readme

高通的cmda语音压缩算法is96a源代码. 针对自己的dsp将最耗时的c改成汇编就几乎是商用代码了.

/**********************************************************************/
/* QCELP Variable Rate Speech Codec - Simulation of TIA IS96-A,service */
/*     option one for TIA IS95, North American Wideband CDMA Digital  */
/*     Cellular Telephony.                                            */
/*                                                                    */
/* (C) Copyright 1993, QUALCOMM Incorporated                          */
/* QUALCOMM Incorporated                                              */
/* 10555 Sorrento Valley Road                                         */
/* San Diego, CA 92121                                                */
/*                                                                    */
/* Note:  Reproduction and use of this software for the design and    */
/*     development of North American Wideband CDMA Digital            */
/*     Cellular Telephony Standards is authorized by                  */
/*     QUALCOMM Incorporated.  QUALCOMM Incorporated does not         */
/*     authorize the use of this software for any other purpose.      */
/*                                                                    */
/*     The availability of this software does not provide any license */
/*     by implication, estoppel, or otherwise under any patent rights */
/*     of QUALCOMM Incorporated or others covering any use of the     */
/*     contents herein.                                               */
/*                                                                    */
/*     Any copies of this software or derivative works must include   */
/*     this and all other proprietary notices.                        */
/**********************************************************************/

Version History
---------------
2.02 - Fixed a minor bug in the rate decision.  Previously, the lowest of
       the "high background noise" thresholds used the incorrect coefficients,
       which caused the threshold to be computed incorrectly.

       Fixed a minor bug in the LPC to LSP conversion.  Previously, the final
       linear interpolation was done incorrectly, causing a slight error.

2.01 - Changed minor bug in CB gain quantization.  Previously, if the
       average of the previous 2 CB gains, in dB, was negative, the
       floor function was not computed correctly.

2.00 - Inserted non-differential LSP quantization to improve the
       performance during frame erasures. (This is IS96-A).  

1.02 - Changed sample io so if the input file has Nx160 samples, the
       encoder will process N frames (rather than N-1).

----------------------------------------------------------------------

This is a C simulation of the TIA IS-96-A speech codec for use with the
TIA IS-95 North American Wideband CDMA Digital Cellular Telephony
Standard.

Compiling the Simulation
------------------------

The simulation was originally written for compilation on Sun Sparcstations,
using the traditional "make" utility.  Typing "make" on a Sparcstation
should compile the various modules and create the executable "qcelp" which
runs the main simulation.  On other platforms, some modification of the
makefile may be needed, or compilation using other methods may be
required.

Running the Coder
-----------------

Once compiled, a brief usage statement can be seen by entering "qcelp"
with no command line arguments.  The coder options and command line
arguments are described in more detail below.

File Types
----------

First a note about the types of files used by this simulation.  Speech
files contain 16 bit "byte-swapped" samples (low byte first).  The samples
are first divided by 4.0 to create an overall maximum value of
2^14.  After decoding, the samples are multiplied by 4.0 to restore 
the original level to the output files. The 2^14 value is slightly 
above the maximum mu-law value assumed in the IS-96-A specification, but 
the simulation performs acceptably with this
slightly larger dynamic range.  As with any digital speech system, 
optimal performance will be achieved when the speech energy is ~20db
below the maximum dynamic range, which makes clipping of the samples
a rare occurrance.  Because of this, best results will be achieved if
the input speech files are approximately 3 bits below the maximum
16 bit dynamic range.

The byte-swapping and division/multiplication by 4.0 is done in the
"read_samples" and "write_samples" routines in the file "io.c".  This
file can be modified to accomodate different file formats.  However,
whatever the file format, the samples used by the coder should have
a maximum value near 2^14.

Packet files are files containing packed frames of speech data.  For each
frame of speech, the file contains 12 byte-swapped, 16 bit words: 1 word
containing the data rate, followed by 11 words containing the packed
speech data (packed in accordance with the packing tables specified in
IS-96-A).  The value for each data rate is:

Value in Packet File        "Rate"       Data Bits per Frame  
------------------------------------------------------------
  4 = 0x0004                   1                171
  3 = 0x0003                  1/2                80
  2 = 0x0002                  1/4                40
  1 = 0x0001                  1/8                16
  0 = 0x0000                 Blank                0
 15 = 0x000f           Full Rate Probable       171
 14 = 0x000e                Erasure               0

Unused bits are set to 0. For example, in a Rate 1/8 frame, the packet 
file will contain the word 0x0100 (byte-swapped "0x0001") corresponding
to Rate 1/8, followed by one 16 bit word corresponding to the 16 data bits
for the Rate 1/8 frame (in byte-swapped form), followed by ten 16 bit
words containing all 0 bits.

Command Line Options
--------------------

 -i infn  : required :	contains the name of the input speech file,
			or the name of the input packet file if only
			decoding is being performed (see the -d option below).

 -o outfn : required :	contains the name of the output speech file name,
			or the name of the output packet file if only
			encoding is being performed (see the -e option below).

 -d			only perform the decoding function.  the input
			file must contain packets of compressed data.

 -e			only perform the encoding function.  the output
			file will contain packets of compressed data.

			If neither the -d or the -e options are invoked,
			the coder performs both the encoding and decoding
			functions by default.

 -h max			sets the maximum allowable data rate to "max", using
			the codes for the rates described in the table 
			above (Rate 1 = 4, etc).

 -l min			sets the minimum allowable data rate to "min", using
			the codes for the rates described in the table 
			above (Rate 1 = 4, etc).

			If neither the -h or -l options are invoked,
			the coder allows the data rate to vary between
			Rate 1 and Rate 1/8 using the rate decision algorithm
			described in the IS-96-A specification.
			In addition, if max != min, the data rate varies
			between "max" and "min" using the same rate decision
			algorithm, where the data rate is set to "max" if
			the selected data rate is >= "max", and the data rate
			is set to "min" if the selected data rate is 
			<= "min".  See the "select_rate" routine in the
			file "ratedec.c" for more information.

 -f frames		stop processing the speech after "frames" 160 samples
			frames of speech. (Note: the frame counter starts
			at 0, not 1)

			If the -f option is not invoked, the coder processes
			all of the speech until the end of the input file
			is reached.

 -p flag 		If flag is set to 0, the postfilter is disabled.

			If the -p option is not invoked, the postfilter is
			enabled during decoding.

 -s sigfn		Read signaling data from the file "sigfn". This
			option allows rates to be determined on a frame
			by frame basis.  The file "sigfn" takes the form

frame_num_1	max_rate_1	min_rate_1
frame_num_2	max_rate_2	min_rate_2
...

			This causes frame number "frame_num_1" to have
			a maximum allowable rate of "max_rate_1" and a
			minimum allowable rate of "min_rate_1".  (Note:
			the first frame is frame 0, not frame 1.)  The same
			holds for frame_num_2, and any frames beyond that
			(Note: the frame_num parameters must be in INCREASING
			order.).  For example, to process the input file
			"input.spch", put the output speech in "output.spch",
			with the 10th frame forced to "blank" and the 15th
			frame forced to be between Rate 1/2 and Rate 1/8, 
			the following could be run.

qcelp -i input.spch -o output.spch -s signal.data

			where the file "signal.data" contains

10	0	0
15	3	1

			In this case, every frame except for frame numbers
			10 and 15 will have the rate selected as normal.

			If the -s option is not invoked, the rate for every
			frame will be selected as normal.


Verifying the Simulation
------------------------

To ensure that the simulation is working properly, several files are included
to verify the performance of the coder.  The file "m01v0000.raw" is an
unprocessed speech file.  The file "m01v0000.pkt" is a packet file, obtained
by running "qcelp -i m01v0000.raw -o m01v0000.pkt -e".  The file 
"m01v0000.mstr.raw" is the speech file, processed by the qcelp algorithm,
obtained by running "qcelp -i m01v0000.pkt -o m01v0000.mstr.raw -d".
Once the simulation is compiled, this file should be processed as 
described above, and the resulting packet file and processed speech file 
should exactly match the files included with this software.