2. Adaptive Playout Scheduling Using Time-scale Modification Yi Liang, Nikolaus Färber Bernd Girod, Balaji Prabhakar

3. Yi LiangAdaptive Audio Playout Outline  QoS concerns and tradeoffs  Jitter adaptation as a playout scheduling scheme  Packet scaling using improved time-scale modification technique  Loss concealment in compatible with adaptive playout  Performance comparison and audio demos

4. Yi LiangAdaptive Audio Playout QoS Concerns at the Receiver Delay Jitter DelayPacket Loss Obstructs proper reconstruction of voice packets at the receiver Impairs interactivity of conversations Impairs speech quality Over best-effort network …

5. Yi LiangAdaptive Audio Playout Playout Algorithm (1) - Fixed Deadline  Use buffer to absorb delay variations and playout voice packets at fixed deadline – jitter absorption  Voice packets received after deadline are discarded Sender Receiver Playout time  12345678 12345678 late loss buffering delay time packetization time

6. Yi LiangAdaptive Audio Playout Buffer Delay vs. Late Loss late loss buffering delay Playout Jitter DelayPacket Loss Fixed playout deadline and jitter absorption:  The playout rate is constant  The tradeoff is between buffering delay and late loss

7. Yi LiangAdaptive Audio Playout Playout Algorithms (2) - Adaptive Playout  Monitor delay variation and adapt playout time accordingly - jitter adaptation  Slow down playout when delay increases to avoid loss; speed up playout when delay decreases to reduce delay Sender Receiver Playout time 12345678 12345678 buffering delay time packetization time slow down, speed up

8. Yi LiangAdaptive Audio Playout Adaptive Playout and Jitter Adaptation Adaptive playout and jitter adaptation  Scaling of voiced packets in highly dynamic way  Playout schedule set according to past delays recorded  Improved tradeoff between buffering delay and late loss  Playout rate is not constant Playout Jitter DelayPacket Loss buffering delay

9. Yi LiangAdaptive Audio Playout Packet Scaling (1)  Based on WSOLA [Verhelst 93]  Improved to scale short individual voice packets  “In-and-out black box operation”, no algorithmic delay, smooth transitions  Preserves pitch output packet 1/20/12/33 4 input packet pitch period 02134 template segment

10. Yi LiangAdaptive Audio Playout Packet Scaling (2)  STD network delay= 20.9 ms Max. jitter=112.0 ms STD total delay= 10.5 ms Packets scaled: 18.4 % Scaling ratio: 50% - 200%  DMOS: 4.5 DMOS scaling : degradation is 5 - inaudible 4 - audible but not annoying 3 - slightly annoying 2 - annoying 1 - very annoying

11. Yi LiangAdaptive Audio Playout Loss Concealment  Based on [Stenger 96]  Using information from both sides, delay minimized to one packet time  Integrates nicely into system when adaptive playout is used  20% random packet loss: Original: Loss: Concealed: i-2i-1i+1i+2 time i-2 i+2 time i lost i-1i+1 L LL 2 L 1.3 L alignment found by correlation

12. Yi LiangAdaptive Audio Playout Comparison of Different Algorithms 1.Method which uses fixed playout time throughout the whole session; 2.Method which estimates delay dynamically but only adjusts playout time during silence periods [Ramjee 94, Moon 98]; 3.Method which dynamically estimates and adjusts playout time, and scales packets within talkspurts using time-scale modification.

13. Yi LiangAdaptive Audio Playout Performance Comparison Traces measured between a host at Stanford and hosts in: 1)Chicago 2)Germany 3)MIT 4)China

14. Yi LiangAdaptive Audio Playout Overall Performance Quality Score Excellent 5 Good 4 Fair 3 Poor 2 Bad 1 Buff. delay Loss rate MOS Alg. 255 ms10%2.6 Alg. 34%3.7 Original4.4

15. Yi LiangAdaptive Audio Playout Conclusions  Improved time-scale technique to work on individual packets with minimum delay  WSOLA based loss concealment integrates nicely into system  Adaptive playout and jitter adaptation significantly reduce buffering delay and late loss, which results in improved overall performance  Small playout rate variation can be traded for lower delay and lower loss rate  Playout scaling depends on audio scaling; scaling of individual packets is almost inaudible