Presentation | 2011-10-06 Language Modeling for Automatic Speech Recognition of Lectures in Elementary and Secondary Education Hiroaki NANJO, Ippei HISAKI, Yuki WADA, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | Automatic speech recognition (ASR) of lectures on elementary and secondary education is addressed. Most of conventional studies of lecture speech recognition target on lectures in universities or oral presentations in technical conferences, in which lecturers make their speech for adult audiences. On the contrary, in elementary school or junior high-school, lecture audience is immature people. Lecturers (teachers) often make utterances in a different way from talks to adult audiences. Specifically, teachers try to select easy words and phrases, some of which are only for kids. For ASR of elementary school lectures, a language model which covers such linguistic phenomena is required. In this paper, suitable vocabulary and language model for elementary school lectures are discussed. Word 3-gram language model trained with texts for adults (Corpus of spontaneous Japanese and one-year newspaper articles) cannot cover a half of 3-grams (about 3000 kinds) appeared in 13 lectures in school. We got higher adjusted testset perplexity about 343. Word 3-gram language model trained with small texts for kids (1.2M words from kids-oriented web sites), we can cover one-third of 3-grams, which are not modeled in the language model for adult. We confirmed that it is significant to collect text corpora for ASR of elementary school lectures. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | Lecture Speech Recognition / Information Support / Elementary and Secondary Education / Language Model / Speaking Style |
Paper # | SP2011-54,WIT2011-36 |
Date of Issue |
Conference Information | |
Committee | WIT |
---|---|
Conference Date | 2011/9/29(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 | Well-being Information Technology(WIT) |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | Language Modeling for Automatic Speech Recognition of Lectures in Elementary and Secondary Education |
Sub Title (in English) | |
Keyword(1) | Lecture Speech Recognition |
Keyword(2) | Information Support |
Keyword(3) | Elementary and Secondary Education |
Keyword(4) | Language Model |
Keyword(5) | Speaking Style |
1st Author's Name | Hiroaki NANJO |
1st Author's Affiliation | Faculty of Science and Technology, Ryukoku University() |
2nd Author's Name | Ippei HISAKI |
2nd Author's Affiliation | Faculty of Science and Technology, Ryukoku University |
3rd Author's Name | Yuki WADA |
3rd Author's Affiliation | Faculty of Science and Technology, Ryukoku University |
Date | 2011-10-06 |
Paper # | SP2011-54,WIT2011-36 |
Volume (vol) | vol.111 |
Number (no) | 226 |
Page | pp.pp.- |
#Pages | 6 |
Date of Issue |