向量和激發線性預測語音編碼之研究: 後級處理與誤差分析

Abstract

向量和激發線性預測 (VSELP) 為一採用於北美雙軌行動通訊系統之 語音編碼法。我們的研究目標在於此語音編碼法解碼端之改進，其中包括 了對於合成語音的後級處理及此語音編碼法的誤差分析。 在後級處理方面，我們採用了一些前人用於CELP的後級處理法實現在 VSELP上，模擬結果數字我們可以得到一些客觀品質的改善及相當的主觀 品質。此外，我們也實驗了其它對次碼框之線性預測係數做線性內插的方 法，從中也得到一些客觀品質的改善。 在誤差分析方面，我 們首先探討了北美雙軌行動通訊系統的誤差偵測及還原法。隨後分析了每 個VSELP位元對錯誤的敏感度及針對了隨機錯誤及Rayleigh衰減頻道造成 的成串錯誤做模擬以得知此錯誤還原法的效果。 VSELP is a speech coding algorithm adopted in American digital cellular system (IS-54). Our research goal is the improvement of decoding, including postprocessing of synthesized speech and error analysis of VSELP algorithm. In postprocessing, we employed the postprocessing method earlier used in the CELP-type coder on VSELP. From simulation results, there are better objective quality and comparable subjective quality than VSELP in IS-54. Besides, we experimented other interpolation method for linear prediction coefficients in subframes and got some objective quality improvement. In error analysis, we first discussed error protection and error recovery methods in IS-54. Error sensitivities of different VSELP bits were also investigated. Then we employed the esror recovery schemes in IS-54 to mitigate transmission errors. Random errors and bursty errors caused by Rayleigh fading channel were simulated to test the error recovery effect.