Jasmine Music Technology - Home

yv enhancer demos
some thoughts before testing vocaloid,
yv enhancer questionnaire
basic scientific and methodological provisions
jmt tips for using yamaha vocaloid
yv enhancer 1.0
related links

Performance Modeling technology can Enhance Yamaha Vocaloid!

2nd Example: LEON can be taught to sing!
April, 28, 2004.


Yes! Another "small step for man" has been made in automatic adding of expression elements to Vocaloid Yamaha using our Performance Modeling technology (PM).

Experimental version of Onyx Arranger and special PMStyle can now calculate curves for Velocity, Pitch Wheel, Noise, Gender Factor, Harmonics, Brightness and Resonance parameters. Moreover, we tried to include into Performance Modeling algorithm imitation of singing with melisma, when long note is replaced by short notes gliding through steps of scale characteristic of soul manner. Of course, this stage is just the troubleshooting of compatibility of controlling commands made by Onyx Arranger and Vocaloid. And creating special styles for Vocaloid is different issue, and a demanding one, too. Nevertheless, the achieved result seems quite hopeful. Well, ...

A reminder: PMStyle used for the second experiment, has calculated Velocity, Pitch Wheel, Noise, Gender Factor, Harmonics, Brightness curves automatically. No manual editing at all!

To analyze the result yourself you can take the following files in mp3 and *.mid format:

Internet_Angel_YVLeon_JMT.mp3 [1,06 Mb] Full demo (voice and background)
IA_Verse1_YVLeon_JMT.mp3 [557 Kb] Verse1 (voice only)
IA_Refrain_YVLeon_JMT.mp3 [321 Kb] Refrain (voice only)
IA_Verse1_YVLeon_(VFile).mid Verse1 - Yamaha Vocaloid (LEON) data file
IA_Refrain_YVLeon_(VFile).mid Refrain - Yamaha Vocaloid (LEON) data file
Internet_Angel_(Processed_Tune).mid PM Processed tune

ALL (3) MIDI files in ZIP

Comments by Alexei Ustinov, CTO Jasmine Music Technology

What is essential? Why LEON, why LOLA, and not Craig David or Avril Lavigne? Is it enough to collect a number of phonemes and ensure the vocal synthesis, without distorting artist's original timbre to get virtual, electronic Sting or Madonna?

Of course not! The identity of wave forms to original voice in itself is just the tip of the iceberg compared to what has to be modeled in attempt to make artificial sound more natural. Besides timbre, every vocalist possesses a unique intonation and an artistic set of performer's techniques, without which singing is basically impossible and without which he or she won't be recognized: Thus, to become closer to real live vocalist, we need not only synthesis, but the full set of performer's nuances, i.e. commands controlling this synthesis!

Imagine that your computer contains perfect 3D images, for example Arnold Schwarzenegger or Sharon Stone, and you can create your videos controlling these objects in the program. But how can you achieve realistic, not robotic movement of these objects? How can one make artificial movement so convincing, that no one distinguishes it from real? Well, here's the rub! It's not enough to have perfect images (convincing realistic voices in our case), one needs to learn to control them... And this is very difficult, for a human being can very easily recognize the fake, when it concerns physical laws of movement... And singing IS physics.

Demo Song details

Tune author - A.Ustinov,
Lyrics - I.Fomchenkov / A.Ustinov,
Arrangement - A.Ustinov (Background was created with the help of Onyx Arranger 2.1 using OOStyle of Coolio's wonderful song "Gangsta's Paradise").

Internet Angel

I don't care much for good signs
And I can't read the future
Pennsylvania bound
Or awaking in Berlin
You are always with me
In the silent display

Oh, my love
Internet Angel, Internet Angel
Oh, my love
Internet Angel, Internet Angel
Oh, my love
Internet Angel, Internet Angel