Back to list

Prosodic and Paralinguistic Speech Parameters for the Identification of Emotions and Stress

Authors Jany-Luig, J.
Year 2017
Thesis Type Doctoral thesis
Topic Audio Signal Processing
Abstract In addition to what we say, it is mainly the way we say it which reveals how we think about the information we are just transmitting. A human listener uses prosodic and paralinguistic cues to decode this information, probably supported by information from facial expressions and gestures. This thesis explores to which extent a computer can solve the task of emotion and stress recognition from the human voice alone. Linguistic and musical approaches to grasp “prosody” are brought together and implemented as algorithms. Based on speech data both from an existing database of emotional speech and a self-created database of speech under stress, 27 different speech parameters are calculated from recorded speech signals without any metainformation. These parameters are then investigated regarding their ability to differentiate between different emotional states or different levels of cognitive stress.
Supervisors Höldrich, R.