Aristotle University Of Thessaloniki (AUTH)
Τεχνικές ψηφιακής επεξεργασίας ομιλίας στην αναγνώριση συναισθημάτων
Abstract
dc:descriptionΤhe subject of this thesis is the recognition of emotions from speech. The investigation began with a review of methods that extract acoustic characteristics and methods for classifying speech into emotional states. The technique adopted for recognizing emotions is as follows. First, statistics are estimated on curves of acoustical characteristics, such as the fundamental frequency, the various frequency band energies, and the formants of the vocal tube. Then, for the classification of speech using the aforementioned characteristics, the Bayes classifier is used, where the probability density function (pdf) each characteristic was modeled as a Gaussian or as a mixture Gaussians. In the experiments, we used two speech databases, one of actors in the Danish language and one of military personnel in the American language. An algorithm for finding the highly correlated acoustic characteristics is proposed, based on the multiple correlation coefficient. The advantage of the proposed method is the exact calculation of confidence intervals of the cross-correlation coefficient. The non-linear Newton-Raphson method is applied twice, one to reverse the pdf of the cross-correlation coefficient, and one for finding the correct number of factors required for the convergence of a hypergeometric function involved in the calculation of pdf of the cross-correlation coefficient. Finally, the algorithm for emotion recognition was applied to a database that consists of speech from students, that are trained in a virtual earthquake environment. The signals were employed, except the speech, was sweating and heart pulses.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2008
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Βερβερίδης, Δημήτριος
- Ververidis, Dimitrios
Subjects
dc:subject × 21- Αναγνώριση συναισθημάτων
- Υπολογισμός ακουστικών χαρακτηριστικών
- Μέθοδοι επιλογής χαρακτηριστικών
- Μέθοδος διασταυρωμένης επικύρωσης
- Μίγματα γκαουσιανών κατανομών
- Συντελεστής πολλαπλής συσχέτισης
- Εφίδρωση
- Παλμοί καρδιάς
- Εικονικό περιβάλλον σεισμού
- Emotion recognition
- Acoustic features extraction
- Feature selection methods
- Cross-validation
- Gaussian Mixture Models
- Multiple correlation coefficient
- Sweat and heart beat rate
- Virtual earthquake environment
- Φυσικές Επιστήμες
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Natural Sciences
- Computer and Information Sciences
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/18935