Acoustic Measures characterizing anger across corpora collected in artificial or natural context

Marie Tahon, Laurence Devillers, LIMSI-CNRS

This paper aims at studying the difference of acoustic manifestations for anger across corpora collected in artificial, manipulated or natural context. It also aims at finding measures for naturalness in emotive corpora. Actually it is quite difficult to evaluate the degree of naturalness of a corpus except considering the task. As for more acted corpora, anger is often stronger; we can imagine that it exists a kind of distance between anger and the all corpus. This distance is here evaluated with basic acoustic descriptors in several corpora collected. We show some differences between the acoustic features obtained for anger in these different contexts and propose measures of naturalness.