EFFECTIVENESS OF SHORT-TERM PROSODIC FEATURES FOR SPEAKER VERIFICATION
Iker Luengo, Eva Navas, Inma Hernáez, Jon Sánchez, Ibon Saratxaga, Iñaki Sainz · 2006
In this work a traditional MFCC based speaker verification system is combined with a prosody based one to determine whether simple short-term prosodic information is useful for improving current state-of-theart ASV. The traditional speaker verification system based in spectral information has an EER of 3.85% when using 1024 mixtures. The prosody based system uses short-term intonation and energy information and achieves an EER of 23.93 % with 128 mixtures. After applying LDA and fusing those scores, a final EER of 3.84 % is achieved. This result does not show a significant improvement when compared with the result of the traditional speaker verification system.