Browse by topic
Type of publication
Improved unsupervised speech recognition system using MLLR speaker adaptation and confidence measurement. V Jornadas en Tecnologías del Habla (VJTH'2008), 2008. pp. 255-258.Mukund Jha and Sourabh Sriom and M'iriam Luj'an-Mares and Carlos-D. Mart'inez-Hinarejos and Alberto Sanch'is Improved unsupervised speech recognition system using MLLR speaker adaptation and confidence measurement. V Jornadas en Tecnologías del Habla (VJTH'2008) Bilbao (SPAIN), Nov, 2008 A robust ASR system needs to perform well in different environment and with different speakers. For this reason speaker adaptation has become an essential part of a state of art ASR system. Here we show how conﬁdence measurement technique can be used to improve the quality of unsupervised speaker adaptation. An initial speaker independent system is adapted to improve the modelling of a new speaker by modifying HMM parameters using Maximum Likelihood Linear Regression(MLLR) technique. Improvement gained from unsupervised speaker adaptation technique are lowered because of their dependency on the accuracy of recognition in ﬁrst pass. We use conﬁdence measures to improve the performance by selective adaptation. We present experimental results on the 8 speakers’ data from Wall Street Journal.