Pandora (R)

[mmalcolm crawford <m.crawford@xxxxxxxxxxxxxx>]

William Clocksin (of Cambridge University) kindly drew my attention to
the following which may be of interest:

        http://www.prosoniq.com/pandora.html

"PANDORA MODULE I - Voice Separation
                package info

This basic module of PANDORA( which is included in the basic engine
allows separating the vocal component from a composite music signal
consisting of other (non-vocal) instruments. If you thought this would
be impossible - think again: PANDORA(s de-voicing algorithm becomes
attracted by all voice and voice-like utterances within a mix of
different instruments and splits the mix into a vocal track and (in the
case of a stereo file) two tracks containing the residual instrumental
mix. The quality achieved meets even the highest pro-level studio
requirements and provides a basis on which a new and previously
unthinkable signal treatment during mastering can be introduced."

The image they show of the voice of Michael Jackson extracted from the
background music suggests that the algorith (whatever it is, and they
stress that it's based on "human hearing", not FFTs) deals primarily
with voiced utterances, and probably not with fricatives.

Has anybody else heard of this, used it, have any comments?

Best wishes,

mmalc.