Toward flexible speech recognition-recent progress at Tokyo Institute of Technology

Abstract

This paper describes the progress at Tokyo Institute of Technology and the author's perspectives for making speech recognition systems more flexible at both the acoustic and linguistic processing levels. Specifically, it describes a broadcast news transcription system, multimodal human-computer interface, neural-network-based HMM adaptation for noisy speech, online incremental speaker adaptation combined with automatic speaker-change detection, message-driven speech recognition and understanding, a Japanese national project on spontaneous speech corpus and processing technology, and speech summarization. For processing spontaneous speech, paradigm shift from speech recognition to understanding where underlying messages of the speaker are extracted will be indispensable, instead of transcribing all the spoken words. Building a large corpus of spontaneous speech to construct reliable acoustic and linguistic models is also crucial. Due principally to the technology of making computers smaller, more powerful and cheaper, the ubiquitous and wearable computing era is expected to come into being in the initial years of the 21st century. In such an environment, speech recognition will be widely used as one of the principal methods of human-computer interaction.