[FFmpeg-user] Erroneous volume level shifts using ffmpeg to convert AC3 in M2TS to WAV

John Pilgrim lists at johnpilgrim.net
Sun Mar 1 19:15:05 CET 2015

Thank you Andy!
ffmpeg -drc_scale 0 -i foo.m2ts foo.wav
worked perfectly and the loudness measurements of the AC3 and WAV extractions now match, as they should.
To wit, http://johnpilgrimdesign.com/client/loudness/SCowR25LKFS/#lifeofsock
Thanks again,

On Feb 28, 2015, at 3:44 PM, Andy Furniss <adf.lists at gmail.com> wrote:

> John Pilgrim wrote:
>> I using ffmpeg to extract audio from non-commercial bluray m2ts video
>> files, for subsequent loudness analysis in AudioLeak or Dolby Media
>> Meter. The m2ts sometimes have AC3 audio and sometimes have linear
>> PCM audio. I wish to output a WAV file. The command I am using is
>> ffmpeg -i foo.m2ts foo.wav The volume levels of the WAV files are not
>> accurate compared to the corresponding AC3 files.
>> Here's what I mean: If I extract the ac3 audio instead, using
>> ffmpeg -i foo.m2ts -acodec copy -f ac3 foo.ac3 and run both through
>> Dolby Media Meter, the loudness and peak levels of the WAV are
>> shifted up/or down compared to the corresponding AC3 file.
>> And this is happening with AC3 files with a DIALNORM metadata value
>> of -31, which per the Dolby specs, should result in NO volume level
>> change, as -31 is the reference point.
>> Does anyone have any insight into what's going on, or recommendations
>> for a better invocation of ffmpeg?
> By default ac3dec will apply full DRC if there is any there it doesn't
> use dialnorm by default.
> To get full range DRC wise do
> ffmpeg -drc_scale 0 -i foo.m2ts
> WRT dialnorm - I am not familiar with Dolby s/w, but I think it's
> possible that some decoders would scale up to a target higher than -31
> depending on how they are set - so just because the stream is -31 I am
> not sure that excludes the possibility that a decoder will adjust -
> target level is a decoder setting.
> _______________________________________________

More information about the ffmpeg-user mailing list