sentiment analysis mechanism is used to identify the tone and attitude of the voice entrant, thereby increasing the success rate of voice recognition.

8. A computer speech recognition method according to claim 7, characterized in that: Create different databases and "speech recognition models" for different accents.

Images