於時域調變域之消除迴響及噪音方法

Abstract

消除迴響和環境噪音一直都是語音訊號處理中很重要的議題，然而以傳統抵銷卷積演算法來達到消除迴響的效果是一連串複雜的過程。本論文中，我們將以機器學習的方式處理迴響訊號和環境噪音，嘗試將時域卷積的過程以近似的方式簡化，以深度類神經網路學習消除迴響的過程。語音理解度與語音品質皆是評量語音訊號處理的重要評量標準，我們同時考慮此兩種評量標準，目的為達到同時提升語音理解度與語音品質，有別於以往以機器聽覺(machine hearing)為最終目的的演算法，我們將人類聽覺(human hearing)視為最終目的，發展同時考慮振幅頻譜(magnitude spectrogram)與相位頻譜(phase spectrogram)之演算法，以複數頻譜之時域變化為特徵，我們提出在時域調變域中聯合消除迴響和環境噪音的演算法，藉由與其他演算法做比較並且嘗試使用不同的資料庫，以分析演算法的效能。

De-reverberation and De-noise to cancel the reverberant effect and environment noise has always been an important task in speech processing, however, time-domain deconvolution algorithms often require a series of complicated pro-cesses and provide no good results. In this thesis, we propose de-reverberation and de-noise algorithms in the temporal modulation domain using a machine learning technique. Inspired by human auditory processing, the time domain convolution operation was first transformed to the temporal modulation domain and a deep neural network (DNN) was used to learn how to de-reverberate and de-noise speech signals in that domain. For human hearing applications, enhancing speech intelli-gibility and speech quality is more critical than enhancing spectral profiles, which are important to machine hearing applications. we propose a joint processing which de-reverberates and de-noises in the temporal modulation domain. Consequently, we analyze the performance of each method by comparing the scores of speech intelli-gibility and quality using two speech corpora.