- 學位論文
混合激發線性預估編碼器之聯合訊源與通道編碼
Joint Source-Channel Coding for MELP Coder
摘要
訊源編碼（source coding）和通道編碼（channel coding）這兩種技術是概念完全相反的方法，前者是使原始語音可以被壓縮的很小而還不失一般品質，後者則是把原始資訊加大，而多加的資訊是為了在傳輸發生錯誤時，可藉由此多加的資訊把錯誤偵測出來或甚至修正回來。本論文使用混合激發線性預估（Mixed Excitation Linear Prediction ，MELP）編碼器與軟輸出維特比演算法（Soft-Output Viterbi Algorithm，SOVA）來達成聯合訊源與通道解碼（Joint Source-Channel Decoding，JSCD）的目的，方法是將混合激發線性預估編碼器輸出的位元統計特性當作是軟輸出維特比演算法的事前機率輸入值，此種通道解碼方式稱為聯合訊源與通道解碼的軟輸出維特比演算法（JSCD SOVA），再加上使用非均勻錯誤保護（Unequal Error Protection，UEP）與均勻錯誤保護（Equal Error Protection，EEP）的方式來保護位元。無論是在主觀或是客觀實驗，發現在低訊雜比（Signal-to-Noise Ratio，SNR）時使用非均勻錯誤保護會比使用均勻錯誤保護還有更佳的效能。除此之外，使用均勻錯保護來保護傳輸位元時，聯合訊源與通道解碼的軟輸出維特比演算法解碼器的效能會勝過傳統使用相同事前機率的軟輸出維特比演算法解碼器。
並列摘要
Source coding and channel coding are definitely different techniques in nature. The former is the method that compresses the original signals into small ones, but still maintains the perceptual quality. On the other hand, the latter adds redundancy for detecting and correcting the transmission error. In this thesis, the mixed excitation linear prediction (MELP) vocoder and the soft-output Viterbi algorithm (SOVA) are used to achieve the purpose of joint source-channel decoding (JSCD). These output statistics of MELP vocoder according to voiced or unvoiced state are regarded as the prior probability of the SOVA decoder. The proposed channel decoding scheme is called the JSCD SOVA. Both unequal error protection (UEP) and equal error protection (EEP) are also included for comparison. The experimental results show that the performance of the UEP is higher than that of the EEP in low signal-to-noise ratio (SNR) condition. In addition, using the EEP to protect transmission bits, the JSCD SOVA decoder performs better than the conventional SOVA decoder with equal prior probabilities under poor SNR environment.