Research And Application Of Text Error Detection And Correction After Speech Recognition

With the rise of online conferences,MOOCs,and self-media videos,the demand for offline long audio generation of subtitles has exploded.The traditional human-based voice transcription method can no longer meet the demand.It is the general trend to apply voice recognition technology to subtitle production.Although speech recognition technology has made great progress in recent years,the recognition of long audio speech still cannot meet the needs of the market.This is mainly due to the fact that long audio speech is difficult to guarantee because of its large content information and voice quality.In practical applications,the same pronunciation of words and phrases often corresponds to different text content due to different usage environments or users.Errors will inevitably occur.The academic research on long audio speech mainly focuses on improving speech recognition technology.However,due to different users and different contexts in practical applications,the current speech recognition technology alone cannot meet the needs of users.Therefore,research on text error detection and correction after long audio speech recognition is still very valuable.This paper proposes a text error correction method based on contextual keywords and user characteristics for speech recognition based on the recognition context of long audio speech recognition and the inaccurate recognition of different users.The context,semantics,pronunciation rules,and user characteristics are integrated into the error correction system to detect and correct the text after speech recognition to improve the accuracy of the text.The main research work and results are:(1)Aiming at the problem of context recognition in long audio speech recognition,the relationship between context core words and context keywords is studied.Using semantic knowledge plus a few context core words,a context knowledge base is constructed,and on this basis,context is used the knowledge base reduces the number of words in the confusion set and improves the speed of the algorithm.(2)Aiming at the problem of repetitive content recognition errors when different users use speech recognition software,the Chinese pronunciation rules and user speaking habits are studied.Adding pronunciation rules and user's speech habits to error detection and correction makes the system "learning" and can continuously improve accuracy during error detection and correction.(3)Designed a speech recognition system based on context keywords and user characteristics,and applied the text error detection and correction method based on the context keywords and user characteristics to the speech recognition practice of teaching videos.Optimize the recognized text.After the system test,the text error correction method after speech recognition based on context keywords and user characteristics proposed in this paper is effective and feasible.Compared with traditional error correction methods,the accuracy rate,recall rate and F1 value of the method proposed in this paper have been improved.