First ISCA ITRW on Auditory Quality of Systems

April 23-25, 2003
Akademie Mont-Cenis, Germany

Corrupted Speech Data Considered Useful

Florian Hammer (1), Peter Reichl (1), Tomas Nordström (1), Gernot Kubin (2)

(1) Telecommunications Research Center Vienna (ftw.), Austria
(2) Signal Processing and Speech Communication Lab (SPSC), University of Technology Graz, Austria

The success of Voice-over-IP technology will crucially depend on the provision of a certain level of perceptual speech quality. In this paper we demonstrate that keeping as much corrupted speech data as possible results in a major step towards this goal. To this end, we explore the combination of two key factors which influence the speech quality as perceived by the user: networking mechanisms and signal processing algorithms. We present three strategies of dealing with voice packets that have been damaged by bit errors. The performance of these strategies is assessed using the standardized PESQ-algorithm. Results of extensive simulations give a detailed insight into the behavior of the different strategies, and thus prove that corrupted speech data in fact have to be considered useful.

Full Paper

Bibliographic reference.  Hammer, Florian / Reichl, Peter / Nordström, Tomas / Kubin, Gernot (2003): "Corrupted speech data considered useful", In AQS-2003, 51-54.