音声を対象とした雑音強度依存の多段接続砂時計型ニューラルネットワーク雑音除去フィルタ

概要

論文の詳細を見る
An adaptive noise reduction filter composed of Cascaded Sandglass-type Neural Network (CSNNRF) is proposed to develop a hearing aid appliance. The number of unit sandglass-type neural networks (SNNs) is controlled adaptively by noise intensity. Usually the hearing aid works outside where noise intensity is altering all the times. An adaptive noise reduction filter controlled by noise intensity is essential to cope with these circumstances. Each SNN has a three-layer structure and consists of the same number of neural units in the input and output layers and a single neural unit in the hidden layer.SNN are connected in cascaded to be CSNNRF. The number of unit SNNs is adaptively determined by our algorithms so that hearing intelligibity for speech signal may be processed more preferablely. To determine the number of SNNs, we regard significance on the improvement of intelligibility more than numerical value itself such as S/N ratio of speech signal.
社団法人電気学会の論文
2003-03-01