Yamaha Releases Singing Synthesis Software

loopdloop writes "The world's first singing synthesis software, Vocaloid, was released by Yamaha this month at the Los Angeles NAMM show. Simply type in the lyrics and notate the vocal expressions to create a completely computer-generated singer. There are also audio demos of the product available." Update: 01/26 21:14 GMT by S : An earlier NYT-authored preview of this software has also been covered on Slashdot.

Re:Deja Vu

[weston]

Did you know about Ingo Titze and "Pavarobbotti"? [ncvs.org]

Actually...

[mOoZik]

I had the chance to try it out at NAMM and it is VERY difficult to get it to "sing." It can probably be used adequately for backup vocals, but again, it takes a lot of work to get it to sound human. Nevertheless, a step in the right direction.

Simone

[Lord_Dweomer]

I think they should have called it SimOne v1.0

Interestingly enough, this made me think of something I read in William Gibsons blog a long time ago. I don't know where it is now though. It was about how in the future, people will be able to take a movie or something on their computer, and tell the computer to replace all the actors heads with dog heads for example, and change what they do and say with simple commands. Perhaps this software is the lower level beginning of making that happen, we'd just need some higher-level controls to make it easy for everybody to use.

Really Bad Synths

[Entropy248]

Just like many Yamaha firsts, this one may have been overhyped a bit. This sounds like a real person singing in the way that a synth brass pad sounds like a trumpet. There is no way in hell you would ever even consider that these noises were made by a human being. Yes, I understand that most of the samples are in Japanese and might not sound normal to me anyway. But, even if you listen to the ONE in English alone, it sounds like the Bell Labs female voice, but screechy and obnoxious instead of like a drugged out cigarette smoker after a trachyotemy (sp?).

first **commercial** software

[stephentyrone]

cool as this may be, it's definitely not the first singing synthesis software. CNMAT [berkeley.edu] at berkeley had a neural-net based additive synthesis engine in the mid 90's that did a pretty good human voice (it could even reproduce the voices of specific individuals, as I recall), and did other instruments as well (a mean viola).

I can't find a link to an actual demo of it simulating a human voice, but here's a page [berkeley.edu] that documents its use to reproduce the sound of a suling (javanese wooden flute). Does a good job too. I've heard it demo'd with a human voice, and it was pretty good (though the neural net needed additional input - the syllable being sung - obviously).

i'm sure that many of the other academic computer music labs around the world had similar software long before yamaha introduced this package. still cool, though.

Not the only person working on this...

[jhhl]

I'm glad someone is releasing this kind of thing - many people gave worked with voice algoritms for years doing vocal synthesis.

• Tellus cassette recording FALSE PHONEMES came out in 1988 or so, from Harvestworks [harvestworks.org]
  
• plenty of good resynthesis is out there, using various vocal synth technologies like LPC.
  
• MBROLA sings pretty well if you feel like coding up the data
  
• FLINGER is open source. It uses the lyric tags in a MIDI stream to drive the synthesis algorithm.
  
• Perry Cook's excellent physical modeler Sing1.2 for the NeXT is very, very old at this point.
  
  So there is prior art spewing out all over the place.
  
  and how could I leave off:
  eddie and eedie [otisfodder.com]? (search for "Some Velvet Morning"

asx and webpage html

[technix4beos]

The "webmaster" who wrote the linked page of demos is linking to ASX files, which in turn link straight to the self-named mp3 files on the server.

In case the direct "save/play" links do not work with your browser and OS, just replace the asx with mp3, and enjoy.