資料隱藏應用於零樹小波分頻壓縮系統之

音訊封包遺失回復技術

 

摘要

 

隨著網路的快速發展,即時性多媒體的應用也愈加廣泛。對於網際網路多媒體即時資料的應用,封包的失是一個嚴重的問題因此需要特別地對即時性多媒體訊號封包作適當的保護處理,降低封包遺失對播放品質的影響。本論文發展資料隱藏應用於零樹小波分頻壓縮系統之音訊封包遺失回復技術,使得重建樂音在不同的封包遺失情形下,皆能有最佳之樂音品質。

本論文以Embedded Zero-tree Wavelet Packet (EZWP) 作為樂音壓縮系統,其以小波封包(Wavelet Packet)分頻方式,結合人耳聲學模型與零樹編碼。利用每個音框開始的前幾次掃描係數當作重要部分,最後一次掃描的更精細值當作不重要部分。基本的想法是將音框資料的重要部分,隱藏於其他音框不重要的部分,並發展區塊資料隱藏法,以降低資料隱藏對於壓縮音質的影響。當封包發生遺失時,可以在解碼端萃取隱藏資料,回復遺失封包的樂音波形。利用EZW 漸進式編碼的特性,非常適合用於資料隱藏,應用於遺失封包恢復技術並不會增加傳輸位元率,並相容位元流語法,且不會嚴重降低原始之音訊封包品質。實驗結果顯示在封包遺失率達到 8% 的情況下,平均有效提升約1.2dB SSNR; 客觀的聽覺測試的結果,亦可證明論文提出的方法能有效改善封包遺失所造成的影響。

 

 

 

Packet Loss Recovery Using Data Hiding for Embedded Zero Tree Wavelet Packet Audio Coding System

 

Abstract

 

Multimedia transmission over Internet is getting popular and increasingly important. Because some networks do not provide quality of service (QoS), the received packet may get lost. As a result, audio quality is often degraded due to packet loss. In this thesis, we propose a packet loss recovery technique using data hiding method applied to Embedded Zero-tree Wavelet Packet (EZWP) audio coding system. The EZWP audio coding system uses wavelet packet decomposition and embedded zero-tree coding based on the psychoacoustic model. We choose several beginning scans of EZW coefficients in each frame as the most significant data, and choose the value list in the last scan as the least significant data. The basic idea is to hide the most significant data into the least significant data. In order to reduce the damage to audio quality, we develop block data hiding method. When packets are lost, we can extract the hided data to reconstruct the most important part of lost audio signals. The hided bitstreams are compatible to the encoding syntax. The data hiding method will not increase bitrate but only slightly degrade the reconstructed audio quality. In simulations, the proposed recovery strategy is able to improve SSNR by 1.2 dB on average. The subjective evaluation tests show that the proposed recovery provides smooth audio signals even at a loss rate of 8%.