INTERSPEECH 2004 - ICSLP
For efficient organization of speech recordings -- meetings, interviews, voice mails, and lectures -- the ability to search for spoken keywords is essential. Today, most spoken document retrieval systems use large-vocabulary recognition. For the above scenarios, such systems suffer from the unpredictable domain and out-of-vocabulary queries. In previous work, we presented phoneme-lattice based vocabulary-independent search of spontaneous speech. In this paper, we propose to combine word-based and phonetic search into a hybrid, and explore two ways: posterior combination (merging search results of a word and a phoneme based system) and prior combination (combining word and phoneme LMs / vocabularies to form a hybrid recognizer). Our best phonetic baseline (FOM 64%) is improved by the hybrid approach to FOM 73% (word-level LM matching test set domain) and 71% (domain mismatch case).
Bibliographic reference. Yu, Peng / Seide, Frank Torsten Bernd (2004): "A hybrid word / phoneme-based approach for improved vocabulary-independent search in spontaneous speech", In INTERSPEECH-2004, 293-296.