Infovista brings machine learning capabilities to QoE measurement in ITU-T SG-12

Experts from Infovista lead ITU-T efforts towards a new voice service quality monitoring and troubleshooting framework for intrusive parametric voice QoE prediction. 

 The ITU-T is one of the world’s leading standards organizations. It coordinates the development of international standards for ICT, driving interoperability and effective global communications. Its output is delivered through dedicated Study Groups and, as a leading innovator in communications network performance and application control, Infovista’s experts make numerous contributions to this valuable work.  

20 years ago, the ITU-T began to address the need to automate the prediction of speech quality through the analysis of speech signals with the publication of the P.861 standard, otherwise known as PSQM or Perceptual Speech Quality Measure. PSQM was launched in 1997 and, since then, standardization efforts have continued to evolve. PESQ (Perceptual Evaluation of Speech Quality), or P.862 followed in 2001 while, in 2011, POLQA (Perceptual Objective Listening Quality Analysis) or P.863 emerged.  

Since then, POLQA has been the de-facto standard for speech quality measurement and has been widely adopted by network equipment manufacturers (NEMs) and network operators. By providing accurate measurements of speech quality, it’s helped operators deliver a better experience to their subscribers and has made a significant contribution to the evolution of networks. 

However, although POLQA has continued to evolve, most recently through enhancements to help address the challenge raised by VoLTE, there are some drawbacks. First, there are costs, with ongoing royalties for which companies using POLQA are liable. Second, there is also the need to calibrate field solutions in which POLQA has been implemented. Finally –  and, crucially – POLQA is dependent on devices, because it has sensitivities that vary according to the audio path frequency characteristics of different handsets. So, as devices have continued to proliferate, it has become more and more difficult for operators and NEMS to keep pace.  

As a result, it’s become increasingly clear that, while POLQA remains of critical importance, additional techniques are required to keep pace with continuing device innovation and the sheer volume of users. That’s why QoE test solutions that can describe the impact of voice quality on the network independently of the device have assumed greater significance for the cost-effective optimization of increasingly complex networks. 

Infovista has been an active leader in predictive speech quality standardization efforts from the outset, making regular contributions to the ongoing activities in the ITU-T for voice and video QoE measurements, as part of the dedicated Study Group 12 (SG-12) efforts. Based on this long experience, Infovista has now made a significant new contribution to the efforts of SG-12, through the launch of sQLEAR, which aims to accelerate speech quality analysis.  

sQLEAR, which stands for Speech Quality by machine LEARning, represents the first outcome from continuing activities in the P.VSQMTF (Voice Service Quality Monitoring and Troubleshooting Framework) work item in ITU-T SG-12, and brings several key innovations to the problem of monitoring and enhancing voice service quality. 

Perhaps the most important of these is the incorporation of machine learning. Until now, machine learning hasn’t been used for the analysis of QoE metrics, such as speech quality, but it offers the promise of dramatically accelerating their analysis and the ability to scale more effectively.  

As such sQLEAR is complementary to POLQA. It allows NEMs and operators to more rapidly determine if problems exist and where they may be found. This pinpointing activity, in turn, paves the way for the intervention of POLQA-based solutions. sQLEAR dramatically increases the efficiency of speech quality evaluation, while providing an accurate voice QoE predictor. It helps operators to deliver the performance they need to retain competitive differentiation and to satisfy the needs of their customers as they seek to monetize their investments in VoLTE, for example.  

In addition, in common with other machine learning-based techniques, sQLEAR relies on training. It can be trained with reference speech samples and with new codecs as they are introduced. This means that it can easily and quickly be adapted for new languages and for ongoing evolution of network standards. Further, sQLEAR is independent of devices, so it does not need to be updated for each new handset that enters the network. So, network operators can adopt sQLEAR across their portfolio, in multiple countries, and benefit from a common QoE evaluation solution. 

sQLEAR natively supports the EVS codec which is deployed in VoLTE, but which is also key to 5G. As such, sQLEAR also provides a future-proof path for the evolution of today’s networks towards a common fixed and mobile core. The innovative use of machine learning means that it can continue to evolve and to meet key use cases, such as video, as well as future services that have other QoE metrics. 

Infovista will continue to work in SG-12 to ensure the standardization of the P.VSQMTF framework on which sQLEAR is based and to ensure that it is available to the community. Our joint efforts will ensure that the real-time measurement of QoE metrics continues to evolve to meet the demands of current and future services, helping to deliver better performance and better quality.  

If you’d like to know more about sQLEAR and how it complements techniques, such as POLQA, why not read our new white paper? Download it here.