Re: Auditory model sound resynthesis

[Malcolm Slaney <malcolm@xxxxxxxxxxxx>]

At 4:02 PM 3/17/95, Thierry Rochebois wrote:
>This method is quite good for harmonic sounds.
>But it is not accurate enough (i.e. it is not as accurate
>as our auditory system) for transients and/or noisy sounds(cymbals
>for example).

A better way to do this is to use an auditory model.  Somebody already
provided a reference to Toshio Irino's paper.

I published some work last year that builds on the results from Irino-san's
paper.  This work also uses previous work in spectrogram inversion
(Griffin) and auditory model inversion (Shamma).

We can now do perfect reconstructions from a cochlear model (Lyon's to be
specific) and from a higher level representation known as a correlogram.
The techniques we use are easily adapted to other cochlear models.

The mathematics of the work was published in the 1994 ICASSP Proceedings
("Auditory Model Inversion for Sound Separation", Slaney, Naar and Lyon,
Volume 2, pp. 77-80.) A postscript version is available from
        ftp://ftp.apple.com/pub/malcolm/CorrelogramInversion.postscript

A higher level version of the paper will be published in the Proceedings of
the 1994 NIPS conference.  It is titled "Pattern Playback in the '90s."  A
preprint is available from
        ftp://ftp.interval.com/pub/papers/malcolm/NIPSPatternPlayback.psc.Z

I hope this helps.

-- Malcolm