Multi-dialect accent Mandarin speech recognition model training method, device and equipment

Abstract

The present application provides a multi-dialect accent Mandarin speech recognition model training method, device and equipment, which relate to the technical field of language recognition. The method includes: obtaining training samples; using labeled standard Mandarin speech data to train to obtain an initial acoustic model, and using text data to train to obtain an initial language model; iteratively training the initial acoustic model based on unlabeled dialect accent Mandarin speech data to obtain a target acoustic model ; Use the text to be trained recognized by the target acoustic model and the initial language model to train a temporary language model, and combine the temporary language model and the initial language model to obtain the target language model; combine the target acoustic model and the target language model into a multi-dialect accent Mandarin speech recognition model. Using a large number of unlabeled Mandarin speech data with dialect accents, iterative training was carried out to obtain a speech recognition model for Mandarin with multi-dialect accents, which improved the accuracy of Mandarin speech recognition for dialect accents.

Description

technical field [0001] The present invention relates to the technical field of speech recognition, in particular to a multi-dialect Mandarin speech recognition model training method, device and equipment. Background technique [0002] With the improvement of the performance of the Internet and other mobile terminals, intelligent products based on speech recognition technology are more and more popular in industrial production and daily life, such as voice dialogue robots, voice assistants, interactive tools, etc. However, my country is a country with multiple dialects. People living in various regions will have a large degree of accent when expressing in Mandarin, which will cause a mismatch with the standard Mandarin model and result in low speech recognition accuracy. [0003] At present, multi-model research can be carried out based on temporal neural networks to realize the recognition of Mandarin accents. Among them, based on the multi-model recognition method, differe...

AI Technical Summary

Problems solved by technology

[0005] The object of the present invention is to, aiming at the deficiencies in the above-mentioned prior art, provide a kind of multi-dialect accent Mandarin speech recognition model training method, device and equipment, so that make full use of the speech data without label to strengthen the training of the model, avoid in practice Due to the limitation of the lack of labeled training sample data in the application, the final recognition accuracy is not high

Images