Application of multiresolution analyses for stop consonants in a cochlear implant model

Abstract : The classical Fourier transform with sliding time window may be inefficient for the characterisation of short events in speech, such as stop consonants, because of a non-adapted time-frequency compromise. In order to investigate this hypothesis, the basic Fourier analysis was compared to two multiresolution methods: a double Fourier transform with two simultaneous windows, varying in size, and a wavelet representation. Our signal processing was tested on the six French stop consonants according to a model of the French cochlear implant Digisonic (which uses the basic Fourier analysis). The stop consonants were analysed with both multiresolution methods, and then processed using cochlear implant specific processing. An automatic recognition was performed, and simulations of stop consonants perceived by implantees were presented to normal-hearing subjects. The results have not shown an improvement with the multiresolution techniques over the classical FFT analysis.
Type de document :
Article dans une revue
Acta Acustica united with Acustica, Hirzel Verlag, 2003, 89 (1), pp.181-184
Liste complète des métadonnées

http://hal.univ-brest.fr/hal-00827884
Contributeur : Mathieu Paquier <>
Soumis le : mercredi 29 mai 2013 - 18:35:45
Dernière modification le : jeudi 8 février 2018 - 11:08:23

Identifiants

  • HAL Id : hal-00827884, version 1

Citation

Mathieu Paquier, Jean-Christophe Béra, Christian Berger-Vachon. Application of multiresolution analyses for stop consonants in a cochlear implant model. Acta Acustica united with Acustica, Hirzel Verlag, 2003, 89 (1), pp.181-184. 〈hal-00827884〉

Partager

Métriques

Consultations de la notice

183