A configuration of remote control system using speech within a priori known noise

概要

論文の詳細を見る
Not only surrounding noise but also the sound radiated by an equipment itself seriously affect the performance of speech recognition when one tries to make voice control system for sound radiating equipment such as a television set. In this paper, a configuration of speech recognition system for this kind of equipment is proposed. This system consists of three components;an adaptive digital filter to reduce noise radiated by the equipment, a speech candidate detector, and an isolated word recognizer. In the experiments not only white noise but also colored noise such as organ solo and male voice are used. In all cases, over 90% of correct recognition ratio is obtained even when speech level is the same as noise level.
社団法人日本音響学会の論文