[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: ICA, a key to solve the puzzle of hearing

From a limited number of microphones, it is possible to extract more than this number of signals - that's basically what stereo and ambisonics (not to mention transaural) does. Using several spaced capsules (as against coincident or virtually coincident)  can work in higher order ambisonics, theoretically, though there are some 'noise' issues at present. Whtether the capsules are directional or omni (or 'in between' - semi-cardioid as used in the SoundField Mic), it's not necessary to assume that switching between capsules is what you want to do. - you're actually just looking for a signal that is optimised for a particular source. By treating your signals in a matrixed fashion, you can rapidly try out succesions of 'decodes' (and these can have frequency dependent components, too). Nevertheless, who or what is steering? - is this a learning algorithm designed to mimic scene analysis? - it would seem that you need a best-fit explanation for the entirety of components in a detected sound field, otherwise you'll have 'bits left over' - isn't that right?
----- Original Message -----
Sent: 25 March 2005 17:43
Subject: Re: ICA, a key to solve the puzzle of hearing

On Fri, 25 Mar 2005, Guoping Li wrote:

 ICA was designed to solve the problem of cocktail party effect using
 higher order statistical methods. Is it possible that we can use this
 technique and implement it in the hearing aid or cochlear implant?

Toth Laszlo wrote:
As far as I remember, when ICA is used for blind signal separation, it
needs at least N input signals (for example, from a microphone array) for
extracting N independent components.

That's true, but there is no reason a priori why one shouldn't build a device that incorporates dozens of microphone capsules in the users clothing, and that could separate dozens of different sources. Then you'd just need a little switch that allows you to quickly scan through those separated sources to find the one you want to listen to. It wouldn't be very "biological", but it might nevertheless work quite well.

Dr. Jan Schnupp
University Laboratory of Physiology             St Peter's College
Oxford University                               New Inn Hall Street
Parks Road, Oxford OX1 3PT, UK                  Oxford OX1 2PL
Tel (01865) 272513                              Tel (01865) 278889
Fax (01865) 272469

No virus found in this incoming message.
Checked by AVG Anti-Virus.
Version: 7.0.308 / Virus Database: 266.8.3 - Release Date: 25/03/05
No virus found in this outgoing message.
Checked by AVG Anti-Virus.
Version: 7.0.308 / Virus Database: 266.8.4 - Release Date: 27/03/05