Model training method and device, voice recognition method and device, server and storage medium

Abstract

The invention relates to model construction in artificial intelligence, and provides a model training method and device, a voice recognition method and device, a server and a storage medium. The model training method comprises the steps: carrying out first signal processing on voice data to obtain first voice data, and carrying out second signal processing on the voice data to obtain second voice data; inputting the first voice data and the second voice data into a feature extraction model to extract a first feature vector of the first voice data and a second feature vector of the second voice data; calculating mutual information between the first voice data and the second voice data according to the first feature vector and the second feature vector; according to mutual information between the first voice data and the second voice data, updating model parameters of the feature extraction model until the feature extraction model converges; and fusing and finely adjusting the converged feature extraction model and the trained speech recognition model to obtain a target speech recognition model. According to the invention, the robustness of the voice recognition model can be improved.

Description

technical field [0001] The present application relates to the technical field of model building, and in particular to a model training method, speech recognition method, device, server and storage medium. Background technique [0002] With the continuous development of the new media industry, the channels of voice data are gradually diversified, and there are different bandwidths and encoding formats. For example, voice data is recorded data with a sampling rate of 8k or 16k, or encoding formats such as ulaw, Alaw, and amr. In some cases, during the transmission of the voice data, processing such as compression is also performed on the voice data. These have brought difficulties and challenges to speech recognition. [0003] The current speech recognition model can only recognize speech data of a single channel. For application scenarios with speech data of different channels, it is necessary to train multiple speech recognition models that match the speech data of each cha...

AI Technical Summary

Problems solved by technology

These have brought difficulties and challenges to speech recognition

[0003] The current speech recognition model can only recognize speech data of a single channel. For application scenarios with speech data of different channels, it is necessary to train multiple speech recognition models that match the speech data of each channel. The robustness of the speech recognition model Poor, and because the training data of different speech recognition models cannot be shared, resulting in a large difference in the accuracy of each speech recognition model, or requiring more training data, there are major drawbacks

Images