read.me

EVRC是用于高通公司的语音编码

**************************************************************************
TIA IS-127 Enhanced Variable Rate Codec, Speech Service Option 3 for
Wideband Spread Spectrum Digital System                            
**************************************************************************
This disk contains the Bit-Exact EVRC fixed point C simulation.
**************************************************************************
Bit-Exact Fixed Point C Source Code Simulation Version 1.21 11 Apr 1997

+ Created separate bin directory for executable.

+ Moved math library specific ".h" files to "dspmath" directory. 
**************************************************************************
Bit-Exact Fixed Point C Source Code Simulation Version 1.2 12 Mar 1997

+ Removed command line options for complexity report (-c) and
  rate-only mode (-r).

+ Added separate random number generator for rate 1/8 encode: e_ran_g().
  Output from running encoder/decoder separately is now bit-exact to
  that of running encoder/decoder together.

**************************************************************************
Bit-Exact Fixed Point C Source Code Simulation Version 1.1 12 Dec 1996

+ Modified durbin.c to correctly handly null inputs (for bit-exactness)

**************************************************************************
Bit-Exact Fixed Point C Source Code Simulation Version 1.0 09 Nov 1996

Changes incorporated into Version 1.0 include (but may not be limited to):

+ Noise supression scaling & block normalization
+ FCB Gain quanziter modularized
+ Bug-fixes in ACELP routines (for l_subfr / L_SUBFR problems) 
+ Re-written main driver/pre_processor memory handling.   
+ Other misc. fixes for 'bit-exact' output.

THE TEST SPEECH FILES INCLUDED WITH THIS RELEASE SHOULD NOT BE CONSIDERED
OFFICAL TEST VECTORS!  TIA Approved test vectors will be provided at a later 
date.

**************************************************************************
Bit-Exact Fixed Point C Source Code Simulation Version 0.99a 09 Aug 1996

Version "a" used for MOS test 08/96. Contains Nokia reduced complexity ACELP
search, updated noise supression scaling, 1/8th rate mods, etc. 


**************************************************************************
Bit-Exact Fixed Point C Source Code Simulation Version 0.95 11 July 1996

There are known known "obvious" bugs in this release.
The code is still being optimized and quality improved.

Current complexity is around 15-16 MOPS for encode, 2-3 MOPS for decode.

Changes/Bug Fixes/etc. to version 0.90 included in this release:

C-File				Desciption of Change      
------------------------------------------------------------------------------
bqiir.c				Re-written by LT to use more percision to improve overall quality
acelp_pf.c			ACELP Gain Problem fixed by Nokia
decode.c			ACELP Gain problem fix also needed to update decode.c..
c8_35pf.c			Fixed energy calc (ps1 & ps2) [Motorola]
lsp2a.c				Added a round() for a variable
auto.c				Optimized/Re-written by LT.
convh.c				Optimized by LT
durbin.c			Re-written & optimized by LT
filter.c			Optimized by LT
fndppf.c			Optimized/Fixed by LT (overflow problems in decimation filter)
getres.c			Optimized by LT
lspmaq.c			Optimized by LT
synfltr.c			Optimized by LT
w2res.c				Optimized by LT
rda.c				Qualcomm fixed background noise est... Needs some further testing.
impulser.c			Fixed Memory[0] inits == 0x800.

**************************************************************************
Bit-Exact Fixed Point C Source Code Simulation Version 0.90 7 June 1996

Known problems in this release:
1) RDA does not handle noise estimate properly. Force full rate operation with
   the "-l 4 -h 4" command line options to disable the RDA.

2) There is a slight gain problem that causes the encoder to generate approx.
1 to 2 db louder output.  This is related to the ACELP codebook search and
shall be corrected by release 1.00.

**************************************************************************

Included in the ./test directory are some test input files (.src files) with
the post-filtered and un-post-filtered speech output files (.pf & .npf) 
and the bit-stream output files (.bit).  Also included are complexity reports
generated with the simualtion.  The encoder are the files with the ".erp"
extension, and the decoder are the files with the ".drp" extensions.  These
files in the ./test directory ARE NOT qualified test vectors, they are only
supplied to insure correct operation on the users system.

The test files were created using gcc 2.6.3 under Linux ("PC Format")

Note that there are two compiler options when compiling the fixed point library
routines in ./dspmath directory.  By default, the library will use double
precision multiply functions which utilize ungsigned multiplies to determine
the result.  As an option, the user may define the "USE_ALT_DP31" compile time
option to use the ETSI double precision multiplies.  Final test vectors will
be supplied which allow the use of either DPF multiplies in an implementation.
See the makefile in ./dspmath for more information.

Complexity estimates can be done with the simulator by using the "-c <filename"
command line option.  If this option is specified, the simulator will create
a report on the complexity of the major modules.  This complexity measure is
an estimate based on the fixed point library weighted mops.

Questions and comments regarding the code should be sent to:
Tom Janiszewski
Lucent Technologies Network Wireless Systems
RM2A-203
67 Whippany Rd,
Whippany, NJ 08809
[e-mail]