8th International Conference on Spoken Language Processing

Jeju Island, Korea
October 4-8, 2004

Multi-Pass ASR using Vocabulary Expansion

Katsutoshi Ohtsuki (1), Nobuaki Hiroshima (1), Shoichi Matsunaga (2), Yoshihiko Hayashi (3)

(1) NTT Corporation, Japan
(2) Nagasaki University, Japan
(3) Osaka University, Japan

Current ASR systems have to limit its vocabulary size depending on available memory size, expected processing time, and available text data for building a vocabulary and a language model. Although vocabularies of ASR systems are designed to achieve high coverage for expected input data, it can not be avoided that input data includes out-of-vocabulary (OOV) words that is OOV problem. In this paper, we propose dynamic vocabulary expansion using conceptual base and multi-pass speech recognition using the expanded vocabulary. Relevant words to content of input speech are extracted based on a speech recognition result obtained using a reference vocabulary. An expanded vocabulary that includes less OOV words is built by adding the extracted words to the reference vocabulary. The second recognition process is performed using the new vocabulary. The experimental results for broadcast news speech show the proposed method achieves 30% reduction of OOV rate and improve speech recognition accuracy.

Full Paper

Bibliographic reference.  Ohtsuki, Katsutoshi / Hiroshima, Nobuaki / Matsunaga, Shoichi / Hayashi, Yoshihiko (2004): "Multi-pass ASR using vocabulary expansion", In INTERSPEECH-2004, 1713-1716.