Audio streaming problems

[Joel Lord]

Thank you, Michiel, you got it in one.

aacparse converts an aac stream of audio to "framed", and clearly that 
was what mpegtsmux needed.  Now that you've pointed that out I see it in 
the documentation, but that was a subtle difference I was missing.

The video part works fine, I guess I'm stuck with a static image I 
didn't really want.  Worse things have happened.

-Joel

On 12/19/2022 3:41 AM, Michiel Konstapel wrote:
> On Monday 19 December 2022 03:38:38 (+01:00), Joel Lord via 
> gstreamer-devel wrote:
> 
>  > I'm trying to take an audio feed from a microphone connected via a 
> USB interface and (ideally) stream it directly out via HTTP. I want 
> multiple end users to be able to get to the stream and listen to it, 
> real time, from their phones. I haven't had any luck finding components 
> that add up to that, but I did find how to produce a mpeg4ts stream, so 
> I added a static image and have that streaming successfully, I added in 
> a clock to have something changing to prove that it was working. But the 
> audio never comes through.
>  >
>  > If I take the mpegtsmux and hlssink off the end of my pipeline and 
> replace it with a filesink I can prove that my audio is working fine, 
> but when I feed it to mpegtsmux and hlssink it seems to get ignored. If 
> I have no image it never creates a second segment or the playlist for 
> the stream, so I can't see anything at all. If I include the image the 
> stream works but it seems to have no audio. I've tried 3 or 4 different 
> audio formats, it doesn't complain that I have a format mismatch and it 
> seems to be working, but nothing comes through.
>  >
>  > Using gstreamer1.0 version 1.18.4 on a Raspberry Pi 4.
>  >
>  > gst-launch-1.0 -v \
>  > mpegtsmux name=mux ! hlssink 
> playlist-root=http://stream.ek:80/stream_files 
> location=/var/www/stream/stream_files/segment%05d.ts target-duration=3 
> playlist-location=/var/www/stream/playlist.m3u8 \
>  > filesrc location=/home/pi/EKPB.png ! decodebin ! \
>  > videoconvert ! video/x-raw,format=I420 ! imagefreeze ! \
>  > clockoverlay halignment=right ! x264enc ! mux. \
>  > alsasrc ! audioconvert ! audio/x-raw,channels=1 ! \
>  > level ! avenc_mp2fixed ! queue max-size-buffers=0 max-size-bytes=0 
> max-size-time=1000000000 ! mux.
>  >
>  > If I take the queue out of the audio path it tosses a few warnings 
> about dropping samples (a lot of samples) and gives me no faith that it 
> will work. Changing the buffer settings on the queue made no difference.
>  >
>  > So for anyone who has made it this far, I'm quite open to better 
> solutions than I've found OR fixing the one that I have that isn't 
> working. I don't actually want anything but the audio feed, and I want 
> it as close to 0 latency as I can get.
>  >
>  > Thanks for your help!
>  >
> 
> Interesting! hlssink expects video key frames to know when it can start 
> a new segment (which has to start with a key frame) so I don't know what 
> happens with an audio only stream. At least for getting things working, 
> it's probably going to be easiest to indeed have a "dummy" video track 
> in there.
> 
> I use the following for HLS audio:
> 
> ... ! avenc_aac ! aacparse ! mpegtsmux
> 
> Maybe give that a go? The aacparse might be adding information needed by 
> the muxer or the sink.
> 
> For x264enc you may need to specify a keyframe interval 
> (key-int-max=<number of frames>) but hlssink will actively ask its 
> upstream for key frames so that might not be required. Maybe also 
> specify a framerate in your video caps, I don't know what the default 
> is. After x264enc I also have a parser:
> 
> ... ! x264enc ! h264parse config-interval=-1 ! ...
> 
> HTH,
> Michiel

-- 
Joel Lord