Speech emotion recognition method combining CGAN spectrogram denoising and bilateral filtering spectrogram enhancement

Abstract

The invention discloses a speech emotion recognition method combining CGAN spectrogram denoising and bilateral filtering spectrogram enhancement. The speech emotion recognition method comprises the steps: S1, acquiring a clean speech spectrogram and a noise-added speech spectrogram; S2, inputting the clean speech spectrogram and the noise-added speech spectrogram into a conditional generative adversarial network based on a matrix distance for training to obtain a denoising model; S3, de-noising the noise-added speech spectrogram by using the de-noising model, respectively performing two different scales of bilateral filtering to obtain a low-scale filtering graph and a high-scale filtering graph, multiplying the difference between the low-scale filtering graph and the high-scale filteringgraph by an enhancement coefficient, and adding the result to the low-scale filtering graph to obtain a detail-enhanced spectrogram; S4, inputting the detail-enhanced spectrogram into a convolutionalneural network model for classification to obtain a classification model; and S5, processing the spectrogram of the voice to be recognized in the S3, and inputting the obtained detail-enhanced spectrogram into the classification model to obtain a voice emotion classification result. According to the invention, speech emotion recognition is effectively realized.

Description

technical field [0001] The invention belongs to the field of speech recognition, and mainly relates to the field of human-computer interaction, specifically, a speech emotion recognition method combined with CGAN spectrogram denoising and bilateral filter spectrogram enhancement. Background technique [0002] Applying speech emotion recognition to the human-computer interaction system, on the one hand, can make the robot have "emotion" like human beings, detect the other party's emotional changes through hearing, and communicate and interact with humans more naturally and intelligently, making the humanized and natural The interaction mode integrating automation and intelligence endows a new type of human-computer interaction system. On the other hand, the system based on voice emotion can provide more novel development ideas for medical treatment, machinery, education, and service, and will further enrich people's daily life, become a human helper, and help people solve pra...

AI Technical Summary

Problems solved by technology

[0003] Speech is often accompanied by various noises in practical applications, and the existence of noise will cause a series of impacts on speech emotion recognition, which will reduce the quality of speech and emotional information.

Images