Modelisation resonantique de la coarticulation des consonnes occlusives sonores
Author
Source
Issue
Vol. 2009, Issue 14-15 (31 Dec. 2009), pp.7-23, 17 p.
Publisher
Centre de Recherche Scientifique et Technique pour le Développement de la Langue Arabe
Publication Date
2009-12-31
Country of Publication
Algeria
No. of Pages
17
Main Subjects
Abstract EN
In speech, sounds are linked to each other and their limits are fuzzy : the speech results of the continuous movement of the vocal apparatus and not from discrete positions.
Among the problems found in the domain of Automatic Speech Processing, there is the difficulties related to the omnipresent variability in continuous speech.
This latter result from the fact that the articulation is in constant evolution and depends even on the context in which the sound is pronounced.
The notion that the same sound can be caracterized by a set of different formants is important.
The exploitation of a model consists to test, compare and interpret variations of features due to variations of command parameters.
For this objective, a corpus of units [ViCV2] ([Vj] = [a, i, u], [V2] a 10 oral vowels and [C] a [b, d, g]) has been recorded in order to study the coarticulation modelling.
We will infer consonantic targets in terms of resonances on the basis of the adjacent vocalic targets as anchorage points of a follow-up of affiliation.
Our study is based on the resonances representation.
Modelling rules will be given.
These are implemented and tested by a rule compiler.
Abstract FRE
Dans le message parle, les sons sont enchaines les uns aux autres, leurs limites sont floucs : la parole resultc du mouvement continue! de I'appareil vocal, et non d*une suite dc positions.
Parmi les problcmes qui rendent le Traitement Automatique de la Parole dclicat, il y a les difTicultcs liees & la variability omniprcsente en parole continue.
Celles-ci rcsultent du fait que l'articulation est en constante evolution et depend mcme du contexte dans lequel le son est prononce.
La notion qu'un mcme son peut etre caractcrisc par une meme serie dc formants diflerents cst importante.
La mise a l'cpreuve puis I'exploitation d'un modelc va consister a tester, interpreter, comparer les variations des caracteristiques de sortie dues aux variations des paramctres de commande.
Dans ce but, un corpus [V1CV2] ([V,] = [a, i, u], [VJ = 10 voyelles orales et [C] = [b, d, g]) a et6 enregistre afm d'&udicr la modelisation dc la coarticulation.
Nous infererons des cibles consonantiqucs en termes dc resonances en partant des cibles vocaliques adjaccntes, comme points d’ancrage d’un suivi d’affiliation.
Notre etude cst basee sur la representation en resonances.
Des regies dc modelisation seront donnecs.
Ces demi6rcs sont implementees ct tcstees par un compilateur de regies.
American Psychological Association (APA)
Guerti, Mhania. 2009. Modelisation resonantique de la coarticulation des consonnes occlusives sonores. AL-Lisaniyyat،Vol. 2009, no. 14-15, pp.7-23.
https://search.emarefa.net/detail/BIM-945794
Modern Language Association (MLA)
Guerti, Mhania. Modelisation resonantique de la coarticulation des consonnes occlusives sonores. AL-Lisaniyyat No. 14-15 (2009), pp.7-23.
https://search.emarefa.net/detail/BIM-945794
American Medical Association (AMA)
Guerti, Mhania. Modelisation resonantique de la coarticulation des consonnes occlusives sonores. AL-Lisaniyyat. 2009. Vol. 2009, no. 14-15, pp.7-23.
https://search.emarefa.net/detail/BIM-945794
Data Type
Journal Articles
Language
French
Notes
Includes bibliographical references : p. 23
Record ID
BIM-945794
