Hi guys,

Yes, it sounds like a clock mismatch: alsa is slower than your
icecast. You need a buffer of some sort. With a normal buffer,
assuming the clock on your computer is pretty much the same as
icecast's clock, you'll still get the drops but "inside liquidsoap" so
you can decide how to cope with them.

For instance you could write
output.icecast(...,fallback([buffer(radio),default])) where radio
contains the alsa stuff, and is thus controlled by the alsa clock, and
default (and the icecast output) are free of alsa and controlled by
the default wallclock. If radio is too slow the fallback will notice a
failure and move on to default for a while.

With the adaptative buffer, the data from radio will be
stretched/compressed (there are parameters to control how much) to
avoid the failure. Obviously, the cost is audio distortion.

Now, I wanted to comment on what liquidsoap may be doing differently
from another tool. In general, we keep track of time more precisely
than other tools, e.g. ices2 for sure. Reading your log again, it
seems that you're already using a buffer, which clearly makes your
setup different from what another tool would do. One thing to try,
then, is to remove buffers completely: then the icecast output will be
alsa-controlled. Liquidsoap won't complain about clock issues, and
streaming will drop if it has too, ie. if icecast kicks you because
you're too slow, which might be less often than currently. But it's
just hiding the problem, pushing it at the level of icecast's buffers.

I hope I'm not proposing too many things to try :)

Cheers,

David

PS: You could also try passing clock_safe=false to your ALSA I/O. It
tells the operators to not use the ALSA clock, but the default
wallclock instead. This will likely result in drops at the source,
though.

------------------------------------------------------------------------------
_______________________________________________
Savonet-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/savonet-users

Reply via email to