Third International Workshop on Models and Analysis of Vocal Emissions for Biomedical Applications (MAVEBA 2003)
In this contribution the recognition of stress and emotional state is analysed by speech signal analysis using Liljencrant-Fant's model. It is based on the knowledge that some parameters of glottal pulses, obtained by this model, are changed owing to stress, hence they are suitable for the detection of speaker's stressed (abnormal) state. Two procedures for the analysis of these parameters are described in detail. The first of them is an analysis of parameters of randomly chosen speech parts (of phonetically constant length) that makes fewer demands on segment selection, the second is an analysis of speech parts going one by one in time. The methods were applied to sound recordings made at stressed oral examinations at a university. The results obtained show the applicability of these parameters and methods especially for speech analysis when we have at our disposal a signal recorded in the normal (steady) state of speaker.
Index Terms. stress, glottal excitation
Full Paper (reprinted with permission from Firenze University Press)
Bibliographic reference. Bostik, M. / Sigmund, M. (2003): "Speaker stress detection by analysis of glottal excitation", In MAVEBA-2003, 87-90.