Non-intrusive instrumental prediction of certain perceptual Quality Dimensions of Transmitted Speech

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

Date/Time: 18.11.2013, 14:15-15:00

SPEAKER: Gabriel Mittag

ABSTRACT:

This bachelor thesis deals with quality prediction of transmitted speech. In previous works, it has been discovered and confirmed, that the perceptual quality of transmitted speech is a multi-dimensional value. These dimensions are discontinuity, noisiness, coloration and loudness. Contrary to integral methods, which only determine the total quality, a dimensional prediction offers various advantages. It additionally delivers analytic information which can lead to the cause of the disturbance. Furthermore a dimensional estimator is more robust against changes in the transmission system. Until now such estimators were mainly assessed with parametric or intrusive instrumental models. The goal of this thesis is to find a way to assess these dimensions with a non-intrusive instrumental model, which works without reference signal. The model presented concentrates on the prediction of the quality-relevant dimension noise.

HOST: Köster Friedemann