Thu Aug 8 14:33:07 EDT 1996 msiegler

Upgrade adc2mfcc -> wave2mfcc.

Allows pipes/devices instead of just files.  Reads in blocks instead
of en-masse.  Uses the math coded by eht Eric Thayer which is more
reliable. Allows conversion of partial files.


8 July 1997,msiegler@cs.cmu.edu
 First recording of versions (duh) and things are very much out of hand.
 (27 June 1997 -- fixed DC offset and dither, now they actually do something

 DC offset is implemented with a causual filter instead of batch mode, and uses
 a fairly straightforward approximation to MAP estimate of mean, assuming
 a Gaussian signal.  This breaks down to a linear weighting of estimated mean
 for less than 10000 points, and full weighting afterwards

 Dither is now half-bit instead of single-bit in size.

29 Oct 1997,msiegler@cs.cmu.edu
 Implementing an FFT overlap-and-add based convolution for front end processing.
 This is intended to allow some sort of arbitrary filter to be applied to the
 incoming audio, for example a telephone channel simulation.

 Will use direct convolution until

         The Numerical Recipies in C convolution and FFT code was used here for expedience,
         but this means that redistribution is not permitted.


version 0.3

4 Nov 1997,msiegler@cs.cmu.edu

 Using in-house FFT code now

 Now allows the use of FIR filtering in the pre-processing stage.  Native 32-bit
 floats are expected.  No special checking of the filter is made.  Uses simple
 FFT code.

 some minor bug fixes to eliminate warnings in compile time.
