Support ADPCM_MS in unknown container

[Dennis]

Summary of the bug:
FFmpeg does not recognize ADPCM_MS audio file when audio header is not present.
How to reproduce:

ffmpeg -i radio-orig.wav
ffmpeg version N-48065-g6375318 Copyright (c) 2000-2012 the FFmpeg developers
  built on Dec 21 2012 09:13:20 with gcc 4.7.2 (GCC)
  configuration: --enable-static --arch=x86_64 --target-os=mingw32 --cross-prefi
x=/home/rdp/dev/ffmpeg-windows-build-helpers/sandbox/mingw-w64-x86_64/bin/x86_64
-w64-mingw32- --pkg-config=pkg-config --enable-gpl --enable-libx264 --enable-avi
synth --enable-libxvid --enable-libmp3lame --enable-version3 --enable-zlib --ena
ble-librtmp --enable-libvorbis --enable-libtheora --enable-libspeex --enable-lib
openjpeg --enable-gnutls --enable-libgsm --enable-libfreetype --enable-fontconfi
g --enable-libass --enable-libutvideo --enable-libopus --disable-w32threads --en
able-frei0r --enable-filter=frei0r --enable-libvo-aacenc --enable-bzlib --enable
-libxavs --extra-cflags=-DPTW32_STATIC_LIB --enable-libopencore-amrnb --enable-l
ibopencore-amrwb --enable-libvo-amrwbenc --enable-libschroedinger --enable-libbl
uray --enable-libvpx --enable-runtime-cpudetect
  libavutil      52. 12.100 / 52. 12.100
  libavcodec     54. 81.100 / 54. 81.100
  libavformat    54. 49.102 / 54. 49.102
  libavdevice    54.  3.102 / 54.  3.102
  libavfilter     3. 29.100 /  3. 29.100
  libswscale      2.  1.103 /  2.  1.103
  libswresample   0. 17.102 /  0. 17.102
  libpostproc    52.  2.100 / 52.  2.100
radio-orig.wav: Invalid data found when processing input

ffmpeg -i radio-header.wav
ffmpeg version N-48065-g6375318 Copyright (c) 2000-2012 the FFmpeg developers
  built on Dec 21 2012 09:13:20 with gcc 4.7.2 (GCC)
  configuration: --enable-static --arch=x86_64 --target-os=mingw32 --cross-prefi
x=/home/rdp/dev/ffmpeg-windows-build-helpers/sandbox/mingw-w64-x86_64/bin/x86_64
-w64-mingw32- --pkg-config=pkg-config --enable-gpl --enable-libx264 --enable-avi
synth --enable-libxvid --enable-libmp3lame --enable-version3 --enable-zlib --ena
ble-librtmp --enable-libvorbis --enable-libtheora --enable-libspeex --enable-lib
openjpeg --enable-gnutls --enable-libgsm --enable-libfreetype --enable-fontconfi
g --enable-libass --enable-libutvideo --enable-libopus --disable-w32threads --en
able-frei0r --enable-filter=frei0r --enable-libvo-aacenc --enable-bzlib --enable
-libxavs --extra-cflags=-DPTW32_STATIC_LIB --enable-libopencore-amrnb --enable-l
ibopencore-amrwb --enable-libvo-amrwbenc --enable-libschroedinger --enable-libbl
uray --enable-libvpx --enable-runtime-cpudetect
  libavutil      52. 12.100 / 52. 12.100
  libavcodec     54. 81.100 / 54. 81.100
  libavformat    54. 49.102 / 54. 49.102
  libavdevice    54.  3.102 / 54.  3.102
  libavfilter     3. 29.100 /  3. 29.100
  libswscale      2.  1.103 /  2.  1.103
  libswresample   0. 17.102 /  0. 17.102
  libpostproc    52.  2.100 / 52.  2.100
[wav @ 000000000225dfe0] max_analyze_duration 5000000 reached at 5002667
Guessed Channel Layout for  Input Stream #0.0 : stereo
Input #0, wav, from 'radio-header.wav':
  Duration: 24:51:18.49, bitrate: 0 kb/s
    Stream #0:0: Audio: adpcm_ms ([2][0][0][0] / 0x0002), 48000 Hz, stereo, s16,
 384 kb/s
At least one output file must be specified

[Dennis]

Original file:
​http://www.djxmmx.net/ffmpeg/2064/radio-orig.wav

File with ADPCM_MS header manually added:
​http://www.djxmmx.net/ffmpeg/2064/radio-header.wav

[Carl Eugen Hoyos]

How was radio-orig.wav produced? Which software plays radio-orig.wav ?

[Dennis]

I do not know how it was produced.
It plays in a video game: Call of Duty Black Ops.
I don't know if they removed the header to save space (doubtful) or for obfuscation purposes.

[Carl Eugen Hoyos]

Do I understand you correctly that the file is found as-is on your drive if Call of Duty Black Ops is installed / that the exact file can be found on the DVD?

[Dennis]

Yes.

The files are inside an archive system, which when read through, make up the file system for the game. You can extract the files with 7-Zip and obtain many more samples. This is most likely an obfuscation attempt.

[Carl Eugen Hoyos]

Could you compare the first 64 bytes of a few samples (or provide them)?
At least the sample-rate is stored in a short header afaict, it may be possible to autodetect the format.

[Dennis]

I will leave it to better men to investigate properly.
For more samples, please feel free to browse the directory of:
​http://www.djxmmx.net/temp/FFmpeg-ADPCM_MS/

All original/unmodified samples from the game.
The only file in which I manually added a header is: ​http://www.djxmmx.net/temp/FFmpeg-ADPCM_MS/radio-header.wav

I can add more samples to that directory if you need.
Just let me know.

Thank you

[Elon Musk]

This bug report is pure nonsense. There is no way to support something that can be anything...

If you want support for raw generic (de)muxer where you manually specify audio codec parameters please open new bug report.

[Carl Eugen Hoyos]

It looks valid to me...

[Carl Eugen Hoyos]

Did you look at the files? They do not look like raw audio to me...

[Elon Musk]

They are not raw audio, they are headerless -raw - adpcm compressed

[Carl Eugen Hoyos]

What do you mean with headerless?
Afaict, the files contain a header that at least shows the number of channels and the samplerate in the file (afaiu, this is essential for decoding).
Or do I misunderstand?

[Elon Musk]

Even if you assumption is correct there is still no way to automagically conclude that it have adpcm.

[Carl Eugen Hoyos]

Maybe "30 08" tells us or the file format only supports adpcm_ms?
Since so far we only have adpcm_ms samples, the assumption looks not dangerous to me in any way.

[Elon Musk]

Only 2 bytes? That is hardly useful. How would you name your new demuxer? 3008?

[Carl Eugen Hoyos]

wav also uses two bytes to identify the codec...

Or do you mean the probing of the file?
Apart from the fact that demuxers without probe function exist, the header is quite long, so it should be possible to write a probe function that is at least not less strict than other existing probe functions.

[Elon Musk]

How long is that header? Can you show here header for all files?

[Carl Eugen Hoyos]

The header is 2096 bytes afaict, mostly filled with "0".

[Carl Eugen Hoyos]

Implemented by Michael, if any additional information is available like the actual file format name or if other codecs can be used, please share it!