This page is a demo page presenting sound examples, links to listening tests and other additional material for my phD thesis on
"Synthesis and expressive transformation of singing voice".
Chapter 3: ISiS: a concatenative singing synthesizer
Examples sounds from databases:
RT - / _ k O e~ k y l p e _ /:
sound 3.1
RT - / _ O v n i _ /:
sound 3.2
MS - / _ O v n i _ /:
sound 3.3
MS - / _ m j O z O t i s _ /:
sound 3.4
EL - / _ m j O z O t i s _ / :
sound 3.5
EL - / _ p a R k i N l a p e~ _ /:
sound 3.6
Synthesis engines:
Exemple of copy synthesis with the SVP engine and RT database:
sound 3.7
Exemple of copy synthesis with the SVP engine and MS database:
sound 3.8
Exemple of copy synthesis with the SVP engine and EL database:
sound 3.9
Exemple of copy synthesis with the PaN engine and RT database:
sound 3.10
Exemple of copy synthesis with the PaN engine and MS database:
sound 3.11
Exemple of copy synthesis with the PaN engine and EL database:
sound 3.12
Chapter 4: Control module: modelization of the synthesis parameters
Exemple of upward release f0 segment:
sound 4.1
Examples of transitions with various parameters from figure 4.9:
Example of resynthesis of the song from the harmonic analysis based on the f0 analysis for the chorus of "Les feuilles mortes" by Edith Piaf:
sound 5.3
This example aims to give an idea of the quality of the f0 estimation obtained for the commercial polyphonic recordings of the corpus, and of the harmonic analysis from which the loudness is estimated.
f0 parameters estimation:
Comparaison of f0 curve analyzed on original recording with the curve obtained from the proposed f0 model after estimation the model parameters on the original curve, for the chorus of "Les feuilles mortes" from Edith Piaf.
The curve are resynthesized with a single sinusoid, allowing to the perceptual difference between the original and the synthetic curve.
Sounds examples of pitch scaling and intensity timbre transformation based on spectral morphing with MFA envelope analysis on sustained vowels:
http://gillesdegottex.eu/Demos/DegottexG2016mfaenvsing/
Example of morphing-based intensity transformation for a singing extract synthesized with our ISiS synthesizer (using SVP engine):
Sounds submitted to the singing synthesis challenge at the Interspeech 2016 conference
- "Les feuilles d'Interspeech" (autumn leaves) - RT - PaN engine - a capella version:
sound 7.2
- "Les feuilles d'Interspeech" (autumn leaves) - RT - PaN engine - with musical accompaniment:
sound 7.3
Other synthesis examples (used in [Feugere2016] for evaluation purposes):
- "Les feuilles d'Interspeech" (autumn leaves) - MS - SVP engine - a capella version:
sound 7.4
- "Au temps d'Interspeech" (summer times) - RT - PaN - a capella version:
sound 7.5
- "Au temps d'Interspeech" (summer times) - RT - SVP - a capella version:
sound 7.6
- "Au temps d'Interspeech" (summer times) - MS - PaN - a capella version:
sound 7.7
Bonus
Research is a long process made of trials and errors. But when it comes to computer sciences and sound, errors may sometimes generate the most interesting and fun results.
To illustrate this statement, you will find below a compilation of my "best failures", most of which result from parametrization errors or bugs in our ISiS synthesis system that occured while developing it.
But those bugs have hopefully all been fixed by now, such that I would unfortunately not be able to reproduce such sounds again (which makes them even more valuable!).