An end-to-end long-term speech recognition method

Abstract

The invention provides an end-to-end long-term speech recognition method. The method includes: selecting a corpus as a training data set, performing data preprocessing and feature extraction on the speech data in the training data set, and generating speech features; constructing an improved RNN-T model that integrates an external language model and a long-term speech recognition algorithm, and The speech features are input into the RNN-T model for training, and the trained and improved RNN-T model is obtained; the trained and improved RNN-T model is used as the teacher model in the mutual learning knowledge distillation algorithm, and the mutual learning knowledge distillation algorithm is used Train the student model in the mutual learning knowledge distillation algorithm, use the trained and verified student model to recognize the long-term speech data to be recognized, and output the speech recognition result. The present invention integrates the three parts of the external language model, the long-term speech recognition algorithm module and the RNN-T model, improves the robustness and generalization ability of the long-term speech recognition model, and optimizes the time and space complexity of the algorithm. Spend.

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] As the most direct and effective way of information transmission, voice is the most important way for people to communicate with each other and convey their thoughts. Automatic Speech Recognition (ASR) technology refers to the correct recognition of 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 popular research field. With the advancement and development of deep learning and speech recognition technology, the end-to-end speech recognition model has significantly improved the accuracy of speech recognition compared to tr...

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