Call quality prediction for audiovisual time-varying impairments using simulated conversational structures

摘要

In this study we present an evaluation and improvement of time integration speech quality models applied to assess fluctuating quality of audiovisual transmission. We first introduce a subjective test methodology to evaluate the user perception of time-varying quality of 90 seconds long sequences that are organized in a simulated conversational structure. We conducted a two-fold user test where in the first part, the quality of simulated videotelephony conversations was assessed. Audiovisual impairments were temporally distributed to follow predefined quality profiles. In the second part of the experiment, subjective ratings of short audiovisual samples (9 seconds) constituent of the simulated conversations are gathered. The results of both experiments show that the end-dialog judgments are closely correlated to the plain average of the short samples. The modeling results for call quality models that predict the quality at the end of a (simulated) conversation are described. These models proved to enhance the prediction accuracy in comparison to the plain average, and an optimization of the models' parameters further refines the correlation of the estimates with the subjective data. The optimized models also showed a higher correlation and a lower prediction error on independent test data.