Expressive control of singing voice synthesis using musical contexts and a parametric F0 model

Demo page

Authors : Luc Ardaillon, Céline Chabot-Canet, Axel Roebel
IRCAM - UMR STMS (IRCAM - CNRS - Sorbonne Universités), Paris, France
contact : luc.ardaillon@ircam.fr

This page is a demo page presenting all the sounds used in the evaluation for the article "Expressive control of singing voice synthesis using musical contexts and a parametric F0 model", submitted to the Interspeech 2016 conference.
You can also still access the listening test with the instructions used for the evaluation using this link.

Test I : CMOS comparing default settings vs style models

The 1st test aimed at evaluating the preference on pairs of synthesis, in regards to expressivity, between a default configuration using the mean parameters of each style and the style models, in a CMOS procedure.


Man voice :

Leroux style
extract default model
1
2
3
4
Distel style
extract default model
1
2
3
4


Woman voice :


Piaf style
extract default model
1
2
3
4
Greco style
extract default model
1
2
3
4


Test II : style recognition

This test aimed at evaluating the recognition rate of the styles used for synthesis, among 2 possible styles, in an ABX test procedure, based only on the F0 variations and phonemes durations.


Man voice :

Originalsynthesis
ExtractLerouxDistelLerouxDistel
1
2
3
4
Woman voice :


Originalsynthesis
ExtractLerouxDistelLerouxDistel
1
2
3
4

The sounds used in this test are under Copyright © 2016 Ircam, Institut de recherche et coordination acoustique/musique.