dolby digital plus 7.1 defective decoding

[Alexander]

Greetings.
Summary of the bug:
I found out that the decoding of ddp 7.1 and ddp 7.1 joc is defective. It looks like the audio data frames from one channel are dropping out into the other channels. It happens in random places. This can be clearly seen in the spectrogram, zooming in on a given defect location shows the absence of signal in one channel and the presence of it in another. It can be clearly audible as clicks on separate channels, but may not be audible when listening on the player in stereo downmix. The defect appears on all versions of ffmpeg with libavcodec decoder higher than 59.37.100.
How to reproduce:

ffmpeg-master-latest-win64-gpl\bin>ffmpeg.exe -drc_scale 0 -i "sample.eac3" -c pcm_s24le "sample.wav"
ffmpeg version N-117868-gecc7d5db9c-20241123
libavcodec     61. 25.102 / 61. 25.102

How to reproduce:

ffmpeg-n7.1-latest-win64-gpl-7.1\bin>ffmpeg.exe -drc_scale 0 -i "sample.eac3" -c pcm_s24le "sample_n7.1.wav"
ffmpeg version n7.1-36-g20c8a3f5ff-20241123
libavcodec 61. 19.100 / 61. 19.100

How to reproduce:

ffmpeg-n6.1-latest-win64-gpl-6.1\bin>ffmpeg.exe -drc_scale 0 -i "sample.eac3" -c pcm_s24le "sample_n6.1.wav"
ffmpeg version n6.1.2-15-ga6a07eebdf-20241123
libavcodec 60. 31.102 / 60. 31.102

How to produce correct results without defects:

ffmpeg-n5.1-latest-win64-lgpl-5.1\bin>ffmpeg.exe -drc_scale 0 -i "sample.eac3" -c pcm_s24le "sample_n5.1.wav"
ffmpeg version n5.1.6-11-gcde3c5fc0c-20241123
libavcodec 59. 37.100 / 59. 37.100

eac3 7.1 sample:

https://drive.google.com/file/d/1H78XIQCBHg9YIj8gpcXWJ4uAVqGvI0sz/view?usp=sharing

[Balling]

I see there is a difference, but most of it is just off-by-one, and otherwise almost not an issue.

Can you find the commit that regressed this?

Can you find a way to remove DRC metadata on the file, I need that. Thanks.

[Alexander]

Can you find the commit that regressed this?

Sorry, I am just a consumer, not a developer.

Can you find a way to remove DRC metadata on the file, I need that. Thanks.

If I understood you correctly, I made a new encode from the lossless original with metadata Line Mode Profile: None and RF Mode Profile: None. I left the DN the same.
I tried to decode it, but the problem remains the same.

https://drive.google.com/file/d/1NqHdIRc03qd-5AsalP83PLv4yAPuh2rz/view?usp=sharing

[Balling]

​https://patchwork.ffmpeg.org/project/ffmpeg/patch/20241124170703.26375-1-jamrial@gmail.com/

Patch

[James]

Should be fixed in 0e07a70611dc109f4e84895ffc79827a91edc045