Comparison OF Wavelet Based Feature Extraction Methods for Speech/Music Discrimination

The speech/music discrimination systems have gaining importance in several intelligent audio retrieval algorithms due to the increasing size of the multimedia sources in our daily lives. This study aims to propose a speech/music discrimination system which utilizes the advantages of the wavelet transform. Also, the performance of the discrete wavelet transform and the dual- tree wavelet transform has been compared with the conventional time, frequency and cepstral domain features used in speech/music discrimination. The speech and music samples collected from common databases, CD recording and internet radios have been classified with artificial neural networks with different feature sets. The principal component analysis has been applied to eliminate the correlated features before classification stage. Considering the number of vanishing moments and orthogonality, the best performance has been obtained with Daubechies8 wavelet among the other members of the Daubechies family. According to the results, the proposed feature set outperforms the traditional ones.