TU Berlin

Quality and Usability Lab2018_01_22_Mishra

Inhalt des Dokuments

zur Navigation

Assessing the Effects of Speech Distortions on the Automatic Prediction of Speaker Personality

LOCATION:  TEL, Auditorium 3 (20th floor), Ernst-Reuter-Platz 7, 10587 Berlin

Date/Time
: 22.01.2018, 14:15-15:00

SPEAKER: Soumya S. Mishra (TU Berlin)

ABSTRACT:

With the abundance of telephone-based voice services in the market and the ever-increasing need of providing tailored solutions based on users' characteristics and preferences, speaker personality research has taken a forward leap. Due to the transmission of audio signals via telephone channels, noise in the form of audio distortions gets added to the speech. The possible effects of distortions on the automatic personality prediction have been overlooked in the literature. In this paper, we experiment with different speech databases labeled with externally-observed or self-reported Big Five personality traits, to which we applied different channel degradations. Our personality classification results show that channel distortions affect the observed personality prediction more than the self-reported personality prediction. This effect varies for the different personality traits. Extraversion is affected the most whereas Openness is the least affected by the distortions, suggesting that Extraversion has a greater influence on energy and spectral speech features which are affected by telephone transmissions.

Navigation

Direktzugang

Schnellnavigation zur Seite über Nummerneingabe