An innovative new approach to measuring and assuring speech quality, based on the unique application of machine-learning techniques.
There's no doubt that voice is a critical service for MNOs. While greater attention is often paid to soaring video consumption, it remains a fact that 70% of MNO revenue continues to come from voice.
The introduction of VoLTE and, more recently, of the high-definition EVS voice codec brought new complexities to the delivery of voice. With 5G rapidly approaching, there will be more challenges to confront, as the introduction of new services and systems will add further complications. Mobile subscribers expect voice services to perform as promised. As a result, MNOs need to continually evaluate how they can maintain and assure voice quality and performance — and protect their revenues in the face of increasingly agile competitors.
Of course, the old adage remains true: you can only manage what you can measure, but today's environment calls for a new approach: one that is less expensive; more easily adaptable to new network conditions, such as new codecs/clients; that is independent of device type and firmware; and which offers even greater accuracy.
Infovista has long been a pioneer in the delivery of carrier-grade assurance solutions and has invested significantly in innovative research to address the new challenges brought by network evolution. These efforts have led to the development of an entirely new model for measuring voice quality: sQLEAR.
sQLEAR, or ‘Speech Quality by machine LEARning', is a hybrid (or intrusive parametric) approach that combines analysis of RTP/I parameters and codec/client information with speech reference samples to calculate an estimate of speech quality as perceived by users. Because of this approach, sQLEAR is able to predict network and codec/client-centric speech quality independently of the acoustic characteristics of devices, and without the need to tune and calibrate each device.
sQLEAR makes novel and innovative use of state-of-the-art machine-learning capabilities to enable rapid adaptation to new network conditions and codecs/clients, as well as new kinds of voice services. This dramatically reduces the time — and cost - it has previously taken to ‘train' speech quality algorithms. As a result, sQLEAR offers MNOs a new model for the continuous evaluation of speech quality, even while they upgrade their network and service portfolio and evolve towards 5G. It ensures that they can deliver consistent quality, while reducing the costs associated with assuring voice services.
In addition to having been developed for the latest voice service technologies that will be adopted in 5G networks, sQLEAR also supports VoLTE services that use the EVS codec (including full bandwidth 24.4KHz), with proven high accuracy.
sQLEAR is based on ongoing ITU-T activity, specifically the P.VSQMTF work item, “Voice service quality monitoring and troubleshooting framework for intrusive parametric voice QoE prediction”. This has the goal of creating a new standard that will assist MNOs through the next phase of network and service evolution, while reducing their costs.
To learn more about this new innovation, please get in touch with the Infovista team.
