A Generic Algorithm for Midcall Audio Codec Switching Haytham Assem, Mohamed Adel, Brendan Jennings, David Malone, Jonathan Dunne and Pat O Sullivan TSSG, Waterford Institute of Technology,Ireland Hamilton Institute, National University of Ireland Maynooth, Ireland IBM Software Lab, Dublin, Ireland
Outline Voice-over-IP Measuring Quality-of-Experience of VoIP Impact of Codec Switching Codec Switching Algorithm Experimental Analysis Conclusion & Future Work Copyright 2013 Waterford Institute of Technology 2
Outline Voice-over-IP Measuring Quality-of-Experience of VoIP Impact of Codec Switching Codec Switching Algorithm Experimental Analysis Conclusion & Future Work Copyright 2013 Waterford Institute of Technology 3
Voice-over-IP Rapid growth of IP networks provided unusual means of communications.voip has evolved rapidly, providing real-time voice and video communications between users through the existing IP network in a cheap cost. VoIP has gained wide acceptance by general Internet users and is increasingly important in the enterprise communications sector. It has become essential to VoIP providers to have testing frameworks to assess users' QoE to provide the best possible service. Copyright 2013 Waterford Institute of Technology 4
Outline Voice-over-IP Measuring Quality-of-Experience of VoIP Impact of Codec Switching Codec Switching Algorithm Experimental Analysis Conclusion & Future Work Copyright 2013 Waterford Institute of Technology 5
Quality-of-Experience of VoIP Achieving voice quality levels for VoIP remains a significant challenge, as IP networks typically do not guarantee delay, packet loss, jitter and bandwidth levels. Packet loss in the IP network is considered one of the most important factors that cause degradation in the overall voice call quality. The maximum quality that can be achieved differs from codec to another under different packet loss rates. Copyright 2013 Waterford Institute of Technology 6
Measuring Audio Call Quality Subjective testing as specified in ITU-T, involves 12-24 participants individually listening to an audio stream of several seconds and rate the audio quality on the scale of 1 (Poor) to 5 (Excellent), with these ratings being used to form a single Mean Opinion Score (MOS). Objective testing provides an alternative for measuring voice quality. It is classified into: Intrusive (e.g. PESQ) Non-intrusive methods (e.g. E-model). Copyright 2013 Waterford Institute of Technology 7
Deriving E-Model for non-itu codecs E-Model does not provide codec impairment values for widely used non-itu codecs like SILK, ILBC, SPEEX, and GSM. We used PESQ to estimate MOS for the above codecs under different packet loss rates ranging from 0-20 %. Non-linear regression model, with parameters a, b, and c is driven for each codec. I e-eff represents the codec impairment value. I e-eff = a log(1+b x Ppl) + c Copyright 2013 Waterford Institute of Technology 8
Measuring Audio Call Quality Copyright 2013 Waterford Institute of Technology 9
Outline Voice-over-IP Measuring Quality-of-Experience of VoIP Impact of Codec Switching Codec Switching Algorithm Experimental Analysis Conclusion & Future Work Copyright 2013 Waterford Institute of Technology 10
Codecs' Performance Copyright 2013 Waterford Institute of Technology 11
Codecs' Performance Possible switching points 23/5/2013 Copyright 2013 Waterford Institute of Technology 12
Switch-over Gaps The switching of codecs during the communication causes a switchover gap. It the time taken between sending the RE-INVITE message from the sender side and receiving the ACK at the receiver side indicating the start of transmission with the new codec. Switch-over gap indicates the response time to switch to another codec. At high packet loss rates, the RE-INVITE message will be at a higher probability of being lost, which will cause multiple retransmissions, and the same also will happen for the 200OK and ACK messages, therefore the switch-over gap will increase more. Copyright 2013 Waterford Institute of Technology 13
Switch-over Gaps Copyright 2013 Waterford Institute of Technology 14
Switch-over Gaps We focus our algorithm on switching codecs in that region (0-10% packet loss) in order to minimize the switch-over gap( avg of 0.5 sec) in order to increase the responsiveness of our algorithm. Copyright 2013 Waterford Institute of Technology 15
Frequent Codec Switching Frequent switching of codecs during a session could cause degradation in the overall call quality. In order to measure the only degradation in the call quality as a result of increasing the number of switches, we selected pairs of codecs which have nearly the same or almost same performance. Copyright 2013 Waterford Institute of Technology 16
Frequent Codec Switching Copyright 2013 Waterford Institute of Technology 17
Outline Voice-over-IP Measuring Quality-of-Experience of VoIP Impact of Codec Switching Codec Switching Algorithm Experimental Analysis Conclusion & Future Work Copyright 2013 Waterford Institute of Technology 18
Codec Switching Algorithm Copyright 2013 Waterford Institute of Technology 19
Outline Voice-over-IP Measuring Quality-of-Experience of VoIP Impact of Codec Switching Codec Switching Algorithm Experimental Analysis Conclusion & Future Work Copyright 2013 Waterford Institute of Technology 20
Experimental Analysis We evaluate Codec Switching Algorithm with packages of codecs with one default codec to start the call with. For our experiments we played a sample audio file for 3mins, with the potential for switching a codec being assessed every 10s. Copyright 2013 Waterford Institute of Technology 21
Experimental Analysis 1% packet loss 5% packet loss Copyright 2013 Waterford Institute of Technology 22 22/28
Outline Voice-over-IP Measuring Quality-of-Experience of VoIP Impact of Codec Switching Codec Switching Algorithm Experimental Analysis Conclusion & Future Work Copyright 2013 Waterford Institute of Technology 23
Conclusion & Future Work Switching codecs during an ongoing voice session can improve user s perceived quality-of-experience due to the fact that different codecs behave differently under different packet loss conditions in the network. Codec Switching has a slightly negative impact which has to be taken into account when designing codec switching algorithm. We intend to extend our algorithm to support wideband audio codecs by applying the newly developed POLQA objective testing method. We are planning to improve our algorithm by studying loss patterns to assess if the frequency and distribution of losses affect codecs quality differently. Copyright 2013 Waterford Institute of Technology 24
Thank you Questions? Questions? Copyright 2013 Waterford Institute of Technology 25