Speech Emotion Recognition Based on Wavelet Transform and Improved HMM

摘要

We proposed a novel speech emotion recognition method by use of Wavelet Transform and Hidden Markov Model (HMM) to classify five discrete emotional states: anger, fear, joy, sadness and surprise. The system is comprised of three main parts, a preprocessing part, a feature extracting part and a recognition part. In the feature extracting part, due to Fourier Transform uses fixed sized windows, we consider using Wavelet Transform to extract the emotion features. In the recognition part, we use improved HMM as the emotion recognizer. We test this method in the Chinese corpus of emotional speech synthesis database. The test result shows that the method is effective and high speed.