[FFmpeg-devel] Handling dual language mono audio encoded as stereo

[Anssi Hannula]

Anssi Hannula kirjoitti lauantai, 17. hein?kuuta 2010 02:35:53:
> Michael Niedermayer kirjoitti lauantai, 20. helmikuuta 2010 03:00:20:
> > On Tue, Feb 16, 2010 at 12:35:02AM +0200, Anssi Hannula wrote:
> > > > On Sun, Feb 14, 2010 at 07:32:08PM +0200, Anssi Hannula wrote:
> > > > > Hi all!
> > > > > 
> > > > > Some nordic DVB channels encode e.g. four mono tracks with
> > > > > different languages into two stereo tracks (mpeg layer 2). The
> > > > > ISO639 language descriptor then has both language codes, separated
> > > > > by a null byte. This is probably a remnant from pre-DVB era, but
> > > > > we should somehow handle it nevertheless.
> > > 
> > > it ["null" byte above] is the audio type code of the first track,
> > > which is always present but is ignored by ffmpeg even without this
> > > patch. There's another audio type code for the second channel after the
> > > second ISO639 language code.
> > > 
> > > The possible values are:
> > > 00 Undefined
> > > 01 Clean effects
> > > 02 Hearing impaired
> > > 03 Visual impaired commentary
> > > 04-FF Reserved
> > > 
> > > I guess we should somehow add this in metadata. Then of course there's
> > > also the case where audio type differs between channels :)
> > 
> > this really is a mess :(
> > why do all these comitees always come up with doing everything in the
> > most painfull and backwardly hacked in way.
> > I guess simply setting the AVStream metadata to things like
> > Channel0/Language="eng"
> > Channel1/Language="jpn"
> > Channel0/Disposition="comment"
> > Channel0/TargetAudience="visually impaired"
> > 
> > is the least annoying way to handle this
> 
> Attached is a patchset that sets
> channel0/language="eng"
> channel1/language="jpn"
> channel1/audio_type="visual impaired"
> language="eng+jpn"
> 
> This time also mpegtsenc gets the support, and there is also a patch for
> ffplay to add '-ach' commandline option and 'c' key binding to switch
> audible audio channel, and a patch for ffmpeg that adds '-achannel' for
> selecting a single channel.
> 
> If e.g. re-encoding channel 1 of above example, the metadata will be
> mangled accordingly:
> language="jpn"
> audio_type="visual impaired"
> 
> Now, about the metadata format... I named the "audio_type" key as per specs
> as avformat.h does currently say "exported exactly as stored in the
> container", though I'm not really sure if it can be applied like this.
> Do you think we should directly mangle it to "target_audience" (and make it
> a generic name listed in avformat.h?) as "visually impaired"?
> If so, would "clean effects" translate to "disposition=clean effects"?

After more thought, I think it'd be enough just using generic tags for 
audio_type.

I guess those could be:
(channelN/)target_audience="visually impaired"
(channelN/)target_audience="hearing-impaired"
(channelN/)disposition="clean effects"

Do you agree?

Especially, is the use of "disposition" here correct? I just took it from your 
disposition="comment" suggestion above, dunno if it fits here like this.

Alternatively, we could just make the generic tag audio_type:
(channelN/)audio_type="visually impaired"
(channelN/)audio_type="hearing-impaired"
(channelN/)audio_type="clean effects"

Is this better?

> I dropped the "commentary" word, as there is no non-commentary "visual
> impaired" value, so broadcasters use the same value whether it is
> commentary or not.

-- 
Anssi Hannula