University of Patras
ΣΥΣΤΗΜΑ ΑΥΤΟΜΑΤΗΣ ΑΝΑΓΝΩΡΙΣΗΣ ΔΙΑΚΡΙΤΗΣ ΟΜΙΛΙΑΣ ΒΑΣΙΣΜΕΝΟ ΣΕ ΦΩΝΗΜΑΤΑ ΚΑΙ ΠΡΟΣΑΡΜΟΓΗ ΤΟΥ ΣΕ ΣΥΝΕΧΗ ΟΜΙΛΙΑ
Abstract
dc:descriptionTHE CURRENT THESIS ADDRESSES THE PROBLEM OF IMPLEMENTING A ROBUST ISOLATED- WORD SPEECH RECOGNITION SYSTEM FOR MEDIUM VOCABULARY, BASED ON TEMPLATE MATCHING PRINCIPLES, WITH THE VIEW OF MAKING IT INDEPENDENT OF VOCABULARY SIZE AND DECREASING COMPUTATIONAL AND MEMORY REQUIREMENTS. A COMPLEMENTARY TARGET OF THIS WORK WAS THE PROBLEM OF ADAPTATION OF SUCH A ROBUST ISOLATED-WORD RECOGNITION SYSTEM TO CONTINUOUS SPEECH. USING PROPERLY STRUCTURED CEPSTRAL COEFFICIENTS, A MULTISECTION CODEBOOK DESIGN APPROACH WAS ADOPTED, FOLLOWED BY THE USEOF THE L.V.Q. ALGORITHM, A MODIFIED DISTANCE CRITERION INCORPORATING TIME CORRELATION INFORMATION, RESULTING TO A CREATION OF A REPRESENTATIVE CODEBOOK AND TO A REDUCTION OF THE COMPUTATIONAL AND MEMORY REQUIREMENTS OF THE RECOGNITION SYSTEM. AT THE SAME TIME THE PERFORMANCE OF THE SYSTEM WAS IMPROVED, BEING COMPARABLE TO SIMILAR SYSTEMS DESCRIBED IN THE LITERATURE. THE EXISTING ISOLATED-WORD REFERENCE DATA ADOPTED TO INCOMING CONTINUOUS SPEECH TRAINING DATA USES A MODIFIED L.V.Q. ALGORITHM. ALSO, BECAUSE THE PROSODY PARAMETERS ARE THE MAIN FACTORS DIFFERENTIATING ISOLATED AND CONTINUOUS SPEECH, A NOVEL TECHNIQUE WAS EMPLOYED TO TRANSFORM THE ISOLATED-WORD PROSODY PARAMETERS TO CONTINUOUS SPEECH INCORPORATING A LINEAR TRANSFORMATION PROCEDURE. FURTHERMORE IT WAS FOUND THAT THE MULTISECTION CODEBOOK DESIGN TECHNIQUE, WHICH WAS USED FOR THE ISOLATED WORD SYSTEM, MINIMISES THE COARTICULATION PROBLEMS OF THE CONTINUOUS SPEECH.
Degree
thesis:*- Grantor dc:publisher
- University of Patras
- Year dc:date
- 1995
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Tsopanoglou, Anastasios
- Τσοπάνογλου, Αναστάσιος
Subjects
dc:subject × 16- ΑΛΓΟΡΙΘΜΟΣ L.V.Q.
- ΑΛΦΑΒΗΤΟ ΠΟΛΛΑΠΛΩΝ ΤΜΗΜΑΤΩΝ
- ΑΝΑΓΝΩΡΙΣΗ ΒΑΣΙΣΜΕΝΗ ΣΕ ΦΩΝΗΜΑΤΑ
- ΠΡΟΣΑΡΜΟΓΗ ΔΕΔΟΜΕΝΩΝ ΑΝΑΦΟΡΑΣ
- ΠΡΟΣΩΔΙΑΚΕΣ ΠΑΡΑΜΕΤΡΟΙ
- ΣΥΓΚΡΙΣΗ ΧΑΡΑΚΤΗΡΙΣΤΙΚΩΝ
- L.V.Q. ALGORITHM
- MULTISECTION CODEBOOK DESIGN
- PHONEME-BASED RECOGNITION
- PROSODIC PARAMETERS
- REFERENCE DATA ADAPTATION
- Template matching
- 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/4960