direkt zum Inhalt springen

direkt zum Hauptnavigationsmenü

Sie sind hier

TU Berlin

Page Content

Attribute-based Speech Quality Measures (DFG)

Lupe

Motivation & Project Description:
A sufficiently high communication quality of a speech service is a necessary condition for its acceptance. In modern telecommunication networks, however, a variety of factors may disturb the quality of the transmitted speech. Examples are linear distortions, (low-bitrate) coders, delays, echoes, packet loss of VoIP links, or the signal processing of various end-user terminals. This DFG-funded research project aims at the development of a new, signal-based approach for predicting the speech quality of modern telecommunication networks. The basic idea is the assumption that the perceived speech quality can be seen as a composition of different, quality-relevant attributes. In order to measure the speech quality or the underlying perceptual components (attributes), subjective tests, i.e. experiments with human listeners are a-priori required. Hence, prediction models can be derived with which the speech quality and their perceptual attributes can be instrumentally estimated by correlating physical parameters or speech-signal features. This instrumental capture allows to predict the speech quality perceived by the user. In this project, especially those network components are considered for which known instrumental quality estimators provide only insufficient results, e.g., for packet-based transmission systems that provoke temporal instationary distortions. Modern signal processing approaches (e.g., coding or noise reduction) are not designed to cope with a signal bandwidth beyond the "classical" telephone band of 3.5 kHz. The transition to packet-based networks of all big telephone companies and the universal data transmission that becomes possible justify the assumption that speech transmission with nearly doubled audio bandwidth (e.g., 50 - 7000 Hz) is on the verge of a breakthrough. Such effects will be taken into account in the research project as well in order to gain new knowledge in the perception of wideband speech transmission quality. Expected Outcome:

  • signal-based approach for the prediction of speech quality on the basis of instrumental estimation of perceptual dimensions
  • extension of existing network planning tools (e.g., the so-called E-model) towards new distortions
Time Frame:

04/2005 - 03/2008
T-labs Team Members:

Sebastian Möller, Alexander Raake, Marcel Wältermann
Students:
Ulrike Stiefelhagen
Partners:
Institute of Circuit and System Theory, Christian-Albrechts-University, Kiel
Funding by:
Deutsche Forschungsgemeinschaft (DFG)MO 1038/2-1 und MO 1038/5-2
Publications:
see list of publications of: Möller/Raake/Wältermann

Zusatzinformationen / Extras

Quick Access:

Schnellnavigation zur Seite über Nummerneingabe

Auxiliary Functions