How to get started:

   Edit the Makefile.

	You should configure a few machine-dependencies and what
	compiler you want to use.

  	The code works both with ANSI and K&R-C.  Use
	-DNeedFunctionPrototypes to compile with, or
	-UNeedFunctionPrototypes to compile without, function
	prototypes in the header files.
 
   Make addtst

	The "add" program that will be compiled and run checks whether
	the basic math functions of the gsm library work with your
	compiler.  If it prints anything to stderr, complain (to us).

   Edit inc/config.h.

   Make

   	Local versions of the gsm library and the "compress"-like filters
	toast, untoast and tcat will be generated.

   	If the compilation aborts because of a missing function,
	declaration, or header file, see if there's something in
	inc/config.h to work around it.  If not, complain.

   Try it

	Grab an audio file from somewhere (raw u-law or Sun .au is fine, 
    	linear 16-bit in host byte order will do), copy it, toast it,
	untoast it, and listen to the result.
    
	The GSM-encoded and -decoded audio should have the quality
	of a good phone line.  If the resulting audio is noisier than
	your original, or if you hear compression artifacts, complain;
	that's a bug in our software, not a bug in the GSM encoding
	standard itself.

Installation

   You can install the gsm library interface, or the toast binaries,
   or both.

   Edit the Makefile
	
	Fill in the directories where you want to install the
	library, header files, manual pages, and binaries.

	Turn off the installation of one half of the distribution
	(i.e., gsm library or toast binaries) by not setting the
	corresponding directory root Makefile macro.

   make install

	will install the programs "toast" with two links named
	"tcat" and "untoast", and the gsm library "libgsm.a" with
	a "gsm.h" header file, and their respective manual pages.


Optimizing

   This code was developed on a machine without an integer
   multiplication instruction, where we obtained the fastest result by
   replacing some of the integer multiplications with floating point
   multiplications.

   If your machine does multiply integers fast enough,
   leave USE_FLOAT_MUL undefined.  The results should be the
   same in both cases.

   On machines with fast floating point arithmetic, defining
   both USE_FLOAT_MUL and FAST makes a run-time library
   option available that will (in a few crucial places) use
   ``native'' floating point operations rather than the bit-by-bit
   defined ones of the GSM standard.  If you use this fast
   option, the outcome will not be bitwise identical to the
   results prescribed by the standard, but it is compatible with
   the standard encoding, and a user is unlikely to notice a
   difference.


Bug Reports

   Please direct bug reports, questions, and comments to
   jutta@cs.tu-berlin.de and cabo@informatik.uni-bremen.de.


Good luck,

   Jutta Degener,
   Carsten Bormann

--
Copyright 1992, 1993, 1994, by Jutta Degener and Carsten Bormann,
Technische Universitaet Berlin.  See the accompanying file "COPYRIGHT"
for details.  THERE IS ABSOLUTELY NO WARRANTY FOR THIS SOFTWARE.