Speech recognition method suitable for noise environment

Abstract

The invention discloses a speech recognition method suitable for a noise environment. According to the speech recognition method, a beam-forming device which integrates minimum variance undistorted response beam-forming and diagonally loaded beam-forming is established for noisy speech signals collected by double micro microphone arrays, the computational complexity is reduced by using a recursivematrix inversion method, and after beam-forming is carried out, a speech signal which is preliminarily denoised is obtained; then a post-modulation domain spectral subtraction method further processes the preliminarily denoised speech signal, the residual noise is removed, the speech distortion is reduced, and the final denoised speech signal is obtained; and moreover, the speech recognition method adopts a convolution neural network to train speech models and extract deep features of speech. According to the speech recognition method, the problem of speech recognition rate decline in the noisy environment is solved, and the speech recognition method has the advantage of good robustness, can be applied to aspects such as household robots, intelligent speakers and speech devices working inthe noisy environment.

Description

technical field [0001] The invention relates to the technical field of voice recognition, in particular to a voice recognition method suitable for noise environments. Background technique [0002] Since the 21st century, my country's speech recognition research has developed rapidly, and a considerable number of outstanding companies have emerged, reaching the world's leading level in some fields and producing products with a high market share. It has been widely used among tourists. Currently, acoustic models based on deep neural networks have significantly improved the performance of speech recognition, especially in near-field conditions. However, far-field and reverberant speech recognition remains a challenging problem for practical applications. [0003] In the practical application environment, robust speech recognition is a common concern in the fields of signal processing and speech recognition, and it is one of the most challenging tasks in recent decades. One ma...

AI Technical Summary

Problems solved by technology

[0005] The object of the present invention is to aim at the problem that the recognition rate drops sharply under the existing noise environment, and provide a kind of speech recognition method applicable to the noise environment, this method can effectively remove the noise component in the noise speech signal under the actual environment, through The convolutional neural network is used to extract high-dimensional hidden features from the original speech data, effectively deal with the variability and richness of the data, and reduce the parameters of neural network training

Images