[FFmpeg-devel] [PATCH 2/2] lavfi/volume: support volume normalization through metadata.

[Jan Ehrhardt]

Nicolas George in gmane.comp.video.ffmpeg.devel (Fri, 1 Mar 2013
22:19:18 +0100):
>Le primidi 11 ventôse, an CCXXI, Nicolas George a écrit :
>> > Do you mean something like "volume=@{metadata_key}" or I'm still in the
>> > fog?
>> Yes, exactly.
>
>As a side note, this kind of method would also allow "half-normalization":
>if you want -23 and you have some parts very loud at -15 and some parts very
>quiet at -31, you may want to make the loud parts less loud and the quiet
>parts less quiet, but still keep the loud parts louder than the quiet parts;
>maybe make them -19 and -27 respectively (halfway). With expression
>evaluation, that would just be volume="(-23-@{loudness})/2".

The filter would be the most flexible if you could inject all
'loudnesses' into the metadata. Then you could things like:

volume=metadata=((lavfi.r128.I+lavfi.r128.S)/2)

By using the average of the Integrated loudness and the Short-term
loudness you also top off the the extremities.

Jan