串流平台語音封包遺失的處理

Abstract

由於互連網路的發達，在網路上即時傳送影像及聲音已經成為了一種趨勢。 串流技術(streaming)就是這種一邊下載一邊播放的技術。網路傳輸訊號時難免會有遺失或是錯誤。對「先下載完再播放」的傳統技術而言，遺失握錯誤的封包，只需重新傳輸即可。但是串流技術為了因應即時播放的邀求，沒有時間重新傳輸遺失或錯誤的封包。在聲音方面，過去有人提出用重複上一個收到的片段來填補遺失或錯誤片段的方法。後來有人提出樣本比對(pattern matching)來改進它。本文作者利用mp3或aac聲音壓縮分頻段的特性，做多頻段樣本比對以填補遺失或錯誤的片段，並將重建的信號與原始訊號比對。結果顯示，多頻段樣本比對在低雜訊的情形下比不上舊有的樣本比對，但是在較高color noise時有較佳的效果。

It has been a tending to real-time transmit audio-video signals on Internet. Streaming is the technology of playing while downloading. The packet loss and crack are unavoidable in Internet transmission. For the conventional technology of play-after-download, the lost or cracked packets can simply retransmitted. However, there is no time to retransmit them in streaming. In the audio area, it was proposed to replace them by the last received frame. It was then proposed the method of pattern matching to improve. In this thesis, it was proposed further the multi-band pattern matching to make use the multi-band compression of the mp3 and aac audio. The reconstructed signals was then compared wth the original signals. It was found that the results of multi-band patt\ern matching are worse the those of the “old” pattern matching when the noise levels are low, themulti-band pattern matching has better when the color noise are higher.