MPEG-4 音訊切片算術編碼之效能分析與改進

Abstract

MPEG-4是由ISO/IEC MPEG所制訂的一套很有效率的多媒體壓縮編碼標準。MPEG-4 第二版的音訊壓縮標準提供了一些新的工具來擴充其功能。其中一個工具稱為切片式算數編碼(Bit-Sliced Arithmetic Coding, BSAC)工具，這個工具提供了編碼率精細可調式的音訊編碼功能，每個可調間距大約為1 kbits/s/ch。這功能對一些頻寬容易變動的通訊系統，例如網際網路或行動通訊來說，是非常有用的。 在本篇論文當中，我們首先研究切片式算數編碼的音質效能及其對於傳輸錯誤的敏感度。接著，我們提出兩種方法試圖改善切片式算數編碼的編碼效率。比較切片式算數編碼和進階音訊編碼(Advanced Audio Coding, AAC)的音質效能之後，我們對實驗結果進行分析，並提出造成兩者效能差異的可能原因。因為算數編碼是一種對傳輸錯誤很敏感的編碼方式，所以我們也研究了切片式算數編碼中的錯誤傳遞問題。 在改善編碼效率方面，我們研究了在切片式算數編碼過程中中會用到的機率模型。我們也設計並測試經由實際聲音訊號產生的機率模型。另一個改善編碼效能的方法是改變每個可調層分配到的位元數。主要觀念在於分配更多的位元數給較低頻的可調層。這個方法將可以看到比較明顯的效能改善。

The MPEG-4 standard defined by ISO/IEC MPEG is a very efficient coding standard for multimedia data. MPEG-4 version 2 provides several new tools for audio coding. One of them is the so-called BSAC (Bit-Sliced Arithmetic Coding) tool, which provides scalable coding with fine granularity. The scalability step is about 1 kbits/s/ch, which would be useful for communication systems with fluctuating bandwidth, such as Internet and mobile communication. In this thesis, we first investigate the quality and transmission error sensitivity of BSAC. Then, we propose two modifications in attempt to improve the coding efficiency. The coded audio quality of BSAC is compared with that of AAC (Advanced Audio Coding). And then we analyze the experimental results and identify the cause of its performance loss. It is well-known that arithmetic coding is sensitive to transmission errors. We thus also investigate the error-propagation problem of BSAC. To improve the coding efficiency, we study the probability models used in the arithmetic coding process of BSAC. Probability models trained from real audio data have been designed and tested although they do not offer significant improvement. Another attempt is changing the bits allocated to each layer. The idea is that the lower frequency bands should receive more bits in coding. The coding gain using this strategy turns out to be much more significant.