home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!zephyr.ens.tek.com!uw-beaver!ubc-cs!destroyer!gumby!yale!yale.edu!qt.cs.utexas.edu!cs.utexas.edu!sun-barr!ames!elroy.jpl.nasa.gov!swrinde!swrinde.nde.swri.edu!kent
- From: kent@swrinde.nde.swri.edu (Kent D. Polk)
- Newsgroups: comp.sys.amiga.audio
- Subject: Re: Halving sample speed while keeping pitch
- Message-ID: <13327@swrinde.nde.swri.edu>
- Date: 28 Aug 92 17:40:25 GMT
- References: <9208241809.AA00474@.nairobi.inel.gov.inel.gov.> <ERICJ.92Aug24131911@lagos.cfsat.Honeywell.COM> <1992Aug25.155155.20785@kong.gsfc.nasa.gov>
- Sender: news@swrinde.nde.swri.edu
- Organization: Southwest Research Institute, San Antonio, Texas
- Lines: 50
-
- In article <1992Aug25.155155.20785@kong.gsfc.nasa.gov> arensb@kong.gsfc.nasa.gov (Andrew Arensburger - RMS) writes:
- >ericj@hwcae.Honeywell.COM (Eric Jacobsen) writes:
- >
- > I did this thing once, by chopping the original signal into 50-
- >sample segments. The segment size should be longer than the wavelength
- >of the signal, but short enough to be almost unnoticeable. Then just play
- >each segment twice, so that
- > "hello world"
- >becomes
- > "hheelllloo wwoorrlldd"
- >
- >I didn't conduct an extensive study of appropriate segment lengths or
- >sampling rates, but it seemed to me that slowing down human speech in this
- >way worked fairly well (sounded like someone speaking verrrry slowwwly,
- >albeit a bit choppy), but made musical instruments sound like hell.
-
- You have hit on some very intriguing aspects of how the aural system handles
- events. If someone is interested in pursuing this type of solution to the
- problem, I have some references that might be of use. I have been studying
- the problems associated with auralization of waveform data (obtaining
- information from waveforms aurally) for several years now. I won't go into
- detail here, but I use high-speed repetition of small signal segments to
- enhance comprehension as well as to place signal information in a certain
- 'context', and it works very well under the correct circumstances. Sample
- rates and segment lengths are critical aspects which can completely alter the
- perceived information.
-
- I would say that the key issues here are to get a handle on how auditory
- streaming works and then understand what categories your sounds fit into so
- you can set the correct sample lengths for the desired effects.
- Unfortunately, I haven't worked much with speech, but from I have read and
- seen, speech would be handled the easiest with this method. Certainly the
- method described for speeding up speech is a 'picket fence' phenomenon where
- contextual information is heavily relied on by the aural system to fill in
- the missing signal elements. Have played a bit with music containing singing
- though. Very interesting results.
-
- I think you would find that a little masking would help improve the
- comprehension for both speech and music, though I expect that the music
- segment lengths would have to be a little different for each contributing
- sound source. (Not very easy :^).
-
- Oh yes. I use "The Web Data Acq. and Signal Processing Suite" for this
- research. If you have tried out the demo and wondered what all those strange
- options in the 'Play' filter were for... (Gotta put a plug in for Pete and
- Dave :^)
-
- Kent Polk: Southwest Research Institute
- Internet : kent@swrinde.nde.swri.edu
- UUCP : $ {cs.utexas.edu, gatech!petro, sun!texsun}!swrinde!kent
-