正弦音長調整在網路語音封包播放時序之應用

Abstract

由於網際網路語音通訊的即時性，接收端會將語音封包暫存在一緩衝器中並延遲其播放時間，以補償網路延遲顫動的影響。針對三種適應性播放時序演算法，包含一個以話務為基礎及兩個以封包為基礎的演算法，在平均緩衝延遲和晚到漏失率之間做取捨衡量。對於以封包為基礎的延遲調整來說，配合使用一種基於諧波正弦表示的語音分析合成機制，對個別的語音封包做適當的音長重建以達到連續播放語音的效果。每個正弦波組成的貢獻包括描述振福、激發相位和聲道系統的參數，控制這些參數以完成音長比例調整。我們的網路模擬是基於單一伺服器的排隊模型，一連串定期的語音封包受網際網路串流的影響，則以一批伯努利串流近似。模擬結果顯示配合正弦音長調整，NLMS時序演算法能使緩衝延遲和晚到漏失率之間有最好的取捨衡量，同時明顯改善其播放語音的品質。

For real-time voice communication over internet networks, voice packets are buffered at a receiver and their playout delayed in order to compensate for the network delay jitter. Three adaptive playout scheduling algorithms, one per-talkspurt based and two per-packet based, are examined for a trade-off between average buffering delay and late loss rate. For per-packet based delay adjustment, proper reconstruction of continuous playout speech is achieved by time-scaling individual voice packets using a new technique based on a sinusoidal representation of the speech production mechanism. The proposed time-scale modification involves the manipulation of functions which describes the amplitude and phase of the excitation and vocal tract system contributions to each sine-wave component. Our network simulation is based on a single-server queueing model in which the impact of the internet traffic on a periodic stream of audio packets is approximated by a batch Bernoulli traffic. Simulation results indicate that with the aid of sinusoidal time-scale modification, NLMS-based scheduling algorithm improves the playout speech intelligibility with the best trade-off between buffering delay and late loss rate.