Third International Conference on Spoken Language Processing (ICSLP 94)

Yokohama, Japan
September 18-22, 1994

A Keyword-Spotting Unit for Speaker-Independent Spontaneous Speech Recognition

Yasuyuki Masai (1), Jun'Ichi Iwasaki (1), Shin'Ichi Tanaka (1), Tsuneo Nitta (1), Masahiro Yao (2), Tomohiro Onogi (2), Akira Nakayama (2)

(1) Multimedia Engineering Laboratory, TOSHIBA Corporation, Japan (2) TOSHIBA Computer Engineering Corporation, Japan

In this paper, we describe a real-time keyword-spotting unit (KeySpot) with an adaptive noise-canceller for speaker-independent, spontaneous speech recognition in noisy environments. KeySpot consists of a DSP (TMS320C30) for adaptive noise-cancellation and acoustic analysis, a special LSI for statistical matrix quantization (SMQ), two SPARC chips ('SPARC1' and 'SPARC2') for HMM based keyword-spotting, and a SPARC chip ('SPARC3') for syntactic analysis. KeySpot was tested under two conditions: a speaker-independent large-vocabulary isolated word recognizer, and a speaker-independent small-vocabulary word spotter. Evaluation results have shown that KeySpot can be used for the speaker-independent, 1000 isolated word recognizer with an accuracy of 96.3%, as well as the 90 word vocabulary word spotter with an accuracy of 94.4% with a response time of 0.3 sec.

Full Paper

Bibliographic reference.  Masai, Yasuyuki / Iwasaki, Jun'Ichi / Tanaka, Shin'Ichi / Nitta, Tsuneo / Yao, Masahiro / Onogi, Tomohiro / Nakayama, Akira (1994): "A keyword-spotting unit for speaker-independent spontaneous speech recognition", In ICSLP-1994, 1383-1386.