Hi John,
Thanks for this report. I actually pushed the changes to decode and pass
metadata from ffmpeg frames when converting back to liquidsoap frame format
in this PR: https://github.com/savonet/liquidsoap/pull/1812
In it, I failed to notice your use case and assumed that we could drop
metadata when converting from liquidsoap frame format to ffmpeg's. I'll be
fixing this shortly.
Romain
Le sam. 21 août 2021 à 13:37, John Warburton <j...@johnwarburton.net> a
écrit :
> As I convert to Liquidsoap 2.0.0, I might be losing metadata through an
>> FFmpeg filter.
>> ...
>> Am I losing it because the conversion to 'raw' discards the metadata,
>> leaving only the audio?
>>
>
> This workaround is doing the job -- but is it correct?
>
> I define a little function to read metadata before the audio is sent
> through ffmpeg.raw.encode.audio, and then insert that metadata directly
> into the Icecast stream using icy.update_metadata
>
> def handle_metadata(m) =
> t = m["title"]
> a = m["artist"]
> icy.update_metadata(host="127.0.0.1", port=8000, password="[PASSWORD]",
> mount="test.aac", [("title", t), ("artist", a)])
> end
>
> Then, the handle_metadata() function is called before track boundaries
> disappear i.e. before any crossfade function.
> radio.on_track(handle_metadata)
>
> Importantly, in the output.icecast call, I switch icy_metadata="false"
> because otherwise the "title" string changes to "Unknown" sometimes. Don't
> know why.
>
> Now, what I'd really like to do is delay the metadata by a few seconds
> because the audio processing involves a 15-second gain detection window.
>
> But that's next.
> JW
>
>>
>> Here is the audio processor, followed by a wrapper using the
>> ffmpeg.raw.encode.audio function to enter the processing, and the
>> ffmpeg.raw.decode.audio function to leave the processing.
>>
>> def audio_process(s) =
>> def mkfilter(graph) =
>> s = ffmpeg.filter.audio.input(graph, s)
>> s = ffmpeg.filter.dynaudnorm(graph, s, gausssize=25, correctdc=true,
>> altboundary=true, maxgain=12., targetrms=1.)
>> s = ffmpeg.filter.aresample(graph, s, sample_rate=192000)
>> s = ffmpeg.filter.alimiter(graph, s, limit=0.95, attack=5.,
>> release=80., asc=true, asc_level=1.)
>> s = ffmpeg.filter.aresample(graph, s, sample_rate=48000)
>> ffmpeg.filter.audio.output(graph, s)
>> end
>> ffmpeg.filter.create(mkfilter)
>> end
>>
>> def audio_process_wrap(s) =
>> a = ffmpeg.raw.encode.audio(%ffmpeg(%audio.raw), s)
>> a = audio_process(a)
>> ffmpeg.raw.decode.audio(a)
>> end
>>
>> with best wishes,
>> John
>>
>
>
> _______________________________________________
> Savonet-users mailing list
> Savonet-users@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/savonet-users
>
_______________________________________________
Savonet-users mailing list
Savonet-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/savonet-users
