Audio Feature Extraction and Classification Based on Wavelet Transform

摘要

There are many methods to classify audio signals through extracting features and in this paper, discrete wavelet transform is proposed. The extracted features include centroid, bandwidth, sub-band energy and silence ratio. Linear discriminant functions are constructed through features analysis. Then the audio signals are classified into 4 groups consisting of pure speech, speech with music, music, and environment sounds. The result shows that it can achieve satisfied classification accuracy while save computation time.