An innovation on polyphonic transcription ("H.Hacihabiboglu" )


Subject: An innovation on polyphonic transcription
From:    "H.Hacihabiboglu"  <H.Hacihabiboglu(at)SURREY.AC.UK>
Date:    Mon, 25 Apr 2005 13:37:41 +0100

Interesting news from the New Scientist!... ____________________________ Ivory encore for dead piano greats * 22 April 2005=20 * Special Report from New Scientist Print Edition=20 * Mick Hamer=20 NEXT month music lovers in Raleigh, North Carolina, will be able to hear = two of the greatest pianists of the 20th century in concert. Both the = pianists, however, are long dead. Zenph Studios, a software company based in Raleigh, has found a way to = take a music recording and convert it into a live concert played on real = instruments. The concert will be a completely faithful rendition of the = original pianists' work. Zenph resurrected a scratchy mono recording of Bach's Goldberg = Variations, made by the Canadian pianist Glenn Gould in 1955 and a = recording of a Chopin prelude by Alfred Cortot in 1928. Cortot died in = 1962, Gould in 1982. The breakthrough that Zenph has achieved is to extract the sounds from = audio recordings and convert them into a high-resolution version of = MIDI, the standard way of coding music for computers. To do so they had = to tackle the problem of polyphonic transcription - distinguishing = several notes played simultaneously. While researchers have been trying = to achieve this for years, previous attempts have managed to identify at = best 80 to 90 per cent of notes correctly - with about 10 per cent = missing and another 10 per cent wrong (New Scientist, 22 December 2001, = p 50). Zenph now says it has found a way to do this, although for commercial = reasons it won't release the details. But the company is confident = enough to have organised the concert, at which a Disklavier Pro piano, = one of a handful of concert grands that can record and play back = high-definition Midi files, will replay Gould and Cortot's work. The = piano will replicate every note struck, down to the velocity of the = hammer and position of the key when it was played. "We have only begun seeing excellent results in the past few weeks," = says John Walker, president of Zenph Studios. "The results are note = perfect." Walker has an impressive track record. Before founding Zenph = in 2002, he was a leading developer of VoIP, the system that allows = phone calls to be carried on the internet. Walker says that the precise timing of notes is almost as important as = identifying the correct notes. One of Zenph's final checks is to play = back the conversion on the Disklavier and to make an audio recording of = it. The engineers then play back a stereo version of the music: one = channel has the original recording, the other has the recording of the = conversion. "If they're different by even a few milliseconds, the ear = immediately identifies that something's wrong - there's a slight echo = effect," Walker says. "Resurrected pieces include a scratchy recording of Bach's Goldberg = Variations, and a recording of a Chopin prelude" "The project at Zenph is definitely very, very interesting," says Anssi = Klapuri of the Tampere University of Technology, Finland, who is one of = the world's leading experts on polyphonic transcription. The company is now working on a recording made at a private party by the = jazz giant Art Tatum two years before his death in 1956. There are many = recordings that have never been released because of some flaw, such as = background noise or an out-of-tune piano. Zenph hopes that recording = companies will use the new technology to make recordings from this type = of material, or to clean up noisy recordings. >From issue 2496 of New Scientist magazine, 22 April 2005, page 27 =20 (http://www.newscientist.com/article.ns?id=3Dmg18624966.700&feedId=3Donli= ne-news_rss20 = <http://www.newscientist.com/article.ns?id=3Dmg18624966.700&feedId=3Donli= ne-news_rss20> ) _____________________________________________ Dr H=FCseyin Hacihabiboglu Research Fellow I-Lab / CCSR University of Surrey, Guildford GU2 7XH Surrey, United Kingdom Tel: +44 1483 683435 Fax: +44 1483 686011 e-mail: h.hacihabiboglu(at)surrey.ac.uk=20


This message came from the mail archive
http://www.auditory.org/postings/2005/
maintained by:
DAn Ellis <dpwe@ee.columbia.edu>
Electrical Engineering Dept., Columbia University