direkt zum Inhalt springen

direkt zum Hauptnavigationsmenü

Sie sind hier

TU Berlin

Inhalt des Dokuments

Perceived Quality of Telephone Speech Featuring Background Noises

LOCATION: Auditorium 1, TEL, Ernst-Reuter-Platz 7, 20th floor

SPEAKER: Raphael Ullmann (Idiap Research Institute, Martigny, Switzerland)


Mobile telephony often takes place in environments with uncontrollable noise levels. Noise reduction techniques process the audio signal transmitted by the phone with the aim of lowering the level of background noise without affecting the user's speech. In practice, such perfect separation is not possible and the right balance between sufficient noise attenuation and tolerable speech distortion must be found. Noise reduction has received renewed interest following the recent deployment of so-called wideband speech, which doubles the transmitted speech (and hence, noise) bandwidth in telephone networks.

In this talk, I will present some results from listening tests that evaluated the perception of noisy speech in both traditional and wideband speech systems.
We used the subjective test method recommended by the International Telecommunications Union (ITU-T P.835) to separately evaluate the perceived noise intrusiveness, speech distortion and overall listening quality of noisy speech signals. Results indicate sub-optimal noise reduction performance at moderate noise levels, and that the used test method allows listeners to evaluate noise intrusiveness independently of speech type and content. A future goal is the development of an algorithm that predicts the perception of noisy speech directly from the signal, yielding an objective quality measure for noise reduction systems.


Raphael Ullmann received the M.Sc. in Micro-engineering from the Swiss Federal Institute of Technology Lausanne (EPFL), Switzerland, in 2006. His Master's thesis dealt with psychoacoustic modeling for Advanced Audio Coding. From 2006 to 2012, he was an Applied Research Engineer in the industry, where he developed algorithms for the analysis of speech signals. He has been one of the main contributors to the ITU-T P.863 algorithm for perceptual speech quality prediction. Since 2012, he is pursuing a Ph.D. at the Idiap Research Institute in Martigny, Switzerland, under the EPFL doctoral program in electrical engineering. His research is focused on the perception and information content of background signals in speech.

HOST: Sebastian Möller

Zusatzinformationen / Extras


Schnellnavigation zur Seite über Nummerneingabe