ISCA Archive Odyssey 2001
ISCA Archive Odyssey 2001

Frequency analysis of speaker identification

Özgür Devrim Orman, Levent M. Arslan

Our main motivation in this work is to investigate subbands based representation of speaker identities and then search for improvements to typical methods (such as MFCCs). Test results obtained via proposed Vector Ranking criteria, have shown that 0-1000 Hz and 3000-4500 Hz frequency bands are more significant in automatic speaker discrimination when compared to other frequencies. We propose a new filter bank for speaker identification systems. Performances of new cepstrum coefficients and MFCCs are compared which show that the proposed feature set results in significantly better SI performance.