Research on Speech Emotion Recognition Based on the Fractional Fourier Transform
Lirong Huang, Xizhong Shen
Abstract
Speech emotion recognition is an important part of human–computer interaction, and the use of computers to analyze emotions and extract speech emotion features that can achieve high recognition rates is an important step. We applied the Fractional Fourier Transform (FrFT), and then constructed it to extract MFCC and combined it with a deep learning method for speech emotion recognition. Since the performance of FrFT depends on the transform order p, we utilized an ambiguity function to determine the optimal order for each frame of speech. The MFCC was extracted under the optimal order of FrFT for each frame of speech. Finally, combining the deep learning network LSTM for speech emotion recognition. Our experiment was conducted on the RAVDESS, and detailed confusion matrices and accuracy were given for analysis. The MFCC extracted using FrFT was shown to have better performance than ordinal FT, and the proposed model achieved a weighting accuracy of 79.86%.