In this report, a new feature extraction method is proposed for speech linguistic information. The new method consists of two stages based on the Time-Frequency Distribution which is one of the nonstationary signal analysis. In the first stage, the accoustic analysis using the Time-Frequency Distribution is implemented as the preprocessing of feature extraction in order to reflect dynamic characteristics of speech signals. In the second stage, the translation from the time-frequency two-dimensional representation into another two-dimensional representation is implemented for getting the important speech characteristic easilly. As the results on some experiments, it is shown that the extracted feature via the proposal method has good availability for the classification of vocal sounds.
展开▼