Read/Search this Article
Abstract
従来から,音声認識における騒音対策法はいくつか提案されてきたが,本研究では,騒音源の方向が特定できない高騒音環境下において,騒音スペクトルの事前情報やピッチ検出が不要な,単一マイクロホンのみを用いる騒音除去法について検討を行う。具体的には,実環境下での音声認識を可能にするため,騒音変動のガウス性や白色性などの前提を必要とせず,しかも騒音の統計情報が未知でかつ音声の時間的変動形態が未知の場合でも適用できる新たな騒音除去法を提案する。さらに,騒音混入下での音声データに本手法を適用することにより,実験的にもその有効性を確認している。
Several noise removal methods for speech recognition have been proposed up to now. In this study, a method to remove the noise by use of single microphone unnecessary to introduce the priori-information on noise spectrum and pitch detection is considered under the existence of noise with high amplitude and the situation that cannot clarify the direction of noise source. More specifically, a new type of noise removal method applicable to the actual speech recognition that the statistics of noise are unknown and fluctuation form of time series of speech signal is also unknown is proposed in the practical form without imposing an assumption of ideal Gaussian white noise. Furthermore, the effectiveness of the proposed method is confirmed by applying it to actual speech signals contaminated by noises.
Journal
- Technical report of IEICE. EA [List of Volumes]
-
Technical report of IEICE. EA 110(285), 67-72, 2010-11-11 [Table of Contents]
The Institute of Electronics, Information and Communication Engineers
Share