bugfix.notes

4.8kcelp语音压缩编码程序

Date: Wed, 24 Mar 93 12:30:44 -0500
From: [e-mail] (Peter Kabal)
Message-Id: <[e-mail]>
To: [e-mail] (Joe Campbell)
Subject: celp distribution
Content-Length: 17181
X-Lines: 539
Status: RO

Joe, I grabbed the latest version from furmint.nectar.cs.cmu.edu.  Vince
Cale has now put everything in a tar.Z file.  I had grabbed the stuff earlier
when it was just loose in some directories.  Some of the stuff has been badly
mangled.  I ran a diff with the latest version, and there is an apology
in the comments to the effect that he has messed up the ifdefs.  Indeed he
has, that is why the C-version does not run.

I suggest you send him (and me), the correct celp.c and makefile.  That
should put the distribution back in order.  There are other bits and
pieces lying around which probably were not in the original distribution.

Back to the Fortran version.  The version I have now runs on both Sun and
DEC without further modification.  The DEC version complains about speech
differences, but after processing the DEC and Sun output files measure
up to within 80 dB.  The DEC output file is slightly longer than the Sun,
but that is presumably due to differences in how they handle partial
records at the end.

Fortran routines for 1016 CELP
The following renames were required:
bitperm.h  ->  bitpermute.h
bitprot.h  -> bitprotect.h
cbgain.tlb -> cbgain.table
cbsearch.f -> cbsearch.F
csub.f     -> csub.F
dcodham.f  -> decodeham.f
disto.f    -> distortion.f
encodham.f -> encodeham.f
gaindcod.f -> gaindecode.f
gainncd2.f -> gainencode2.f
gainncod.f -> gainencode.f
ldelaynr.f -> ldelay_nr.f
lsp34.tbl  -> lsp34.table
lspd34.f   -> lspdecode34.f
mexcite.f  -> mexcite.F
mtrxgen.f  -> matrixgen.f
pgain.tbl  -> pagain.table
ptdcode.f  -> pitchdecode.f
ptncod.f   -> pitchencode.f
postfilt.f -> postfilt2.f
read_err.f -> read_error.f
readme.mke -> readme.make
savesg.f   -> save_sg.f
smcbgn.f   -> smoothcbgain.f
smpgn.f    -> smoothpgain.f
smt.f      -> smoothtau.f
wrterr.f   -> write_error.f

In addition, the following routines were more extensively modified and
renamed:
celp.f     -> celp.F     changes + ultrix changes
cli.f      -> cli.F      changes + ultrix changes
iodisk.f   -> iodisk.F   changes + ultrix changes
makefile.mke -> Makefile
psearch.f  -> psearch.F  changes + ultrix changes

Sun refers to Fortran under SunOS 4.1.3.
DEC refers to DEC Fortran 3.2 running under Ultrix 4.2.

Changes:
(1) "implicit undefined" is a Sun'ism.  The equivalent for DEC is
    "implicit none".  However, the best strategy is not to have such a
    statement at all.  Both Sun & DEC Fortran accept the -u option which
    turns on this option without having to modify the source.  This is
    the preferred way for portable code.  This affects ALL fortran routines.
(2) DEC Fortran (and the F77 standard) do not allow null character strings.
    Since strings are padded with blanks anyway, the portable solution is
    to use ' ' instead of ''. (cli.F)
(3) Both Sun and DEC use the extension .F to signify that the file contains
    conditionals which need the C-preprocessor.
(4) Consider a file which is read only.  For Sun an OPEN succeeds on this
    file, and only if one attempts to write will an error be signalled.
    For DEC, an open fails unless the "readonly" option is specified in
    the open.  The readonly option is ignored by Sun, but with a warning.
    I have put conditionals (ifdef ultrix) around such changes.
(5) The interpretation of the record size for unformatted reads is according
    to the F77 standard in terms of "processor defined units".  Sun uses
    bytes and DEC uses longwords (4 bytes).  I have put conditionals around
    such changes. (iodisk.F)
(6) The routine celp.f (renamed to celp.F) had a number of unexplained
    errors.  First the header had been mangled so that comment characters
    were not in place on some statements.  Second, the reference to
    bitprotect treated it as a character, when in fact it is an integer.
    Third the only routine available is postfilt2, yet the code calls out
    postfilt.
(7) There are some incompatibilities for an unformatted read of a character
    string in routine cli.F.  The F77 standard indicates that on the
    external record, the string should be in quotes.  I changed this to a
    formatted read.
(8) The code in psearch.F had some consistency checks on parameter values.
    With DEC Fortran, these were optimized right out of the code.  This
    made part of the code unreachable, causing a compilation error.

Here are my diffs (not including the "implicit undefined" stuff which applies
practically to all files).

*** celp32/celp.f	Mon Oct 26 17:19:36 1992
--- ../celp-1016/celp32/celp.F	Wed Mar 10 17:32:26 1993
***************
*** 7,13 ****
  *	resulting from this code.  Further reproduction or distribution   *
  *	of this code without prior written permission of the U.S.         *
  *	Government is prohibited.  See NCS Technical Information Bulletin *
! *	on FS 1016 CELP Speech Coding.					  *                                                                         *									  *
  ***************************************************************************
  C
  C ROUTINE
--- 7,14 ----
  *	resulting from this code.  Further reproduction or distribution   *
  *	of this code without prior written permission of the U.S.         *
  *	Government is prohibited.  See NCS Technical Information Bulletin *
! *	on FS 1016 CELP Speech Coding.                                    *
! *                                                                         *
  ***************************************************************************
  C
  C ROUTINE
***************
*** 146,152 ****
  C**************************************************************************
  C*-
          program celp
! 	implicit undefined (a-z)
  	include 'ccsub.com'
  convex	#include "ccsub.com"
          real sold(maxll), snew(maxll), ssub(maxll)
--- 147,153 ----
  C**************************************************************************
  C*-
          program celp
! 
  	include 'ccsub.com'
  convex	#include "ccsub.com"
          real sold(maxll), snew(maxll), ssub(maxll)
***************
*** 285,292 ****
--- 286,295 ----
  c	This issue can be reviewed by reading the UNIX man pages
  c	on ieee floating point (man -k ieee).
  c
+ #ifdef sun
  	call abrupt_underflow()
  #include <f77/f77_floatingpoint.h>
+ #endif
  cSUN?	call ieee_handler("set", "underflow", SIGFPE_DEFAULT)
  c
  c	*** parse the command line
***************
*** 377,385 ****
  c
  c			*bit stream channel file
  	if (analy) then
! 	   open(unit=25, file=ofile(1:oflen)//'.chan')
  	else
  	   open(unit=25, file=ifile(1:iflen)//'.chan', status='old')
  	end if
  c
  c			*quantizer design (data collection) files
--- 380,393 ----
  c
  c			*bit stream channel file
  	if (analy) then
! 	   open(unit=25, file=ofile(1:oflen)//'.chan', status='new')
  	else
+ #ifdef ultrix
+ 	   open(unit=25, file=ifile(1:iflen)//'.chan', status='old',
+      -	        readonly)
+ #else
  	   open(unit=25, file=ifile(1:iflen)//'.chan', status='old')
+ #endif
  	end if
  c
  c			*quantizer design (data collection) files
***************
*** 387,393 ****
--- 395,406 ----
  copt	open(unit=21, file='pgain.data')
  c
  c			*read stochastic code book vector file (by 20)
+ #ifdef ultrix
+ 	open(unit=11, file='./codebook.h', status='old', err=112,
+      -	     readonly)
+ #else
  	open(unit=11, file='./codebook.h', status='old', err=112)
+ #endif
  	i = 1
  	do 110 j = 1, (2*(maxncsize-1)+l)/20
  	   read(11, *, err=112) (x(k), k=i,i-1+20)
***************
*** 400,406 ****
--- 413,424 ----
  115	continue
  c
  c			*read adaptive code book index (pitch delay) file
+ #ifdef ultrix
+ 	open(unit=12, file='./pdelay.h', status='old', err=122,
+      -	     readonly)
+ #else
  	open(unit=12, file='./pdelay.h', status='old', err=122)
+ #endif
  	do 120 i = 0, maxpd-1
  	   read(12, *, err=122) pdelay(i)
  120	continue
***************
*** 412,419 ****
--- 430,444 ----
  c			*read pitch delay coding tables for bit assignment
  c			*pdencode.h for encoding, pddecode.h for decoding
  c			*generate pdtabi for delta delay coding
+ #ifdef ultrix
+ 	open(unit=13, file='./pdencode.h', status='old', err=132,
+      -	     readonly)
+ 	open(unit=14, file='./pddecode.h', status='old', err=142,
+      -	     readonly)
+ #else
  	open(unit=13, file='./pdencode.h', status='old', err=132)
  	open(unit=14, file='./pddecode.h', status='old', err=142)
+ #endif
  	do 130 i = 0, maxpd-1
  	   read(13, '(z)', err=132) pdencode(i)
  	   read(14, *, err=142) pddecode(i)
***************
*** 427,441 ****
  145	continue
  c
  c			*read bit protection file
  	open(unit=15, file='./bitprotect.h', status='old', err=152)
! 	   read(15, *, err=152) bitprotect(1:codelength2)
  	close(15)
  	go to 155
  152	   stop ' celp:  Problem with file "bitprotect.h"'
  155	continue
  c
  c			*read bit permutation file (by 12)
  	open(unit=16, file='./bitpermute.h', status='old', err=162)
  	i = 1
  	do 160 j = 1, streambits/12
  	   read(16, *, err=162) (bitpermute(k), k=i,i+11)
--- 452,476 ----
  145	continue
  c
  c			*read bit protection file
+ #ifdef ultrix
+ 	open(unit=15, file='./bitprotect.h', status='old', err=152,
+      -	     readonly)
+ #else
  	open(unit=15, file='./bitprotect.h', status='old', err=152)
! #endif
! 	   read(15, *, err=152) (bitprotect(i),i=1,codelength2)
  	close(15)
  	go to 155
  152	   stop ' celp:  Problem with file "bitprotect.h"'
  155	continue
  c
  c			*read bit permutation file (by 12)
+ #ifdef ultrix
+ 	open(unit=16, file='./bitpermute.h', status='old', err=162,
+      -	     readonly)
+ #else
  	open(unit=16, file='./bitpermute.h', status='old', err=162)
+ #endif
  	i = 1
  	do 160 j = 1, streambits/12
  	   read(16, *, err=162) (bitpermute(k), k=i,i+11)
***************
*** 916,922 ****
  c
  c	      *** post filtering