音訊修補技術應用於暫態雜訊緩解與封包丟失補償

Abstract

由於科技進步免持式通訊系統(hands-free communication systems)、人工電子耳(Cochlear implants)、機器聽覺(machine listening)、通訊網路電話(voice over internet protocol)等較新穎的系統應用蓬勃發展；其中有效降低噪聲(noise reduction)、音訊品質提升(audio enhancement)與通訊網路封包丟失補償(packet loss concealment in VOIP)為主要面臨的問題，本論文主要特色是假設已知干擾音與丟失封包位置為前提，利用兩旁鄰近音訊在時頻域平面上的單弦波特性(harmonic property)進而實施強度線性內插(magnitude spectrum linear interpolation)並結合短時傅立葉變換自偶性質(STFT-self-duality)促使其相位復原(phase retrieval)，希望經過一些優化的參數設定能夠達成找到最長可修復的丟失音訊片段且緩解暫態雜訊之目的。

From the past till now, several applications of navel system have developed vigorously such as hands-free communication systems, cochlear implants, machine listening and voice over internet protocol. Moreover, most of significant technic's problems of those applications are noise reduction, audio enhancement and packet loss concealment in VOIP. In this thesis, we concentrate the problem on restoration quality based on the location of interference sound assumed to be known. On the other hand, the trade- off between time’s or frequency’s resolution will influence the quality of magnitude interpolation. Through setting the parameter of window’s waveform and using the harmonic property of vowel that we transfer the problem into phase retrieval based STFT-self-duality, which we would like to find out the longest length for restoration.