CD-HMM algorithm performance for speaker identification on an Italian database

The focus of this work is on the performance analysis of a text dependent closed set speaker identification system for the Italian language. Two identification algorithms, based on LPC and LPC-cepstral feature extractors followed by a continuous density hidden Markov model (CD-HMM) classifier, have been implemented and tested on the Italian database SIVA the MUSER. The database consists of 360 phone calls made by 20 different male speakers from different Italian regions. The false identification probability for the two algorithms has been evaluated for different training sets, different spoken words and a variable number of states of the CD-HMM classifier. Results show that, in any of the considered conditions, the LPC-cepstral based system performs better than the LPC based one and that, in the best working condition, the false identification probability turns out to be of the order of 1.5 per cent