A perceptual quality model intended for adaptive VoIP applications

Abstract

SUMMARYQuality models predict the perceptual quality of services as they calculate subjective ratings from measured parameters. In this article, we present a new quality model that evaluates Voice over IP (VoIP) telephone calls. In addition to packet loss rate, coding mode and delay, it takes into account the impairments due to changes in the transmission configuration (e.g. switching the coding mode or re-scheduling the playout time). Moreover, this model can be used at run time to control the transmission of such calls. It is also computationally efficient and open source.To demonstrate the potential of our model, we apply it to select the ideal coding and packet rate in bandwidth-limited environments. Furthermore, we decide, based on model predictions, whether to delay the playout of speech frames after delay spikes. Delay spikes often occur after congestion and cause packets to arrive too late. We show a considerable improvement in perceptual speech quality if our model is applied to control VoIP transmissions.

Real-Time, Non-intrusive Evaluation of VoIP

“…al. [9] present a look-up table based VoIP quality estimation model. The various MOS and corresponding parameter values would be stored in a lookup table.…”

An Analysis of the Impact of Playout Delay Adjustments introduced by VoIP Jitter Buffers on Listening Speech Quality

“…Hoene et al in [5] used PESQ to investigate the impact of asingle adjustment (0-1000msec)i nan8second sample (typical of delay spikes)d uring both silences and active speech on speech quality.Heshowed that PESQ predicts significant impacts during active speech butthat adjustments of up to approx. 320 ms are not noticeable during silences.…”

“…Examples of the latter approach are described in [3,4] and such techniques can be found in popular Vo IP applications such as GoogleTalk and Skype. Other research has attempted to optimise buffer size and in particular,the trade off between late packet loss and delay based on customised objective models [5,6,7,8] Finally,previous research by one of the authors has proposed ahybrid playout strategy that utilises synchronised time in order to implement an informed fixed delay playout wheneverpossible thus minimising the need for playout adjustments whilst minimising late packet loss impairments. It reverts to an adaptive approach when delays become excessive.D etails of this approach can be found in [9].…”

“…Hoene et al also validated the use of the PESQ model to predict the impact of adjustments during active speech. In subsequent research by Hoene et al [5], PESQ wasused to estimate the impact of single large adjustments during both silences and active speech, and regarding the former,s hows howa djustments of up to approximately 320 ms are deemed not noticeable. Finally,research using subjective listening tests by Vo ran [11], suggests that very large adjustments (430 ms)are noticeable and then examines the impact of various general impairments, butn ot specifically silence period adjustments.…”

“…• Comprehensive study to build on Hoene's et al work in [5] and investigate the impact of such silence-period adjustments (i.e. typical of Vo IP jitter buffers)o no bjective listening MOS scores (MOS-LQO), specifically predicted by PESQ.…”