For Full-Text PDF, please login, if you are a member of IEICE,|
or go to Pay Per View on menu list, if you are a nonmember of IEICE.
Committee-Based Active Learning for Speech Recognition
Yuzo HAMANAKA Koichi SHINODA Takuya TSUTAOKA Sadaoki FURUI Tadashi EMORI Takafumi KOSHINAKA
IEICE TRANSACTIONS on Information and Systems
Publication Date: 2011/10/01
Online ISSN: 1745-1361
Print ISSN: 0916-8532
Type of Manuscript: PAPER
Category: Speech and Hearing
active learning, query by committee, LVCSR, progressive alignment,
Full Text: PDF(584KB)>>
We propose a committee-based method of active learning for large vocabulary continuous speech recognition. Multiple recognizers are trained in this approach, and the recognition results obtained from these are used for selecting utterances. Those utterances whose recognition results differ the most among recognizers are selected and transcribed. Progressive alignment and voting entropy are used to measure the degree of disagreement among recognizers on the recognition result. Our method was evaluated by using 191-hour speech data in the Corpus of Spontaneous Japanese. It proved to be significantly better than random selection. It only required 63 h of data to achieve a word accuracy of 74%, while standard training (i.e., random selection) required 103 h of data. It also proved to be significantly better than conventional uncertainty sampling using word posterior probabilities.