Path: utzoo!attcan!uunet!mcsun!ukc!cam-eng!ajr
From: ajr@eng.cam.ac.uk (Tony Robinson)
Newsgroups: comp.dsp
Subject: Re: DSP Hearing Aids?
Message-ID: <15797@rasp.eng.cam.ac.uk>
Date: 22 Oct 90 16:55:10 GMT
References: <33682@nigel.ee.udel.edu> <952@eplunix.UUCP> <10810@goofy.Apple.COM>
Sender: ajr@eng.cam.ac.uk
Organization: Cambridge University Engineering Department, UK
Lines: 25

In article <10810@goofy.Apple.COM> zarko@apple.com (Zarko Draganic) writes:
>There was a neat project going on when I was in college last year (I 
>didn't get the chance to work on it). The basic idea was to apply pattern 
>recognition techniques to the input audio signal, and recognize the voiced 
>plosive phonemes (which are very troublesome in certain types of hearing 
>imparments involving large high frequency attenuations). The recognized 
>phonemes would then be substituted with "in-band" phonemes, synthesized or 
>taken from a foreign language. Gradually the user would map the new, 
>in-band phonemes to the lost voiced plosives.  Don't know how far it's 
>come or if anyone other than the U. of Waterloo is working on it; know of 
>any similar efforts?

Well, I will be working on something similar.  The basic idea is that
telephone quality speech is bandlimited, 300Hz to 3.3KHz, and as a result is
quite intelligible but not very natural.  It is intelligible because the
power spectra above 3.3KHz is mostly redundant, so it should be possible to
regenerate a plausible high frequency spectra and patch the speech.  This is
most likely to work for unvoiced speech as then you don't have to worry about
the phases of the added high frequencies, but it is unvoiced speech that
suffers most from the bandlimiting anyway, so that should be okay.

Maybe all this has been done before :-(


Tony Robinson.