作者: A. R. Abu-El-Quran , J.S. Gammal , R. A. Goubran , A. D. C. Chan
DOI: 10.1109/ICSENS.2007.4388565
关键词:
摘要: In this paper, we propose a robust talker identification (TI) system that can identify speakers in different reverberant environments. We describe reverberation sensing (RSS) determines the approximate level of surrounding environment, and then selects TI engine trained similar environment; greater accuracy is achieved when training test environments are similar. system, there six engines five one non-reverberant environment. Performance evaluated using new twelve environments, which all within time (RT60) range 0 to 1.8 seconds. The RSS achieves average 95.5%; an enhancement 9% over