PCNN spectrogram feature integration based emotion voice recognition system

Abstract

The invention relates to the technical field of voice recognition and provides a PCNN spectrogram feature integration based emotion voice recognition system. According to the invention, windowing andframing are performed on voice signals, discrete Fourier transform is performed and a spectrogram of the voice signals is drawn; a PCNN model is constructed and the spectrogram is processed through apulse coupling neural network; performing convolution on a PCNN graph and 5-scale 8-direction Gabor wavelets and extracting Gabor amplitude features so as to obtain 40 Gabor spectrograms; extracting uniform mode LBP features on each Gabor spectrogram and acquiring a feature vector as indicated in the description through cascading connection of histograms in the 40 Gabor spectrograms. According tothe invention, Fourier analysis is made on the voice signals; the voice signals are converted to the spectrogram; the spectrogram is processed through the pulse coupling neural network; 40 Gabor spectrograms are obtained through convolution on the spectrogram and the 5-scale 8-direction Gabor wavelets; a local two-value mode feature and a local Hu matrix feature are extracted; and a support vectormachine is adopted for classification and recognition after the two features are integrated.

Description

technical field [0001] The invention relates to the technical field of speech recognition. Background technique [0002] With the rapid development of information technology, more and more attention has been paid to human-computer interaction. Emotional speech recognition, as a key technology of human-computer interaction, has become the focus of research in this field. Emotional speech recognition is a speech recognition technology in which computers extract and analyze the emotional information of human voices to make judgments on human emotional states. It is widely used in many fields such as business, medical care, and education. [0003] At present, the acoustic features used for speech emotion recognition can be roughly summarized into prosodic features, spectrum-based correlation features and timbre features. Prosodic feature distinguishes speech emotion through features such as duration, fundamental frequency, and energy, and its emotion recognition ability has bee...

AI Technical Summary

Problems solved by technology

Among these features, the time-domain characteristics and frequency-domain characteristics of the speech signal play an important role, but there are relatively few studies on the combination of the time-domain and frequency-domain correlation of the speech signal.

Images