Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

Merging search spaces for subword spoken term detection

Poster at Interspeech 2009 - ICSLP, tenth International Conference on Spoken Language Processing, Brighton, 6.-10. September 2009
: Mertens, T.; Schneider, D.; Köhler, J.

Fulltext urn:nbn:de:0011-n-1059681 (159 KByte PDF)
MD5 Fingerprint: 4cbaf5ad38f59abc0c4a62bc66cc08a8
Created on: 6.10.2009

2009, 4 pp.
International Conference on Spoken Language Processing (ICSLP) <10, 2009, Brighton>
Poster, Electronic Publication
Fraunhofer IAIS ()
speech recognition; spoken term detection; pronunciation variation

We describe how complementary search spaces, addressed by two different methods used in Spoken Term Detection (STD), can be merged for German subword STD. We propose fuzzysearch techniques on lattices to narrow the gap between subword and word retrieval. The first technique is based on an edit-distance, where no a priori knowledge about confusions is employed. Additionally, we propose a weighting method which explicitly models pronunciation variation on a subword level and thus improves robustness against false positives. Recall is improved by 6% absolute when retrieving on the merged search space rather than using an exact lattice search. By modeling subword pronunciation variation, we increase recall in a high precision setting by 3% absolute compared to the edit-distance method.