home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!psgrain!hippo!ucthpx!uctvax.uct.ac.za!clrmic08
- From: clrmic08@uctvax.uct.ac.za
- Newsgroups: comp.sys.ibm.pc.soundcard
- Subject: Re: Speech recognition
- Message-ID: <1992Aug12.143348.202347@uctvax.uct.ac.za>
- Date: 12 Aug 92 14:33:48 +0200
- References: <levitt.712956324@informatics> <1992Aug9.222744.202250@uctvax.uct.ac.za> <Bsr1Cn.LHr@usenet.ucs.indiana.edu>
- Organization: University of Cape Town
- Lines: 33
-
- In article <Bsr1Cn.LHr@usenet.ucs.indiana.edu>, mfleig@silver.ucs.indiana.edu (Mark W. Fleig) writes:
- > In article <1992Aug9.222744.202250@uctvax.uct.ac.za> clrmic08@uctvax.uct.ac.za writes:
- >
- > Okay. Speaking as an utter eletrical-idiot, (Computer Science: A Study
- > in How to Study Computer Science) how might this apply to the current
- > crop of sampling boards we have out there. PAS16/SB/Ultrasoon. I have
- > no idea if these boards do anything but store the digital samples, and
- > actual audio analysis sounds unlikely. Am I wrong with this notion?
- >
- > Some of the original ideas of template matching seemed the most likely
- > candidates, can we fashion software to take what these boards give us,
- > possibly dissect a sampled single word file, pull out the good bits,
- > and match it up to a fuzzy database? Would anyone like to give a quick
- > lesson on what goes into an .SND/.WAV file?
-
- > ,----------------------------. "Therefore, make peace with your god, whatever
- > : Mark Fleig \ you conceive him to be. Hairy thunderer, or
- > : mfleig@silver.ucs.indiana.edu\ cosmic muffin. With all its hope and renewal,
- > `-------------------------------' the world continues to deteriorate. Give up."
-
- Here in the darkest of Africa we don't get those kind of cards. Soundblaster
- and Adlib cards are available but very very expensive. We used a card made by a
- local company. It's got 16 single ended channels capable of sampling at 30Khz,
- we only used four but ideally more channels should be used. I don't know if any
- of these cards that you've mentioned have got built in filters.
-
- The part about matching templates is dead right, after filtering the audio we
- took samples for each word stored them as templates and then after each
- recognition we update each template.
- Which ever way you go about it try and do most of the processing in analogue!
-
- GOOD LUCK !!!!
-
-