home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.sys.amiga.audio
- Path: sparky!uunet!snorkelwacker.mit.edu!thunder.mcrcim.mcgill.edu!poly-vlsi!sevigny
- From: sevigny@vlsi.polymtl.ca (Benoit Sevigny)
- Subject: Re: Halving sample speed while keeping pitch
- Message-ID: <1992Aug28.022825.13059@vlsi.polymtl.ca>
- Sender: news@vlsi.polymtl.ca (USENET News System)
- Organization: Ecole Polytechnique de Montreal
- References: <1992Aug26.133525.26944@vlsi.polymtl.ca> <1992Aug26.134200.27011@vlsi.polymtl.ca> <ERICJ.92Aug27101148@lagos.cfsat.Honeywell.COM>
- Date: Fri, 28 Aug 1992 02:28:25 GMT
- Lines: 29
-
- In article <ERICJ.92Aug27101148@lagos.cfsat.Honeywell.COM> ericj@hwcae.Honeywell.COM (Eric Jacobsen) writes:
- >In article <1992Aug26.134200.27011@vlsi.polymtl.ca> sevigny@vlsi.polymtl.ca (Benoit Sevigny) writes:
- >
- > To half the speed, simply overlap your FFTs by 512.
- >
- >To clarify: the initial spectrogram FFTs overlap 50% in time. This does
- >sound like a good way to do it, but I wonder about boundary effects
- >when you transform back to the time domain. Is this something you've
- >actually had experience with, particularly with speech?
- >
- > Of course you can transpose by any factor by spacing your FFT's starting
- > point by x/(FFT's length) where x is your transposing factor
- >
- >Ayup, pretty cool, assuming it works. If it does it offers a nice
- >solution to the problem. It might be better to use Hartley transforms
- >instead of Fourier transforms, though, to avoid real->complex->real
- >conversion problems.
- >
- I agree that there might be a continuity problem. I think it could be
- solved by transforming (whatever transform) every point to its frequency
- domain and then by averaging each component over a segment corresponding to
- the transposition factor (e.g. 768pts/1024pts for 0.75 factor) before you
- transform the data back to its time domain. I think this should result in
- a more continuous wave at the expense of processing time, of course!
- --
- ===============================================================================
- /// Benoit Sevigny, software | Skipping every x86 lines in your sources
- \\\/// engineering student | might remove both your most irreductible
- \\V/ sevigny@info.polymtl.ca | bugs and bottlenecks :-)
-