Speech recognition model training method, speech recognition method and system

Abstract

The embodiment of the invention discloses a training method of a speech recognition model, a speech recognition method and a system, and relates to the technical field of speech recognition. This embodiment includes: inputting audio training samples into an acoustic encoder, encoding and representing the audio training samples, and determining the acoustic coding state vector; inputting a preset vocabulary into a language predictor to determine a text prediction vector; Input the text mapping layer to obtain the text output probability distribution; calculate the first loss function according to the target text sequence corresponding to the audio training sample and the text output probability distribution; input the text prediction vector and the acoustic encoding state vector into the joint network to calculate the second loss function , perform iterative optimization according to the first loss function and the second loss function until the stopping condition is satisfied. This embodiment adjusts the training and prediction process of the speech recognition model, improves the modeling capability of the semantic recognition model, and thus improves the accuracy of the speech recognition model.

Description

technical field [0001] The present application relates to the technical field of speech recognition, in particular to a speech recognition model training method, speech recognition method and system. Background technique [0002] The speech recognition model based on Transducer has been widely used at home and abroad, and its typical feature is that it can directly adapt to streaming speech recognition tasks. Although it introduces a language predictor, its language modeling ability is insufficient. After research, it is found that the language predictor does not play a role similar to a language model in real reasoning, but more assumes the function of eliminating duplicate labels. There is room for further improvement in the ability to model dependencies between languages. Contents of the invention [0003] In order to solve the above technical problems or at least partly solve the above technical problems, embodiments of the present invention provide a speech recogniti...

AI Technical Summary

Problems solved by technology

Although it introduces a language predictor, its language modeling ability is insufficient. After research, it is found that the language predictor does not play a role similar to a language model in real reasoning, but more assumes the function of eliminating duplicate labels. There is room for further improvement in the ability to model dependencies between languages

Images