A data processing method, device, equipment and storage medium

Abstract

Embodiments of the present invention disclose a data processing method, device, device, and storage medium, wherein the method includes: a server acquires a sequence of speech frames according to a preset time window, determines characteristic information of the sequence of speech frames, and determines according to the characteristic information The input information of the first layer of time-truncated self-attention network, for any layer of time-truncated self-attention network, input the input information and the output of the previous layer of time-truncated self-attention network into any layer of time-truncated self-attention network. Self-attention network to train the speech recognition model and get the trained speech recognition model. Through the above embodiment, the input information of the time-truncated self-attention network of the first layer is input into the time-truncated self-attention network of each layer for training, thereby reducing the loss of speech feature information during the training of the speech recognition model. , to improve the speech recognition rate of the speech recognition model.

Description

technical field [0001] The present application relates to the technical field of speech recognition based on artificial intelligence, and in particular, to a data processing method, apparatus, device and storage medium. Background technique [0002] Speech recognition technology, the purpose of which is to receive human speech signals and make machines responsible for converting the speech signals into text. For speech processing, the entire process can be divided into four parts: front-end processing, acoustic model modeling, language model and dictionary modeling, and decoding. [0003] With the research and development of artificial intelligence technology, especially deep learning, the current speech recognition is divided into three types, one is the probability model method, the second is the deep learning method, and the third is the application of self-attention mechanism. For the probabilistic model method, the model cannot use the context information of each frame...

AI Technical Summary

Problems solved by technology

For the probabilistic model method, the model cannot use the context information of each frame, that is, it cannot use historical information to assist the current task; for the deep learning method, although the model can achieve a better convergence effect, due to the recurrent neural network (Recurrent Neural Network , RNN) itself has a loop structure, more RNN units make the training time longer, and it is difficult to parallelize; and the current self-attention mechanism overcomes the above-mentioned problems to a certain extent, but it is affected by the voice in this method. The time windowing technique of the signal will lead to the loss of information

Images