Creation of Time-Varying Voiceprint Database
Linlin Wang, Thomas Fang Zheng · 2010
Performance degradation with time varying in speaker recognition is a generally acknowledged phenomenon and it is widely assumed that speaker models should be updated from time to time to maintain representativeness. However, lack of a longitudinal voiceprint database which specially focuses on the time-varying effect has prevented researchers from finding out reasons behind this phenomenon. In this paper, an on-going voiceprint database creation project is presented, aiming to examine solely the time-varying impact on speaker recognition and explore hidden factors causing possible performance degradation. In this speech database, speakers are requested to utter in a reading way with fixed prompt texts instead of free-style conversations throughout 16 sessions in a period of about three years. Sessions are of gradient time intervals where initial ones are of shorter time intervals and following ones of longer and longer time intervals. Initial experimental results on the partially completed database are also presented, which demonstrated time-varying effect evidently.