Can Emotions be Synthesized without Controlling Voice Quality?

Marc Schröder

In: PHONUS PHONUS. 4 Seiten 37-55 1999.


The present study addresses the question whether it is in principle feasible to convey emotion in synthesized speech using a restricted parameter set which can usually be controlled in concatenation based synthesizers. Using copy synthesis, the prosodic structure of one sentence uttered with five emotional expressions (anger, joy, fear, sadness, and neutral) was transferred to synthetic stimuli. Perception tests show that for some synthetic stimuli, the high recognition rates for the corresponding natural stimuli are almost reproduced, while for other stimuli the emotional information is lost. In a free association perception test, a tendency towards the perception of the unintended category "disappointment" was found that only varied to a limited extend across stimuli.

Weitere Links

Deutsches Forschungszentrum für Künstliche Intelligenz
German Research Center for Artificial Intelligence