Low-resource speech recognition and keyword search (KWS) are important topics for speech technologies. However, their performance often suffers from out-of-vocabulary (OOV) keywords. Subword units like syllables are useful in handling this issue. This report introduces a weighted finite state transducer (WFST) based syllable transduction framework for OOV handling in KWS. Syllable lattices are generated by performing syllable decoding and OOV keywords are entered into a pronunciation dictionary using a word-to-syllable pronunciation prediction system. Syllable lattices are then transduced into word lattices using both in-vocabulary word pronunciations and OOV pronunciations. Experiments on 5 languages provided by IARPA Babel project are presented, and it is shown that syllable transduction can effectively spot OOV keywords. Combination of this approach with two other OOV handling methods further improves keyword search performance.
Syllable based Lattice Transduction for Keyword Search
Researchers may make free and open use of the UC Berkeley Library’s digitized public domain materials. However, some materials in our online collections may be protected by U.S. copyright law (Title 17, U.S.C.). Use or reproduction of materials protected by copyright beyond that allowed by fair use (Title 17, U.S.C. § 107) requires permission from the copyright owners. The use or reproduction of some materials may also be restricted by terms of University of California gift or purchase agreements, privacy and publicity rights, or trademark law. Responsibility for determining rights status and permissibility of any use or reproduction rests exclusively with the researcher. To learn more or make inquiries, please see our permissions policies (https://www.lib.berkeley.edu/about/permissions-policies).