Digital Signal Processing and System Theory

Talk Befkadu Temesgen Gebru

Automated Quality Assessment of Text-to-Speech Signals Using Acoustic Markers (in English)

Date: 10.05.2012, 13:00 h - 14:00 h, Room: Aquarium

Befkadu Temesgen Gebru
CAU, Kiel, Germany

Details (talk will be in English)

In today's competitive environment, perceived speech quality of Text-to-Speech (TTS) signals is the main criterion for the customers to judge the Quality of Service (QoS). In this thesis work different speech signal analysis methods are tested in an effort to develop objective TTS-signal quality estimators. The effect of signal degradation on short-term and long-term temporal dynamics of the TTS signals are captured to derive quality measures. Furthermore, irregularities in the periodicity of the TTS signals is analyzed with general variogram analysis. Finally, cross-validated feature selection using the data from the quality measures obtained is developed in an effort to build non-intrusive TTS-signal quality estimators.