Automatic speech recognition method and system based on artificial intelligence

A technology of automatic speech recognition and artificial intelligence, applied in speech recognition, speech analysis, instruments, etc., can solve problems such as low efficiency of professional vocabulary recognition and inaccurate recognition of professional vocabulary, so as to reduce professional misunderstandings, improve professionalism, and improve search speed effect

Active Publication Date: 2021-06-11
深圳奇实科技有限公司
8 Cites 0 Cited by

AI-Extracted Technical Summary

Problems solved by technology

[0006] The present invention provides an automatic speech recognition method and system based on artificial intelligence to solve the probl...
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Method used

In addition, by setting the noise judgment value of described voice input signal, if the noise judgment value G of each signal data is greater than preset judgment threshold value, then this signal data is judged as noise point, the data set M in Each signal data is judged one by one. If it is a noise point, it will be eliminated. If it is not a noise point, it will be retained. The retained signal data will form the final processed signal. The noise points are further reduced by way of the noise judgment value to improve the quality of the voice input signal, thereby providing high accuracy of automatic voice recognition.
The beneficial effect of above-mentioned technical scheme is: the scheme that adopts the present embodiment to provide can improve the precision and accuracy rate to professional vocabulary recognition, especially strengthen the accuracy, precision of video conferencing record in professional field, especially professional Improve the professionalism of enterprises in relevant professional fields, and more importantly, reduce professional misunderstandings caused by automatic recognition of professional vocabulary and speech recognition, and prevent misunderstandings caused by speech recognition and cause major losses. At the same time, since the vocabulary classification template is used as the basis, the search rate of professional vocabulary is improved, and the recognition efficiency of automatic speech for professional vocabulary is improved.
The working principle and beneficial effect of above-mentioned technical scheme are: the scheme that present embodiment adopts is the process that spelling error is carried out to the text of input, after passing through acoustic model and language model, the text of its output may have the mistake of spelling For problems of other forms, in order to ensure the accuracy and professionalism of automatic speech recognition, it is necessary to correct the spelling of the output text. By setting the spelling error correction model, it is ensured that the final output text has no formal spelling errors and automatic speech recognition is improved. the accuracy.
The working principle of above-mentioned technical scheme is: the scheme that the present embodiment adopts is to carry out random sampling by the speech characteristic parameter that speech signal to be recognized extracts, and obtains recognition result based on acoustic model and language model to the parameter of sampling, again to all The above recognition results are based on the vocabulary classification template to judge whether it belongs to the speech recognition involving professional vocabulary. If so, it means that the speech signal to be recognized is speech related to the profession, and the recognition of such speech requires a relatively professional vocabulary library to provide the basis support. Therefore, the speech feature parameters are input to the professional vocabulary acoustic model and the professional vocabulary language model, the comprehensive information is decoded through the search of the output layer, and the corresponding text is output; The weight of vocabulary has been re-matched to increase the probability of obtaining professional vocabulary. For speech recognition that does not belong to professional vocabulary, ordinary automatic speech recognition technology is carried out, that is, the speech feature parameters are input into the acoustic model and language model, the recognition result is obtained through decoding search, and the corresponding text is output.
The working principle of above-mentioned technical scheme is: the scheme that the present embodiment adopts is to carry out random sampling by the speech characteristic parameter that speech signal to be recognized extracts, and obtains recognition result to the ...
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Abstract

The invention discloses an automatic speech recognition method and system based on artificial intelligence. The method comprises the steps: comparing vocabularies in a recognition result with professional vocabularies in a vocabulary classification template by employing the vocabulary classification template, acquiring the proportion of the professional vocabularies in the vocabularies in the recognition result, and judging whether professional vocabulary speech recognition is needed or not according to the proportion. According to the scheme provided by the invention, the accuracy and accuracy of the recognition of the professional vocabularies can be improved, particularly the accuracy, precision and specialty of video conference records in the professional field are enhanced, and the specialty of enterprises in the related professional field is improved, More importantly, misunderstanding of vocabulary recognition caused by automatic voice recognition of professional vocabularies is reduced, and heavy loss caused by misunderstanding of voice recognition is prevented. Meanwhile, on the basis of the vocabulary classification template, the search rate of the professional vocabularies is improved, and then the recognition efficiency of the automatic voice for the professional vocabulary is improved.

Application Domain

Technology Topic

Image

  • Automatic speech recognition method and system based on artificial intelligence
  • Automatic speech recognition method and system based on artificial intelligence
  • Automatic speech recognition method and system based on artificial intelligence

Examples

  • Experimental program(10)

Example Embodiment

[0080] Example 1:
[0081] The embodiment of the present invention provides an automatic voice recognition method based on artificial intelligence. figure 1 For a flow chart of an automatic speech recognition method based on artificial intelligence, please refer to figure 1 The method includes the following steps:
[0082] Step S101 receives the voice signal to be identified;
[0083] Step S102, the speech signal to be identified is pre-processed to obtain a voice input signal;
[0084] Step S103, perform the conversion of the speech input signal to the frequency domain, extract speech feature parameters;
[0085] Step S104, random samples of the speech feature parameters, obtain several sample feature parameters;
[0086] Step S105, input the sample feature parameters to the acoustic model and the language model, and the decoded search acquisition recognition result is obtained;
[0087] In step S106, the identification result is input to the vocabulary classification template, and the vocabulary in the identification result is compared to the professional vocabulary in the word collection classification template, and the proportion of professional vocabulary in the identification result is obtained;
[0088] In step S107, it is judged whether or not the predetermined value is exceeded; if the result is that the result is that step S108 is performed, if the result of the determination is not, step S109 is performed.
[0089] Step S108, input the speech feature parameter to a professional vocabulary sound model and a professional vocabulary language model, and the output layer is decoded, and the corresponding text is decoded, and the corresponding text; the professional vocabulary model and the professional vocabulary language model The weight of professional vocabulary has been reappeared, and the probability of obtaining professional vocabulary;
[0090] Step S109, input the speech feature parameter into the acoustic model and the language model, and the decoding search acquires the identification result, and outputs the corresponding text.
[0091] The operation principle of the above technical solution is that the scheme employed in this embodiment is to randomly sample the speech feature parameters extracted by the identification speech signal, and the sampled parameters acquire the identification result, again to the identification result. Based on vocabulary classification templates, whether or not they belong to speech recognition involving professional vocabulary. If it is, the speech signal to be identified is to provide basic support for such voice to provide a relatively professional vocabulary. Therefore, the speech feature parameters are input to the professional vocabulary sound model and the professional vocabulary language model. After the search for the output layer, the comprehensive information is decoded, and the corresponding text; the professional vocabulary sound model and the professional vocabulary language model The weight of the vocabulary has been reappeared and the probability of obtaining professional vocabulary. When it is judged that the speech recognition is not a professional vocabulary, the normal automatic speech recognition technology is performed, and the speech feature parameter is input to the acoustic model and the language model. After decoding the search, the identification result is obtained, and the corresponding text is output.
[0092] It should be noted that the speech input signal performs the conversion of the time domain to the frequency domain, extracts the speech feature parameters, and can be used to extract the speech characteristics in the manner, which can be used, and obtain the spectral spectrum by Meer frequency. Then, the sound spectrum is filtered through the filter.
[0093] Further, the extraction of the speech features can also be extracted by the principle of depth convolutional neural network to obtain speech feature parameters.
[0094] In addition, a brief introduction and description of the vocabulary classification template, the phrase classification template contains a professional vocabulary database, and the professional vocabulary in different industries is also different. Therefore, different categories of professional vocabulary can be set according to the different industries. The desired professional vocabulary classification can be searched from different sorting databases.
[0095] The scheme provided in this embodiment can be applied very widely, such as meeting records involving a highly professional industry meeting, and some products are displayed automatic identification, if they involve video or recording in the professional field, etc. An occlusion that requires automatic speech recognition can be employed in this embodiment.
[0096] The beneficial effects of the above technical solution are: the schemes employed in this embodiment can improve the accuracy and accuracy of the professional vocabulary recognition, especially the accuracy, precision, especially professional, and improvement of video conference records in the field of professional. The company's professionalism in the relevant professional areas is to reduce the major misunderstandings due to automatic identification of professional vocabulary, prevent misunderstanding due to speech recognition. At the same time, due to the basis of vocabulary classification templates, improve the search rate of professional vocabulary, thereby increasing the identification efficiency of automatic speech of professional vocabulary.

Example Embodiment

[0097] Example 2:
[0098] Based on the first embodiment, after the corresponding text, the corresponding text is included, including:
[0099] Enter the text of the output to the spelling error correction model, obtain the text after error correction;
[0100] Output of the error after the end text is output.
[0101] The working principle and beneficial effect of the above technical solution is that the scheme used in this embodiment is the process of spelling the error correction of the input text. After the acoustic model and the language model, the text thereof may exist in the form of a spelling error. Question, in order to ensure the accuracy and professionalism of automatic voice recognition, the spelling of the output text is required to correct the output of the output. By setting the spelling error correction model to ensure that the final text of the output does not have the form of spelling errors, improve the accuracy of automatic voice recognition. .

Example Embodiment

[0102] Example 3:
[0103] Based on the first embodiment, the word constituent template construction method includes:
[0104] Get a large number of professional vocabulary that belong to different industries;
[0105] The professional vocabulary uses convolutional neural networks to classify training in accordance with the industry to which the professional vocabulary belongs;
[0106] Get the classification result and store the classification result in a classified database, and constitute the vocabulary classification template.
[0107]The operation principle of the above technical solution is that the scheme used in this embodiment is a description of the construction method of the vocabulary classification template. By getting a large number of professional vocabulary, the convolutional neural network is used to classify the above-mentioned professional vocabulary according to the industry-based, which is different from the professional vocabulary contained in different industries. Different professional vocabulary will be conducted by vocabulary classification templates. Classify and store the classification results in the classified database, which is easy to query the corresponding professional vocabulary during subsequent processes.
[0108] The beneficial effects of the above technical solution are: Sumber and classify professional vocabulary by using the solutions provided in this embodiment, and improve the search rate of professional vocabulary, and improve the identification efficiency of automatic voice of professional vocabulary by improving the search rate of professional vocabulary by means of vocabulary classification templates. . In addition, the scheme provided by this embodiment can improve the accuracy and accuracy of professional vocabulary recognition, in particular, the accuracy, precision, especially professionalism of video conferencing in the field, and improves enterprises in the relevant professional fields. Professionalism, more importantly, reduced the professional misunderstanding due to the automatic identification of professional vocabulary, preventing significant loss due to misunderstanding due to speech recognition.
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

no PUM

Description & Claims & Application Information

We can also present the details of the Description, Claims and Application information to help users get a comprehensive understanding of the technical details of the patent, such as background art, summary of invention, brief description of drawings, description of embodiments, and other original content. On the other hand, users can also determine the specific scope of protection of the technology through the list of claims; as well as understand the changes in the life cycle of the technology with the presentation of the patent timeline. Login to view more.
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Similar technology patents

Pretreatment kit for salbutamol sample based on immunomagnetic beads

PendingCN108508199AReduce interference from background substancesImprove accuracy and precisionMaterial analysisChemistryImmunomagnetic bead
Owner:沭阳康源泰博生物科技有限公司

Classification and recommendation of technical efficacy words

  • Improve accuracy and precision
  • Improve accuracy

Cassette-based dialysis medical fluid therapy systems, apparatuses and methods

InactiveUS20050209563A1Improvement for dialysisImprove accuracyMedical devicesPeritoneal dialysisAccuracy improvementDialysis
Owner:BAXTER INT INC +1
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products