How does real affect affect affect recognition in speech?


Share/Save/Bookmark

Truong, Khiet Phuong (2009) How does real affect affect affect recognition in speech? thesis.

open access
[img]
Preview
PDF
6MB
Abstract:The aim of the research described in this thesis was to develop speech-based affect recognition systems that can deal with spontaneous (‘real’) affect instead of acted affect. Several affect recognition experiments with spontaneous affective speech data were carried out to investigate what combination of acoustic (and also lexical) features and classifier yields best performance. Laughter and subjectivity in meetings, and Arousal (active vs. passive) and Valence (positive vs. negative) were some of the affect types that we aimed to recognize. While it is very convenient to use acted emotional speech in emotion recognition studies, the use of posed affective data undermines the ecological validity of these studies.
Following several attempts to capture real affect in the field, we elicited spontaneous vocal and facial expressions with people playing videogames. Furthermore, we believe that the evaluation of affect recognizers has been an underexposed topic, hence, we have tried to raise awareness by applying a standardized evaluation scheme to the field of speech-based affect recognition and by proposing a cross-validation scheme that matches more closely real world situations. The findings of our research indicate that real affect drastically affects the development of speech-based affect recognizers with respect to data collection and annotation, which in turn affects the performance of these recognizers. The subjective and context-sensitive character of spontaneous affect, and the subtleness of spontaneous affective expressions appear to be complicating factors for the machine learning-based approach that is commonly adopted in automatic affect recognition.
Item Type:Thesis
Faculty:
Electrical Engineering, Mathematics and Computer Science (EEMCS)
Research Group:
Link to this item:http://purl.utwente.nl/publications/67355
Official URL:http://dx.doi.org/10.3990/1.9789036528801
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page