Real-Time, Non-Intrusive Evaluation of VoIP Using Genetic Programming
1 like207 views
This document describes research into using genetic programming to develop models for real-time, non-intrusive evaluation of VoIP quality. The researchers used network traffic parameters and a dataset of 3360 distorted speech files to train and test models. The best models depended on only 1-3 variables and more accurately estimated speech quality compared to existing standards like PESQ, performing better than prior approaches using neural networks, regression, or lookup tables. The genetic programming approach produced counter-intuitive but high-performing models suitable for real-time VoIP quality monitoring without intrusive measurement.
Real-Time, Non-Intrusive Evaluation of VoIP Using Genetic Programming
- 1. REAL-TIME, NON-INTRUSIVE EVALUATION OF VOIP USING GENETIC PROGRAMMING MUHAMMAD ADIL RAJA1 ATIF AZAD2 COLIN FLANAGAN1 CONOR RYAN2 1 WIRELESS ACCESS RESEARCH CENTRE DEPARTMENT OF ELECTRONIC AND COMPUTER ENGINEERING 2 BIO-COMPUTING AND DEVELOPMENTAL SYSTEMS DEPARTMENT OF COMPUTER SCIENCE AND INFORMATION SYSMTEMS UNIVERSITY OF LIMERICK, LIMERICK, IRELAND RESEARCH GOAL Derivation of a VoIP listening Quality estimation model as a function of transport layer metrics. Genetic Programming based Symbolic Regression is used Using the ITU-T Recommendation P.862, PESQ , as the reference system PESQ is the de-facto standard for speech quality estimation. ON DATA COLLECTION The VoIP traffic parameters include loss rate, burstiness measure, frame duration, packetization interval and bit-rate. A total of 3360 distorted speech files were created for each combination of network traffic parameters. 1177 35% were used for training 503 15% were used for testing 1680 50% were used for speaker independent validation SIGNIFICANCE OF VOIP PARAMETERS FIGURE : VoIP system VOIP QUALITY MONITORING MODELS MOS − LQOGP = −2.46 × log(cos(log(br)) + mlrVAD ×(br + fd/10)) + 3.17 (1) MOS − LQOGP = −2.99 × cos(0.91 × sin(mlrVAD) +mlrVAD + 8) + 4.20 (2) Equation (1) Equation(2) Data MSEs σ MSEs σ Training 0.0370 0.9634 0.0520 0.9481 Testing 0.0387 0.9646 0.0541 0.9501 Validation 0.0382 0.9688 0.0541 0.9531 SCATTER PLOTS SCATTER PLOTS On Performance of ITU-T P.563 FIGURE : VoIP system FIGURE : VoIP system SIGNIFICANCE OF MODELS 1 The model is a good approximation to PESQ. 2 Suitable for real-time and non-intrusive estimation of speech quality whereas PESQ is NOT. 3 Simple models; depend on 3 and 1 variable respectively. 4 Performs significantly better than ITU-T P.563 THE HUMAN COMPETITIVE ELEMENT 1 The results are better than past approaches that include: Various flavors of Artificial Neural Networks based models. Non-linear least squares estimation. Multi-state Markov Chains. Look-up tables 2 All the above approaches (excluding ANNs) used a restricted set of network traffic parameters. 3 By far it is the first ever application of GP on the problem of speech quality estimation, and has produced better results. 4 The results are counter-intuitive. 5 The results solve a formidable problem and outperform previously existing solutions.