readme.doc

基本定点运算程序库

README.TXT for AMR speech encoder and decoder     
===============================================================================

       GSM AMR-NB speech codec   R98   Version 7.6.0   December 12, 2001
                                 R99   Version 3.3.0                
                                 REL-4 Version 4.1.0                


Installation
------------

  Assuming all delivered files are in the same directory, issuing "make"
  or "make VAD=VAD1" on a Unix system compiles the source code and
  builds two standalone programs called "encoder" and "decoder" which
  use the ENS VAD option. To compile with the Motorola VAD, use
  "make VAD=VAD2"
  
  To compile with weighted MOPS count enabled use
  "make [VAD=VAD#] MODE=WMOPS".
  
  To verify the correct installation the files spch_unx.inp,
  spch_unx.cod, and spch_unx.out are provided for the ENS VAD option and
  the files spch_un2.cod and spch_un2.out for the Motorola VAD option.
  These files contain data in MSBbyte first order. The corresponding
  files spch_dos.* and spch_do2.* contain the same data in LSByte first
  order. A mode control file (allmodes.txt) is also supplied.
  
  To verify the encoder, use
  
     encoder -dtx -modefile=allmodes.txt spch_unx.inp tmp.cod
  or
     encoder -dtx -modefile=allmodes.txt spch_un2.inp tmp.cod

  for the ENS or the Motorola VAD option, respectively, and compare the
  output tmp.cod to spch_unx.cod or spch_un2.cod.
  
  To verify the decoder, use
  
     decoder spch_un2.cod tmp.out
  or
     decoder spch_unx.cod tmp.out

  for the ENS or the Motorola VAD option, respectively, and compare the
  output tmp.out to spch_unx.out or spch_un2.out.

  On LSByte first systems, spch_dos.*/spch_do2.* must be used instead of
  spch_unx.*/spch_un2.*
  
  A shell script "amr_chk.csh" is provided for Unix systems which
  performs these tests automatically. The command line syntax is as
  follows:

    amr_chk.csh [-vad2] unix

  for MSB first systems (such as SUN workstations), or

    amr_chk.csh [-vad2] dos

  for LSB first systems (such as Linux on a PC).
  By default, it is assumed that the executable was compiled with
  the ENS VAD option. To check the Motorola VAD version, the option
  "-vad2" must be given on the command line.

  Note that this verification procedure only performs a very basic
  installation verification and is not a proof of overall bitexact
  operation.


Command line options for encoder program
----------------------------------------

  The general syntax is
  
     encoder [options] amr_mode            input_file bitstream_file
  or
     encoder [options] -modefile=mode_file input_file bitstream_file

  In the first case, amr_mode must be one of MR475, MR515, MR59, MR67,
  MR74, MR795, MR102, or MR122. In the second case, the text file
  mode_file must contain the mode names to be used (one line per
  speech frame, each line containing one of the mode names mentioned
  above). An example file iterating trhough all mode names is supplied
  (allmodes.txt).
  The following options are recognized:
  
     -dtx  enables DTX operation



Command line options for decoder program
----------------------------------------

  The general syntax is
  
     decoder [option] bitstream_file output_file
  
  The following option is recognized:
  
     -rxframetype      expect RX frame type instead of TX frame type in
                       input file. This only useful for simulations with
                       other components between encoder and decoder.



File formats
------------

  The encoder input and decoder output files contain are 8 kHz, 16
  bit/sample data.  The output file length is always a multiple of 320
  bytes because of the frame length of 160 samples. The encoder will
  only process whole input frames, i.e. for input files whose length
  is not a multiple of 160 samples the last <size> - n*160 samples are
  not processed.
  
  The encoder output/decoder input files ("bitstream files") contain
  frames which are structured as follows:
  
     1 word     frame type
   244 words    encoded speech parameter bitstream (one bit per word,
                each word contains either 0x0001 or 0x0000), unused
                bits written as 0x0000 for modes < MR122
     1 word     mode information
     4 words    unused (written as 0x0000 by encoder)
  
  The frame type written by the encoder (TX frame type) can have one
  of the following values:
  
     0x0000  TX_SPEECH_GOOD (normal speech frame)
     0x0001  TX_SID_FIRST   (first SID frame after speech)
     0x0002  TX_SID_UPDATE  (SID update frame in DTX)
     0x0003  TX_NO_DATA     (empty frame in DTX)
  
  If invoked without the -dtx option, the encoder only produces type
  0x0000 (TX_SPEECH) frames.
  
  The decoder normally accepts the output format and frame type
  produced by the encoder. However, using the option -rxframetype, it
  can be forced to expect frames with RX frame type instead of TX
  frame type. In that case, it can NOT directly read files produced by
  the encoder because the frame types do not match. The RX frame type
  coding is as shown below:
  
     0x0000  RX_SPEECH_GOOD               (normal speech frame)
     0x0001  RX_SPEECH_DEGRADED           (degraded speech frame)
     0x0002  RX_ONSET                     (ONSET frame)
     0x0003  RX_SPEECH_BAD                (bad speech frame)
     0x0004  RX_SID_FIRST                 (first SID frame after speech)
     0x0005  RX_SID_UPDATE                (SID update frame in DTX)
     0x0006  RX_SID_BAD                   (corrupted SID frame)
     0x0007  RX_NO_DATA                   (no data received)
  
  The coding of the mode information field in the bitstream files is
  as follows:
  
     0x0000  MR475   4.75 kbit/s
     0x0001  MR515   5.15 kbit/s
     0x0002  MR59    5.90 kbit/s
     0x0003  MR67    6.70 kbit/s
     0x0004  MR74    7.40 kbit/s
     0x0005  MR795   7.95 kbit/s
     0x0006  MR102  10.20 kbit/s
     0x0007  MR122  12.20 kbit/s


Motorola VAD
------------

Summary:    Version 2.0.1 incorporates the Voice Activity Detector option 2.
            Either option 1 or option 2 can be compiled via the respective
            command "make VAD=VAD1" or "make VAD=VAD2".  This package has
            been verified to be backward compatible with version 2.0.0.  That
            is, version 2.0.0 encoder is bit-exact to version 2.0.1 when
            compiled with the VAD=VAD1 argument.  It has also been verified
            that version 2.0.1 compiled with the VAD=VAD1 argument is
            bit-exact to version 2.0.1 compiled with the VAD=VAD2 argument
            when run in non-DTX mode.  Only the vad_flag output of the vad2()
            function will affect the subsequent state of the encoder process.
            No other outputs are generated, nor are any encoder state
            variables directly modified by the use of the VAD2 option.

            The co-existence of the two VAD options is achieved through the
            use of conditional compiler directives.  That is, for every
            instance where the existing code has been modified, there is
            a corresponding "#ifdef VAD2" where code has been added, and a
            corresponding "#ifndef VAD2" where code related to the VAD1
            option is to be omitted.  It should be noted that the conditional
            compile was purposefully triggered from a single define to avoid,
            for example, the cases where both (or neither) a VAD1 and/or VAD2
            compiler directives were defined.

            Furthermore, the integration impact of the VAD2 option was kept
            to a minimum by allowing the typedef for the vadState structure
            to be dependent on the compiler directive.  This was accomplished
            by "typedef-ing" vadState1 for VAD1, and vadState2 for VAD2, and
            then further defining the typedef to be used as:

                #ifndef VAD2
                #define vadState vadState1
                #else
                #define vadState vadState2
                #endif

            The result is that very few modules are affected by the change
            because the functions passing the state pointers do not care
            how the structure is defined.  Only the low level routines which
            use the specific elements of the state structures need to care
            about the elements of the vadState structure.


C code changes since 7.3.0
--------------------------

One change request on the AMR C-code reference was approved by the joint
SMG11/TSG-SA4 meeting #14 / #9 in Puerto Vallarta, Mexico, January 24-28 2000.

CR ID:
   A021

Subject of change:
   Avoidance of pulse cancellation in FCB excitation

Files affected:
   C code files c2_11pf.c, d2_11pf.c

For details of the changes, see the CR.


C code changes since 7.4.0
--------------------------

CR ID: 
   A023
   
Subject of change:
   Correction of potential bug in AMR decoder due to the usage of standard
   C abs() function.

File affected:
   dtx_dec.c

For details of the changes, see the CR.


CR ID: 
   A024
   
Subject of change:
   Correction of comfort noise parameter interpolation bug of AMR decoder

File affected:
   dtx_dec.c 

For details of the changes, see the CR

CR ID: 
   A025
   
Subject of change:
   Correction of mode state bug in AMR decoder

File affected:
   decoder.c, sp_dec.c, sp_dec.h, coder.c

For details of the changes, see the CR

CR ID: 
   A026 rev 1
   
Subject of change:
   Correction of the TX_Type and RX_Type identifiers

File affected:
   frame.h, sid_sync.c, decoder.c, dec_amr.c, dtx_dec.c, strfunc.c 

For details of the changes, see the CR

CR ID: 
   A027 
   
Subject of change:
   Correction of potential bug in AMR due to the usage of standard
   C abs() function (VAD2).

File affected:
   vad2.c

For details of the changes, see the CR


C code changes since 7.5.0
--------------------------

CR ID: 
   A028
   
Subject of change:
   Correction of RX-DTX handling of NO_DATA frames in AMR decoder

File affected:
   dtx_dec.c

For details of the changes, see the CR.

CR ID: 
   A029
   
Subject of change:
   Correction in AMR decoder to avoid division by zero in RX-DTX Handling

File affected:
   dtx_dec.c

For details of the changes, see the CR.


$Id $