ID Fuel and New Scientist are both running a story about Zenph Studios. It's a New England startup run by a couple of dotcom millionaire geeks who helped to develop VoIP. The angle they're selling is that they've found a way to convert old, scratchy recordings (of dead piano players, mainly), into high-definition MIDI files. They trigger a Disklavier Pro piano with the MIDI file, and re-create the performance, 'bringing the dead back to life'. So far, I'm thinking 'so what?'. But behind the PR silliness, they're claiming to have solved the old problem of polyphonic transcription which, in the past, has only been possible with a worse-than-useless 80% success rate. It's easy enough to convert a monophonic sequence of notes into MIDI, but it's been pretty much impossible to do it with chords. Apparently that's what the Zenph team (three programmers and one piano expert) have cracked.
If it's true, then it could be a breakthrough as important as the invention of sampling. Imagine a system that could automatically create a 100% accurate MIDI rendition of any piece of music. You could analyse a Jimi Hendrix guitar part, edit the notes a little, and replay it with pan-pipe samples over a new track, with the feel & vibe completely intact. Guitar synths would not longer need a hex pickup, and the piano bar would be obsolete. This is all in the future (there's no suggestion that this is a real-time system at the moment, and it only works with piano) but it's still very exciting. I just wish they'd explain how it works...
OMG, is that Jesus-spam? How droll.
ReplyDeleteI think this new tech is going to make for some interesting changes... I mean, what motivation is there for someone to learn to play the piano to a masters level, if there's software out there that can just take your work and turn it into a template?
OTOH, there's the interesting possibility that they can abstract two functions out of this software - the "notes being played" bit, and the "piano is played like this" bit... it'd be interesting if they can abstract other instruments, so instead of just "this is the same piece of music played with pan-pipe samples", you could get "here's the same piece of music, played as though a pan-pipe player were playing it..."
It'd be great for a studio musician/engineer... but will put a lot of session musicians out of work...
Hey, you know like some cultures have Christmas on a different day? Does that work with April Fools' Day as well?
ReplyDeleteThie story is without doubt Cold Fusion for the piano. That first comment looks like a Google translation of some kind of self-help greetings card.
The notion that this technology would put session musicians out of job is just stupid - the transcribing of monophonic sound to MIDI is quite good, yet it don't replace having a real saxophone player, trumpet player or a player of nearly any other monophonic instrument.
ReplyDeleteThere is just so mamy aspects of the performance that cannot be captured by MIDI - just think about Jimi Hendrix - how on earth could all his techniques and effects be transcribed as MIDI data? The piano is propably among the easiest instruments to apply this kind of processing to, but even here there are problems with techniques such as half-pedalling, use of sostenuto pedal, etc.
This is amazing technology. I love piano recordings and will be very happy to hear recreations. I've heard player piano rolls played via a Disklavier and was pleased; given that the original player died 100 years ago!
ReplyDeleteOf course this won't affect studio musicianship. Real-time emotionally resonant interpretation of music can only come from a real-live emoter.
Maybe it's just me, but the crackle and bad sound quality of those old recordings is half the appeal of listening to them.
ReplyDeleteDude, they have plugins for that too!
ReplyDelete;)
Hendrix rendered in pan-pipe samples "with the feel and vibe intact?"
ReplyDeleteNot likely, I think...
Maximus
Voltage
Polyphonic transcription with 80% success rate sounds quite amusing to me.
ReplyDeleteAny software tips?
I don't see anywhere on their site a claim that their 'polyphonic transcription' is automated by computer in any way. I'm imagining a human plugging in notes, and using the original recording for timing. Is it possible that you are all reading something more into this? Or can someone correct me with information from their site? They are after venture capital, by the way, and have past histories of getting it.
ReplyDelete