Presentation 1999/6/18
Statistical Language Model Incorporating Kana-Characters and Phrases
Masao Utiyama, Hiroshi Matsumoto,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) This papre proposes a type of statistical language model which deals with out-of-vocabulary words (OOVs) encountered in very large continuous speech recognition. The proposed language model has a dictionary which contains syllables and phrases as well as words. By using this model, OOVs in speech signals are recognized as sequences of syllables. Then, the transcribed syllable sequences are translated into ordinary words by using a large dictionary which may contain OOVs of the speech recognition system. The experiments showed that the recognition accuracy of a bigram language model increased from 64.35% to 72.53% and that of a trigram language model increased from 70.73% to 82.14%. The recall of the detection of sentences which contain OOVs is 93.0% and the precision is 91.7%. The recall of the detection of OOVs 75.4% and the precision is 78.3%. The ratio of successfully transcribed OOVs is 65.3%.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) statistical language model / speech recognition / out of vocabulary word
Paper # SP99-38
Date of Issue

Conference Information
Committee SP
Conference Date 1999/6/18(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Speech (SP)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Statistical Language Model Incorporating Kana-Characters and Phrases
Sub Title (in English)
Keyword(1) statistical language model
Keyword(2) speech recognition
Keyword(3) out of vocabulary word
1st Author's Name Masao Utiyama
1st Author's Affiliation Shinshu University()
2nd Author's Name Hiroshi Matsumoto
2nd Author's Affiliation Shinshu University
Date 1999/6/18
Paper # SP99-38
Volume (vol) vol.99
Number (no) 121
Page pp.pp.-
#Pages 8
Date of Issue