Back to search

Aristotle University Of Thessaloniki (AUTH)

Σύγχρονες τεχνικές ανάλυσης και σύνθεσης φωνής και ομιλίας με τη χρήση οπτικών και ακουστικών πληροφοριών

Abstract

dc:description

The aim of the current thesis is to study technical methods of information processing associated with the face-based audiovisual communication.Following a study of different approaches, referring to anatomical, neurological, semantic, emotional and systemic influence factors, the exact object is defined in terms of purely linguistic content. In relation to the latter, a detailed characteristics specification of a native speakers control sample was determined, suitable for analysis of written form, auditory and visual stimuli.Regarding the written form a novel phonetic transcription representation is proposed, with the use of an acrylic directional graph instead of the classic serial sequence of phonetic symbols. The methodology was applied to the case of the Greek language, by means of a rule-based transcription system development.Furthermore, regarding the acoustic stimulus two novel fuzzy segmentation methods of the corresponding signal are proposed. Their design is based on both the mapping of the mechanical characteristics of the vocal tract in the audio signal features domain and the repetition of similar acoustic patterns during the pronunciation of a self-contained linguistic message. Two of the special features these methods present are the absence of a training phase and the capability of using additional phonetic transcription information.Moreover, regarding the visual stimulus, a methodology for extracting normalized images of the visible articulators region during speech via image processing techniques, is described. The proposal involves an affine transformation factorization using quaternions, combined with a single keyframe framework. Based on the fore mentioned infrastructure, the introduced methods feature counteracting the cumulative detection error and three-dimensional depth evaluation of a set of characteristic points, focusing on head pose parameters estimation in each frame. The methodology is completed by using the latter in a visible articulators isolation algorithm.Subsequently, a series of evaluation results are presented using an audiovisual recording sample consisted of DVB-T native speakers nightly news transmissions. An original interface implementation developed to meet the current and future needs of sample collection, excerpts annotation and required methods interoperability is also presented. Finally, reference is made to the proposed methods integration potential as well as to a roadmap for the creation of a visible articulators motion assessment model, during the pronunciation of language messages.

Degree

thesis:*
Grantor dc:publisher
Aristotle University Of Thessaloniki (AUTH)
Year dc:date
2012

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Avdelidis, Konstantinos
  • Αβδελίδης, Κωνσταντίνος

Subjects

dc:subject × 22

Rights

Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/29051

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

Avdelidis, Konstantinos; Αβδελίδης, Κωνσταντίνος. Σύγχρονες τεχνικές ανάλυσης και σύνθεσης φωνής και ομιλίας με τη χρήση οπτικών και ακουστικών πληροφοριών. Aristotle University Of Thessaloniki (AUTH), 2012. http://hdl.handle.net/10442/hedi/29051