Abstract
Discrimination between speech and music signals is an important problem in efficient digital radio broadcasting, particularly for variable bit rate applications such as Internet radio. This paper presents a speech/music discrimination system based on a Mel frequency cepstral coefficient (MFCC) front end and a GMM classifier. This system can be used to select the optimum coding scheme for the current frame of an input signal without knowing a priori whether it contains speech-like or music-like characteristics. An analysis of speech and music error rates for different numbers of MFCCs (from 8 to 28) is presented. For the 46 minute evaluation database used in this experiment, an accuracy of up to 97.14% for music and 93.87% for speech can be attained.