Détection du fondamental de la parole en temps réel : application aux voix pathologiques

Fadoua Bahja 1, 2
2 PAROLE - Analysis, perception and recognition of speech
Inria Nancy - Grand Est, LORIA - NLPKD - Department of Natural Language Processing & Knowledge Discovery
Abstract : This thesis is part of researches aimed at determining the fundamental frequency of speech signals. The first contribution is related to the development of real time pitch detector algorithms, based on an implicit circular autocorrelation of the glottal excitation. Among all the pitch detection algorithms described in the literature, few of them are able to tackle correctly all the problems of pitch tracking. For this reason, we expanded our scope of investigation and proposed new algorithms based on wavelet transforms. To evaluate the performances of the proposed algorithms, we used two databases : Bagshaw and Keele. The results we obtained prove that our developed algorithms compare favourably with the best reference pitch detector algorithms described in the literature. The second contribution of this thesis concerns the implementation of a voice conversion system in order to enhance the pathological voice. In this case, we talk about a correction system. Our main contribution, concerning voice conversion, lies in the prediction of Fourier cepstral coefficients related to the excitation signal. This new kind of prediction allowed us to implement conversion systems whose results, either they are objective or subjective, validate the proposed approach.
Complete list of metadatas

Cited literature [94 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00927147
Contributor : Joseph Di Martino <>
Submitted on : Tuesday, January 14, 2014 - 1:02:09 PM
Last modification on : Friday, August 9, 2019 - 3:10:32 PM
Long-term archiving on : Tuesday, April 15, 2014 - 4:19:18 PM

Identifiers

  • HAL Id : tel-00927147, version 1

Citation

Fadoua Bahja. Détection du fondamental de la parole en temps réel : application aux voix pathologiques. Traitement du signal et de l'image [eess.SP]. Université Mohammed V-Agdal UFR Informatique et Télécommunications Laboratoire LRIT Unité associée au CNRST, URAC 29, Faculté des sciences, 2013. Français. ⟨tel-00927147⟩

Share

Metrics

Record views

1003

Files downloads

8646