From gos.ukc.ac.uk!harrier.ukc.ac.uk!ukc!mcsun!uunet!tut.cis.ohio-state.edu!pt.cs.cmu.edu!k.gp.cs.cmu.edu!dandb Tue Apr  3 16:48:35 BST 1990
Article 482 of comp.dsp:
Path: gos.ukc.ac.uk!harrier.ukc.ac.uk!ukc!mcsun!uunet!tut.cis.ohio-state.edu!pt.cs.cmu.edu!k.gp.cs.cmu.edu!dandb
>From: dandb@k.gp.cs.cmu.edu (Dean Rubine)
Newsgroups: comp.dsp
Subject: Re: Manipulating sampled sounds.
Message-ID: <8631@pt.cs.cmu.edu>
Date: 28 Mar 90 07:02:25 GMT
References: <00933E08.7E4E7D40@max.berkeley.edu> <14504@cbnewsc.ATT.COM> <695@xdos.UUCP>
Organization: Carnegie-Mellon University, CS/RI
Lines: 25

In article <695@xdos.UUCP> doug@xdos.UUCP (Doug Merritt) writes:
>I'm about to write some spectrograph software, so I'd appreciate hearing
>about such limitations/pitfalls and what workarounds are best. All of
>my current reference books discuss only single FFT's, so I'm kind of
>winging it on the "stitching together" part of it.

     There are basically two methods of resynthesizing a signal from
successive (overlapped) Fourier Transforms: the Overlap Addition Method and the
Filter Bank Summation Method.  What you want to do is check out some of the
(vast) literature on the Short Term Fourier Transform.  A good starting place
is L. Rabiner and R. W. Schafer's "Digital Processing of Speech Signals",
chapter 6.  They show, among many other things, the conditions necessary to be
able to reconstruct the original signal from successive, overlapped Fourier
transforms.

   Of course, if you're just making a spectrograph program to get pretty
amplitude vs frequency vs time plots, there's no need to do any "stiching
together", is there?  The stitching part comes after you manipulate the
transforms and then want to reconstruct a signal from these transforms.

-- 
ARPA:       Dean.Rubine@CS.CMU.EDU	
PHONE:	    412-268-2613		[ Free if you call from work ]
US MAIL:    Computer Science Dept / Carnegie Mellon U / Pittsburgh PA 15213
DISCLAIMER: My employer wishes I would stop posting and do some work.