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

Date/Time: 16.06.2014, 14:15-15:00

SPEAKER: Anna Kuzmina

ABSTRACT:

It is assumed that the quality of transmitted speech can be modeled on the basis of perceptual dimensions. In previous works the dimensions "coloration", "discontinuity", "noisiness" and "loudness" have been identified. The assessment of these dimensions offers the advantages of diagnostic information and robustness. However, until now this assessment is mostly done with parametric or intrusive instrumental models.  In the present thesis a prediction of the two dimensions "coloration" and "discontinuity" is investigated on the basis of only the output signal without a reference, also known as non-intrusive prediction.

HOST: Friedemann Köster