TU Berlin

Quality and Usability Lab2016_07_04_Malkova

Inhalt des Dokuments

zur Navigation

Comparison of different feature selection algorithms for speech quality estimation

LOCATION: Auditorium 2, TEL, Ernst-Reuter-Platz 7, 20th floor

Date/Time: 04.07.2016, 14:15-15:00

SPEAKER: Ekaterina Malkova (TU Berlin) 

ABSTRACT:

Traditionally, speech quality is evaluated using subjective tests. Due to the limitations of this method (costs and time consumption), objective speech quality assessment methods have become a very active research area. These methods describe quality on several dimensions such as discontinuity, noisiness, coloration, and loudness.

In this talk, I give an account on how the number of features for two speech quality dimensions can be reduced using machine learning algorithms. From the reduced data an optimal feature set is formed for each dimension. The feature sets are further used to compare different mapping algorithms that predict the quality value for noisiness and coloration.

Navigation

Direktzugang

Schnellnavigation zur Seite über Nummerneingabe