University of Patras
Αναγνώριση συναισθημάτων από ομιλία με χρήση τεχνικών ψηφιακής επεξεργασίας σήματος και μηχανικής μάθησης
Abstract
dc:descriptionThe present dissertation entitled "Emotion Recognition from Speech Using DigitalSignal Processing and Machine Learning Techniques" was submitted to theUniversity of Patras, in partial fulfillment of the requirements for the degree of Doctorof Philosophy.The dissertation is structured as follows:In Chapter 1, an introduction to human-computer interaction is presented. Theimportance as well as the necessity of emotion recognition from speech is outlined,together with the main directions in the research area of emotion recognition. Finally,the contribution of the Thesis and its structure are presented.In Chapter 2, the reader is introduced to more details on the area of emotionrecognition from speech. Specifically, the globally inherited emotion models areintroduced, and there follows an introduction to emotion recognition from speechsignal, utilizing the acoustic information, which is topic of the present Thesis.In Chapter 3 a initial study on a complete end-to-end emotion recognition fromspeech system is presented, which is based on decision trees. The evaluation of thesystem is performed on acted speech data both without noise and in various type ofnoise conditions, and compared to subjective evaluation of humans.In Chapter 4 the process of designing, and implementing a real world speech corpusof emotions, which serves the purpose of conducting research in the area of emotionrecognition from real-life speech is described. The system suggested is designed todetect negative emotional states in real world conditions, both in speaker dependentand speaker independent mode using Gaussian Mixture Models, (GMM). Thesuggested architecture utilizes low and high level descriptors for constructing thefeature vector.In Chapter 5 a emotion recognition system based on Universal Background Model,(UBM-GMM), is evaluated-integrated and evaluated for different kind of real-worlddata: for the purpose of supporting patients with mental disorders and evaluated fordetecting emotions in atypical event situations, respectively. Further, the improvementof an emotion recognition system utilizing speech parameter selection is presented.In Chapter 6 a novel method for modeling co-occurrence of affective states, as theyappear in real life speech is proposed. The proposed architecture leads to significantimprovement over a number of affective states.In Chapter 7 a summary and conclusion of the work together with future directionsare presented.
Degree
thesis:*- Grantor dc:publisher
- University of Patras
- Year dc:date
- 2012
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Kostoulas, Theodoros
- Κωστούλας, Θεόδωρος
Subjects
dc:subject × 20- Αναγνώριση συναισθημάτων
- Αναγνώριση συναισθηματικών καταστάσεων
- Μοντελοποίηση συναισθημάτων
- Διαλογικό σύστημα
- Παραμετροποίηση ομιλίας
- Επιλογή παραμέτρων ομιλίας
- Διεπαφή ομιλίας
- Συνυπάρχοντα συναισθήματα
- Emotion recognition
- Affect recognition
- Emotion modeling
- Dialogue system
- Speech parameterization
- Feature selection
- Speech interface
- Co-occuring emotions
- Επιστήμες Μηχανικού και Τεχνολογία
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Engineering and Technology
- Electrical Engineering, Electronic Engineering, Information Engineering
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/30450