Speech understanding model generation method and intelligent speech interaction method based on pragmatic information

Abstract

The invention discloses a speech understanding model generation method and an intelligent speech interaction method based on pragmatic information, and the speech understanding model generation method comprises the steps: processing speech, and obtaining speech data in a coding sequence form; presetting pragmatic information classification nodes; correlating the voice data in the coding sequence form with pragmatic information classification nodes; and generating a voice understanding model by using the voice data in the coding sequence form and the pairing data of the pragmatic information classification nodes. According to the method, pragmatic information is directly understood from voice, and information loss caused by the fact that the voice is transferred into characters is avoided; the method is not limited by characters, and a set of voice interaction architecture and a corresponding model can support various language environments such as different dialects, small languages and mixed languages; training corpora and a training voice understanding model are collected according to the hierarchy of pragmatic information classification nodes of voice interaction, and the data volume required by training is greatly reduced; and through simple association operation, the voice obtained in voice interaction is used for rapid iteration of the voice understanding model.

Description

technical field [0001] The present invention relates to speech intelligent processing technology, in particular to a speech understanding model generation method and an intelligent speech interaction method based on pragmatic information. Background technique [0002] The current natural language understanding (NLU) is realized on the basis of automatic speech recognition (ASR) technology. Speech recognition is the transcription of speech into text. That is to say, first obtain the text sequence corresponding to the speech from the speaker's speech through speech recognition, and then obtain the speaker's meaning (semantics) expressed by the speech by analyzing the text sequence. [0003] On the one hand, speech recognition technology itself has its own technical difficulties. For the same words, words, phrases and sentences, due to the influence of the speaker's pronunciation characteristics, speaking habits, speaking scenes, context, real-time emotions and other factors,...

AI Technical Summary

Problems solved by technology

Although the current speech recognition technology can train specific speech recognition models for specific dialects, specific accents, and even specific people, it cannot realize automatic switching between these models, so it cannot fundamentally solve the problem of inaccurate recognition caused by individual pronunciation. accurate question

[0004] On the other hand, the current natural language processing (NLP) performs semantic analysis on the basis of the text sequence obtained from speech recognition. Whether it is through keyword matching technology, context or knowledge graph, or deep learning technology, it cannot make up for it. Inaccurate semantic understanding caused by deviation of speech recognition results

Images