Skip to Main content Skip to Navigation

Multipitch estimation methods for the separation of speech and musical signals

Abstract : This thesis deals with the multipitch estimation problem for speech and musical mixtures where the number of sources is unknown. In the speech context, we propose an iterative method which successively estimates the pitches. The complex "voiced/unvoiced" nature of the mixtures is characterized by a two-part model which consists in "sums of harmonic sinusoids + autoregressive process". The pitch estimation is based on the maximization of a penalized likelihood term. This likelihood also allows to define a voicing detector used to estimate the number of sources. In the musical context, we propose three new methods which simultaneously determine the pitches. Based on a spectral peak classification of the mixture, they differ by the classification technique. All the methods allow to estimate the number of sources. Moreover they are able to take into account the spectral overlaps which characterize the musical mixtures and can be used on musical chords.
Document type :
Domain :
Complete list of metadata
Contributor : Ecole Télécom ParisTech Connect in order to contact the contributor
Submitted on : Friday, September 24, 2004 - 8:00:00 AM
Last modification on : Friday, July 31, 2020 - 10:44:07 AM
Long-term archiving on: : Thursday, September 30, 2010 - 6:11:32 PM


  • HAL Id : pastel-00000723, version 1



Julie Rosier. Multipitch estimation methods for the separation of speech and musical signals. domain_other. Télécom ParisTech, 2003. English. ⟨pastel-00000723⟩



Record views


Files downloads