以估測型態為基準之下三角型轉換器於語音編碼之應用

Abstract

在本文中, 我們提出用轉換編碼的方式來做語音編碼,我們使用了 一種名稱為"以估測型態為基準之下三角型轉換器"(PLT)。 它和Kahurene - Loeve 轉變 ( KLT ) 十分相似, 它們擁有許多相同的特質 。但是在運算複雜度上,PLT來的比KLT小。在這裡我們用了一種新的架構 來實驗我們的理論系統,此架構我們稱之為MINLAB。使用這個架構,我們 可以使我們的系統能夠有(1) 噪音增益是唯一(unitary) (2) 系統完美的 實現 (3) 可以使用於有損耗或者沒有損失的壓縮

In this thesis, we proposed a transform coding method for speech coding,that is using a new nonunitary transform named Prediction-based Lower triangular Transform(PLT)\cite{PLT} to speech coding. It is like the Kahurene-Loeve Transform(KLT) that they has the same decorrelation property. But PLT's implementational cost is less than one half of KLT. The PLT can be factorized into simple building blocks. We implement PLT by using the minimum noise structure that makes the coding gain of PLT is the same as KLT's. The minimum noise structure has the following properties. (i)Its noise gain is unity. (ii)Structurally PR implementation. (iii)It can be apply for lossy or lossless compression. Performance of this transform coding method will be test by some speech data and compare with other compression ways. 1.1 Outline 1.2 Notations 2 The Prediction-based Lower Triangular Transform(PLT) 2.1 Preliminaries and Review 2.2 The Prediction-based Lower Triangular Transform(PLT) 2.3 Using Ladder to implement PLT 3 MINLAB in PLT 3.1 Minimum Noise Structure for Ladder-Based Biorthogonal (MINLAB) Coders 3.2 MINLAN for PLT 3.3 PLT for Lossless Data Compression 3.4 PLT for AR(1) Inputs 4 Application of PLT in speech coding 4.1 Quantizer 4.2 PLT Coder 4.3 Compare with KLT 4.4 Compare with ADPCM 5 Conclusion