Back to search

Πανεπιστήμιο Μακεδονίας

Η μηχανική μάθηση στην ηχοποίηση εκφραστικής χειρονομίας με τη χρήση στοχαστικών μοντέλων

Abstract

dc:description

A wider scope of this thesis is to investigate the gestural know-how of a musician and specifically of a pianist, which includes not only the acquisition of theoretical knowledge but also that of practical skills. Emphasis is given, in this thesis, in the expressive gesture and its variations. The gesture recognition is accomplished by using machine-learning algorithms and motion capture technologies. According to the literature review, several research approaches have been conducted aiming not only at the recognition of the performed gesture but also at the sonification of the gesture. However, a weakness that emerges is that the existing methodologies fail to take into account expressive variations that may occur during the performance of the incoming gesture with respect to the template gesture. This results in the recognition not being correct throughout the performance of the expressive gesture and in overlaps occurring between the different classes of classification. As a consequence, the sonification of the expressive gesture is not fluid and immediate.For this purpose, the proposed thesis presents a novel methodology which aims at a) the stochastic modeling, b) the gesture recognition and c) the sonification of the expressive gesture of the user, taking into account possible variations that may occur during the performance of the expressive gesture. This is achieved with the development of the Expert Operational Model, through which the confidence bounds are extracted. The added value of the Expert Operational Model, and thus of the confidence bounds, is that, during the recognition, the system prevents numerical errors that may occur due to variations, made either intentionally or not, and which may also be regarded as expressive elements of the performance of the gesture. The recognition of the expressive gesture is implemented by using machine-learning algorithms and specifically the Particle Filter algorithm. In addition, sound synthesis methods are applied to gesture sonification, providing the user the ability to resynthesize and manipulate the sound continuously and in real-time.The evaluation of the proposed methodology in comparison with established techniques and machine-learning algorithms, shows higher percentages of recognition, accuracy and similarity between the produced sound and the original. Another observation is that the quality of the produced and resynthesized sound in real-time, directly depends on the quality of the recognition of the expressive gesture. The better the performance of the incoming expressive gesture, the better, more fluid and without oscillations is the recognition of the expressive gesture. Hence, the re-synthesis of the sound is better and more fluid. Finally, the positive results of the evaluation, along with the proposed theoretical framework, confirm the efficient use of the confidence bounds in the recognition and sonification of the expressive gesture.

Degree

thesis:*
Grantor dc:publisher
Πανεπιστήμιο Μακεδονίας
Year dc:date
2016

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Volioti, Christina
  • Βολιώτη, Χριστίνα

Subjects

dc:subject × 19

Rights

dc:rights
Statement dc:rights
  • BY_NC
Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/42382

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Volioti, Christina; Βολιώτη, Χριστίνα. Η μηχανική μάθηση στην ηχοποίηση εκφραστικής χειρονομίας με τη χρήση στοχαστικών μοντέλων. Πανεπιστήμιο Μακεδονίας, 2016. http://hdl.handle.net/10442/hedi/42382