A detection approach to search-space reduction for HMM state alignment in speaker verification

Authors
Citation
Q. Li, A detection approach to search-space reduction for HMM state alignment in speaker verification, IEEE SPEECH, 9(5), 2001, pp. 569-578
Citations number
28
Language
INGLESE
art.tipo
Article
Categorie Soggetti
Eletrical & Eletronics Engineeing
Journal title
IEEE TRANSACTIONS ON SPEECH AND AUDIO PROCESSING
ISSN journal
1063-6676 → ACNP
Volume
9
Issue
5
Year of publication
2001
Pages
569 - 578
Database
ISI
SICI code
1063-6676(200107)9:5<569:ADATSR>2.0.ZU;2-C
Abstract
To support speaker verification (SV) in portable devices and in telephone s ervers with millions of users, a fast algorithm for hidden Markov model (HM M) alignment is necessary, Currently, the most popular algorithm is the Vit erbi algorithm with beam search to reduce search-space; however, it is diff icult to determine a suitable beam width beforehand. A small beam width may miss the optimal path while a large one may slow down the alignment. To ad dress the problem, we propose a nonheuristic approach to reduce search-spac e. Following the definition of the left-to-right HMM, we first detect the p ossible change-points between HMM states in a forward-and-backward scheme, then use the change-points to enclose a subspace for searching, The Viterbi algorithm or any other search algorithm can then be applied to the subspac e to find the optimal state alignment. Compared to a Full-search algorithm, the proposed algorithm is about four times Faster while the accuracy is st ill slightly better in an SV task; compared to the beam search algorithm, t he proposed algorithm can provide better accuracy with even lower complexit y. In short, for an HMM with S states, the computational complexity can be reduced up to a factor of S/3 with slightly better accuracy than in a full- search approach. This paper also discusses how to extend the change-point d etection approach to large-vocabulary continuous speech recognition.