Hacker News Reader

Comment by camtarn | Hacker News Reader

Slightly different. Gorillaz band members are usually voiced by one real artist (although this artist can change from album to album), whereas Hatsune Miku is one step removed - while there was a real voice actress providing the foundation for her voice, the actual lyrics sung don't pass through any human mouth.

The persona of Hatsune Miku also seems to have evolved a little more organically: the style and personalities of Gorillaz are basically driven by Damon Albarn and Jamie Hewlett; Hatsune Miku started out as an image on a software box, but got adopted by the internet community who produce videos based on her, with quite diverse personality and style.

As for artificial intelligence composition, Emily Howell is probably the closest we've got at the moment: http://arstechnica.com/science/news/2009/09/virtual-composer...

View on HN

I want to see the crossover!

EDIT: Really, I want to see her animated with the keyframes by Hewlett.

Cope's earlier work with EMI is also pretty awesome; the idea of using as a training set a composer's body of work in order to produce new music in that same style is very cool.

http://artsites.ucsc.edu/faculty/cope/mp3page.htm

So do they record and then playback, or is it voiced fully by computers?

Fully voiced by the computer. Think of a midi keyboard that also has a 'syllable' component for each note: So for example you make her sing the syllable 'Ka' in middle C. There are also a dozen or so other knobs that control how it is sung at that particular point in time.

So the singing is synthesized, but the band is real?

Backwards world :)

Entirely computers. Hatsune Miku and similar software are basically voice synthesizers geared towards music, as I understand it, not hugely unlike other text-to-speech programs but with more control of pitch, tone, and so forth.

Also, as far as making artificial intelligences compose and perform, it might be interesting to hook up Hatsune Miku to a Markov chain generator[1], and feed it a bunch of j-pop as input. Would probably take some fine-tuning and modification to get it to give output that the synthesizer would understand (as it would be dealing with both words and notes), but it's not too far fetched.

[1] Which isn't an AI by any means, being more of way of generating text based on statistical correlations between words in the input body of data, but it comes up with some rather interesting things.

You input the lyrics, notes, scale, tempo and dynamics; the software will produce a voice based on those parameters. Then you can record it to share it as an audio file. Here's a quick FAQ on the official site: http://www.vocaloid.com/en/technical_faq.html

Reply on news.ycombinator.com