A Study on Speech Emotion Recognition Based on CCBC and Neural Network

This paper described a novel speech emotion recognition approach aiming at improving speech emotion recognition rate. Seven discrete emotional states (anger, disgust, fear, joy, neutral, sadness, surprise) are classified throughout the work. Firstly, series preprocessing of speech signals are done. Secondly, extracting features are done, and then we consider incorporating Canonical Correlation Based on Compensation (CCBC) to cope with the mismatch between training and test set. The mismatch between training and test conditions can be simply clustered into three classes: differences of speakers, changes of recording channel and effects of noisy environment. Finally, we evaluated the system using Back-propagation Neural Networks (BPNN). Results are given using the Chinese Corpus of emotional speech synthesis database, recognition experiments show that the method is effective and high speech for emotion recognition.