Speech dereverberation method based on deep features of generative adversarial network

Abstract

The present invention discloses a speech de-reverberation method based on the deep features of the generated confrontational network. The main steps of the method are: firstly, performing weighted prediction error WPE on the speech and performing signal preprocessing; secondly, data feature extraction: performing MFCC on the speech data Feature extraction and phoneme-specific bottleneck feature BNF extraction without reverberant speech; then WPE speech signal preprocessing: build a generative confrontation network, and the MFCC features containing reverberant speech are mapped to the deep features of clean speech through the generative confrontation network; finally through The Kaldi toolbox uses traditional GMM‑HMM for forced alignment, and then uses deep neural networks for acoustic model training and decoding. The present invention combines the signal processing method and the deep learning framework based on the generation confrontation network, so that the system can combine the respective advantages of the two to produce a better speech reverberation effect.

Description

technical field [0001] The invention relates to the field of speech signal processing, especially for the problem of recognition performance degradation caused by environmental reverberation in far-field speech recognition, and proposes a speech reverberation method based on deep features of generative adversarial networks. Background technique [0002] In recent years, emerging industries such as smart home, dialogue robots, and smart audio have flourished, which has brought about great changes in people's lifestyles and the way people interact with machines. Voice interaction, as a new way of interaction, has gained popularity in these emerging fields a wide range of applications. With the application of deep learning in speech recognition, the recognition performance has been greatly improved, the recognition rate has exceeded 95%, and the recognition effect has basically reached the level of human hearing. However, the above are limited to near-field conditions, and the...

AI Technical Summary

Problems solved by technology

Although the current neural network method can establish a good nonlinear mapping, it is difficult to achieve the desired effect only by using a fully connected neural network. Secondly, the use of the most basic feature mapping method cannot learn speech very well. Semantic information, constructing a good network structure and studying the deep semantic information of speech will produce a good recognition performance improvement for speech recognition, which has practical significance for speech recognition in complex scenes

Images