[FFmpeg-devel] [PATCH 2/2] avformat/riffdec: follow the MS docs more strictly for setting wav channel layouts

[Tobias Rapp]

On 21/03/2024 10:40, Marton Balint wrote:

>
>
> On Thu, 21 Mar 2024, Tobias Rapp wrote:
>
>> On 19/03/2024 20:14, Marton Balint wrote:
>>
>>>
>>>
>>>  On Tue, 19 Mar 2024, Michael Niedermayer wrote:
>>>
>>>>  On Sun, Mar 17, 2024 at 08:57:29PM +0100, Marton Balint wrote:
>>>>>  - Only parse the defined masks in dwChannelMask, unless
>>>>>  strict_std_compliance
>>>>>    is less than normal. This matches with the behaviour of the wav 
>>>>> muxer.
>>>>>  - Ignore additional bits in dwChannelMasks as the MS documentation
>>>>>  suggests [1]
>>>>>  - Assume UNKNOWN channels for missing bits as the MS documentation
>>>>>  suggests [1]
>>>>>
>>>>>  [1]
>>>>>  https://learn.microsoft.com/en-us/previous-versions/windows/hardware/design/dn653308(v=vs.85)#details-about-dwchannelmask 
>>>>>
>>>>>
>>>>>  Signed-off-by: Marton Balint <cus at passwd.hu>
>>>>>  ---
>>>>>   libavformat/riffdec.c | 28 +++++++++++++++++++++++++---
>>>>>   1 file changed, 25 insertions(+), 3 deletions(-)
>>>>
>>>>  breaks:
>>>>  ./ffmpeg  -i ~/tickets/2859/5.1plusdownmix.wav -ac 2 -t 100 -bitexact
>>>>  -c:a aac -y /tmp/2859-frenchspeack-nolibfaac.mp4
>>>
>>>  After the patch this is file will need -strict unofficial to work, 
>>> since
>>>  the downmix channels are not officially recognized in the 
>>> dwChannelMask.
>>>
>> I think downmix channels are part of the RF64 specification, see EBU 
>> Tech 3306 section 3.1:
>>
>> """
>> 3.1 Enhancement for a PCM stereo down mix
>>
>> No PCM stereo signal is included in the basic Wave Format Extensible.
>>
>> To include a stereo channel the following is added:
>>
>> #define SPEAKER_STEREO_LEFT    0x20000000
>> #define SPEAKER_STEREO_RIGHT    0x40000000
>> """
>>
>
> I was reluctant to add these, because the recommendation which 
> superseded this, ITU BS.2088 does not mention these masks.
>
> Nevertheless, you are right, these should be recognized to support 
> historical RF64 files. Or we should not even make the distinction of 
> RF64?
>
WAV files smaller than 2/4GiB could start with the usual "RIFF" bytes 
but still contain these mask bits. From a quick glance at ITU BS.2088 it 
seems it could use the "chna" chunk to achieve a mapping like "5.1 + 
downmix".

Having said that, I have not stumbled over "X plus downmix" WAVE files 
in production yet. So no strong opinion from my side on whether this 
should work without "-strict unofficial", or not.

Regards, Tobias