FS1016的低複雜度碼簿搜尋演算法

Abstract

隨著對於語音封包傳輸方式的需求量增加，低位元率語音編碼器逐漸成為研究的趨勢，在1991年㆓月，美國通用服務管理部門公佈了聯邦標準1016 (FS1016)，它制定了利用4.8kbps碼簿激發線性預測編碼(CELP)將類比語音轉換成數位資料時的需求；然而，當將其使用在即時系統㆗的最大障礙是在於它執行碼簿搜尋時的大量運算複雜度。 在這篇論文㆗，我們會提出新的碼簿搜尋演算法來改善FS1016的運算複雜度，我們不但會利用運算次數的數量來作為比較我們所提出的新演算法與原來演算法的依據，還會透過將演算法實作成微軟視窗系統㆘的FS1016 ACM驅動程式的方式來驗證其結果。為了完結我們實驗室在語音封包傳輸領域㆗的研究，我們會將快速碼簿搜尋演算法與過去所提出的遺失封包重構演算法整合於同㆒個ACM驅動程式。 i

With the increasing demand for packet-voice transmission, low bit-rate speech coders gradually become a research trend. In February 1991, General Services Administration published Federal Standard 1016 (FS1016) which specifies the requirements for the conversion of analog voice to digital data by a method of 4.8 Kbps CELP. However, the major obstacle for applying it to real-time applications is the mass computational complexity in codebook search. In this thesis, we improve the FS1016 complexity by presenting a new codebook searching algorithm. The improvement of our new algorithm over the original one is not only illustrated by the reduction of the number of computations required in principle, but also demonstrated through a so-implemented FS1016 ACM driver under Windows Operating System. To conclude the research on the area of packet-voice transmission in our lab, a combined ACM driver with our fast searching algorithm, and a packet-loss recovery scheme developed previously has also been implemented.