Inhalt des Dokuments
zur Navigation
Other Conference Papers
go back to overview
| Zitatschlüssel | soloducha2013b |
|---|---|
| Autor | Soloducha, Michal and Skowronek, Janto and Raake, Alexander |
| Buchtitel | Proceedings of AIA-DAGA 2013 |
| Seiten | 463–466 |
| Jahr | 2013 |
| ISBN | 978-3-939296-05-8 |
| Ort | Meran |
| Adresse | Merano, Italy |
| Monat | mar |
| Notiz | other medium: CD/DVD |
| Wie herausgegeben | full |
| Organisation | Deutsche Gesellschaft fur Akustik |
| Zusammenfassung | Two full-reference speech quality assessment models are currently in use for evaluating telephony systems, for example during intrusive service monitoring: PESQ (ITU-T Rec. P.862) and P.OLQA (ITU-T Rec. P.863). Recent studies have shown that, depending on which speech samples are used for testing a system, different results may be obtained. The aim of this paper is to systematically investigate how quality predictions by these models vary for different speech samples of different languages under different real-life network conditions. Several measurements have been conducted using a test network using different VoIP clients. Utterances in 9 different languages such as English, German, Italian and Japanese from ITU-T Rec. P.501 were used for this purpose. The results show that there is a measurable language dependency of the quality scores. The paper analyses the range of prediction-variation resulting from the use of different samples, in comparison with the variation resulting from different network settings. Based on this comparison, implications for network operators and system manufacturers are being discussed. |
go back to overview