Two-Stage Speech/Non-Speech Classification of Telephone Signals

摘要

Robust speech/non-speech classification is very useful in the pre-processing stage for speech recognition and content-based audio retrieval where the database is composed of various audio files. In this paper, a two-stage speech/non-speech classification algorithm for telephone signals is provided by combining three simple methods. Short-time energy plus pitch period is used in the first stage and the output is well classified at the second stage which applies the AdaBoost algorithm and MFCC features. Experiments show the effectiveness and efficiency of the algorithm.