基於Gram-Schmidt正交化的MPEG Surround降混音器與去相關器之設計

Abstract

MPEG Surround乃一項低位元率多聲道音訊壓縮標準。其壓縮的原理是透過降混音(down-mix)處理將多聲道訊號耦合成雙聲或單聲道訊號，並計算出聲源定位的空間參數(spatial parameter)，來達到減少聲道數與紀錄聲場之目的。解碼端透過去相關器產生的去相關訊號(decorrelated signal)，並根據空間參數進行升混音(up-mix)處理來重建聲源定位與空間寬廣度的環繞效果。因此，編碼端如何將多聲道耦合成雙聲或單聲道訊號，與去相關訊號的生成，將影響重建聲音的品質。 此篇論文，使用Gram-Schmidt正交化的概念，改進去相關器與二轉一降混音模組的運作方式。方法的改進效果將透過升混音後訊號與原始訊號的聲道間能量差與相關性差值變化，及ODG(Objective Difference Grade)客觀量測與MUSHRA主觀測試來驗證。

MPEG Surround (MPS) coding is an efficient method for multichannel audio coding. In an MPS encoder, downmixing from multichannel signals into a less number of channels is an efficient way to achieve high compression rate. In decoder, an upmixing module combining with the decorrelator is the key module to reconstruct the multichannel signals. This thesis considers the design of the downmixer and the decorrelator with the assistance of the Gram-Schmidt orthogonal process. The performance of the proposed downmixer and decorrelator is verified through the differences of Channel Level Difference (CLD) and Inter-Channel Coherence (ICC) between the upmixed signals and the original signals. Also, the objective and subjective quality measurements are conducted.