A method and system for intelligent speech recognition based on deep learning

Abstract

The present invention proposes an intelligent speech recognition method based on deep learning. Firstly, the speech information is acquired; the noise elimination is performed on the acquired speech information by using a fusion noise elimination model, and the noise elimination is obtained. The fusion noise elimination The model is obtained by combining two noise elimination models with the voice endpoint detection algorithm; the voice information after noise elimination is input into the staged learning enhancement network structure, and the enhanced voice information is obtained; the staged learning enhancement network structure includes multiple The target layer, the target layer adopts a linear activation function, and the hidden layer is an LSTM-RNN network; the enhanced voice information is input into the voice model for voice recognition; the method provided by the invention can eliminate noise while retaining necessary Target voice, improve the robustness of voice enhancement in various complex environments, with a small amount of calculation.

Description

technical field [0001] The field of speech recognition of the present invention particularly refers to an intelligent speech recognition method and system based on deep learning. Background technique [0002] In recent years, the artificial intelligence boom triggered by deep learning is affecting and changing people's lifestyles. People are no longer satisfied with the human-computer interaction of a single text and instruction, but look forward to the more convenient voice interaction. Fast way to communicate. Voice has become an indispensable information medium. However, in the actual transmission process of speech, background noise and human voice interference will have a certain impact on speech, which will reduce the quality and intelligibility of speech, and also bring challenges to subsequent applications, such as speech recognition and speaker recognition. Wait. In a complex application environment, as the front-end interface of voice applications, voice signal p...

AI Technical Summary

Problems solved by technology

Compared with traditional speech, the speech enhancement algorithm based on DNN (Deep Neural Network) can achieve great performance improvement, especially in the case of dealing with non-stationary noise. However, the supervised speech enhancement algorithm based on DNN is in practice In the face of real noise scenes, differences in speaking styles, and low signal-to-noise ratio (Signal-to-Noise Ratio), there are generalization problems, such as voice loss, low intelligibility, etc.

Images