AUDIO SEGMENTATION BASED ON THE WAVELET TRANSFORMATION

In content-based audio retrieval, the audio segmentation is one of essential steps. Segmenting audio in terms to certain features will facilitate computer processing and make it easier to carry on recognition and retrieval. This paper does the wavelet transformation to audio and analyzes its approximate coefficients, then extract two wavelet-based features which are the mean approximate amplitude and the approximate zero-crossing rate. Experiments of segmentation are presented with four kinds of audio types, speech, pure music, song and silence. The results indicated that this method is not only simple but also can achieve a good effect.