Skip to Main content Skip to Navigation
Theses

Contribution à l’identification automatique du locuteur sur des critères acoustiques et phonétiques

Odile Mella 1
1 PAROLE - Analysis, perception and recognition of speech
INRIA Lorraine, LORIA - Laboratoire Lorrain de Recherche en Informatique et ses Applications
Abstract : Methods for achieving automatic speaker recognition may be classified into two categories: pattern recognition based approaches that implicitly use interspeaker and intraspeaker variability of speech and approaches which explicitly take into account the sources of interspeaker and intraspeaker differences. The latter examine linguistic units in order to extract features which are relevant for speaker characterization. The aim of the present study is precise/ y to analyse the relative effectiveness of the three first formants of French vowels in the context /p-vowel-R/. These selected trigrams are a part of a larger set of preselected acoustic and phonetic parameters which lead us to record and digitalize a set of seventeen sentences, uttered four times by ten male speakers coming from the same region. We begin to expound the required know/egdes about speech production process and intraspeaker and interspeaker variability for understanding issues in speaker characterization. We then present a review of the studies in automatic speaker recognition according to the above classification. With regards of methods adapted from the domain of speech recognition, we restrict our review to the more recent works. In the other hand, we try to be more exhaustive for studies in speaker characterization. Following this, we develop the different stages of our study. After describing and praving the prese/ected acoustic and phonetic parameters, we deal with the the hand-Iabeling of our corpus according to strict rules. We examine furthermore the effectiveness of the three first formants of French vowels in the context /p-vowel-R/. For that purpose, we firstly propose an automatic method to determine reliable values of the three frequencies of the first formants of selected vowels. We besides discuss the re/iability of the results. For every vowel, we then analyse the relative effectiveness of each combination of formants frequencies and differences between these frequencies. This analyse is based on the computation of three "relevance indicators" from the results of speaker identification experiments.
Document type :
Theses
Complete list of metadata

Cited literature [61 references]  Display  Hide  Download

https://hal.inria.fr/tel-01739696
Contributor : Odile Mella <>
Submitted on : Wednesday, March 21, 2018 - 11:51:55 AM
Last modification on : Friday, February 26, 2021 - 3:28:06 PM
Long-term archiving on: : Thursday, September 13, 2018 - 8:07:30 AM

File

SCD_T_1993_0411_MELLA.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : tel-01739696, version 1

Collections

Citation

Odile Mella. Contribution à l’identification automatique du locuteur sur des critères acoustiques et phonétiques . Informatique et langage [cs.CL]. Université de Nancy I, 1993. Français. ⟨NNT : 1993NAN10411⟩. ⟨tel-01739696⟩

Share

Metrics

Record views

203

Files downloads

1601