作成者 |
|
|
|
|
本文言語 |
|
出版者 |
|
|
発行日 |
|
収録物名 |
|
巻 |
|
開始ページ |
|
終了ページ |
|
会議情報 |
|
出版タイプ |
|
アクセス権 |
|
Crossref DOI |
|
概要 |
In throat microphone (TM), two skin attached piezo-electric sensors can capture speech sound signals from the tissue vibration. Because of their small bandwidth, throat microphone recorded speech is r...obust to the surrounding noise but suffers from intelligibility and naturalness problems. This study addresses the issue of improving the perceptual quality of the throat microphone speech is based on the statistical mapping between the features of TM and AM speech using the Artificial Neural Network approach for correction of vocal tract parameters and spectral envelope. The target is for natural man machine communication especially for vocal tract affected people. This paper exploits the nonlinear mapping property of Multi-Layered Feed Forward Neural Network (MLFFNN) for estimation of high-frequency components (4-8kHz) from the low-frequency band (0-4kHz) of TM signal. The proposed algorithm is tested using ATR503 Dataset. The simulation results show a noticeable performance in the field of speech communication in adverse environments.続きを見る
|