End-to-end long-time speech recognition method

Abstract

The invention provides an end-to-end long-time speech recognition method. The method comprises the following steps: selecting a corpus as a training data set, and carrying out data preprocessing and feature extraction on voice data in the training data set to generate voice features; constructing an improved RNN-T model fusing an external language model and a long-term speech recognition algorithm, and inputting the speech features into the RNN-T model for training to obtain a trained improved RNN-T model; and taking the trained improved RNN-T model as a teacher model in a mutual learning knowledge distillation algorithm, training a student model in the mutual learning knowledge distillation algorithm by using the mutual learning knowledge distillation algorithm, identifying long-term voice data to be identified by using the trained and verified student model, and outputting a voice identification result. According to the method, the external language model, the long-term speech recognition algorithm module and the RNN-T model are fused, so that the robustness and generalization ability of long-term speech recognition of the model are improved, and the time and space complexity of the algorithm is optimized.

Description

technical field [0001] The invention relates to the technical field of speech recognition, in particular to an end-to-end long-term speech recognition method. Background technique [0002] Voice, as the most direct and effective way of information transmission, is the most important way for people to communicate with each other and convey their thoughts. Automatic Speech Recognition (ASR) technology refers to correctly recognizing speech signals as corresponding text content or commands, allowing machines to understand human language and perform related operations. With the wide application of computers, ASR technology has become a key technology to realize simple and convenient human-computer intelligent interaction, and has gradually become a hot research field. With the advancement and development of deep learning and speech recognition technology, the speech recognition model based on end-to-end has significantly improved the accuracy of speech recognition compared with...

AI Technical Summary

Problems solved by technology

In the process of training the RNN-T model, the lack of linguistic knowledge and insufficient language modeling ability (that is, the insufficient training of the prediction network) make the training of the RNN-T model difficult.

[0010] (2) The robustness of long-term speech recognition is poor

However, the effect of sequence-level knowledge distillation is easily affected by information such as parameter facilities and model initialization, and the model generalization ability is poor.

Images