With the advent of speech communication systems transmitting the full audible frequency band (0–20,000 Hz), traditional approaches for narrowband (300–3,400 Hz) speech quality estimation, service planning and monitoring come to their limits. Recently, signal-based as well as parametric tools have been developed for fullband speech quality prediction. These tools estimate overall quality, but do not provide diagnostic information about the technical causes of degradations. In the present paper, we evaluate approaches for diagnostically monitoring the quality of super-wideband and fullband speech communication services. The aim is, first, to estimate technical causes of degradations from the degraded output signals, and, second, to combine the estimated causes with parametric quality prediction models to obtain a quantitative diagnostic picture of the quality-degrading aspects. We evaluate approaches for non-intrusively identifying coding schemes and packet-loss, and compare estimated quality to the predictions of an intrusive signal-based model.
Cite as: Möller, S., Hübschen, T., Michael, T., Mittag, G., Schmidt, G. (2020) Non-Intrusive Diagnostic Monitoring of Fullband Speech Quality. Proc. Interspeech 2020, 2872-2876, doi: 10.21437/Interspeech.2020-1125
@inproceedings{moller20_interspeech,
author={Sebastian Möller and Tobias Hübschen and Thilo Michael and Gabriel Mittag and Gerhard Schmidt},
title={{Non-Intrusive Diagnostic Monitoring of Fullband Speech Quality}},
year=2020,
booktitle={Proc. Interspeech 2020},
pages={2872--2876},
doi={10.21437/Interspeech.2020-1125}
}