Has someting been changed with merging sequenced sources in cross?

[JohanJNL]

For quite some time I've tried to improve crossfading in Liquidsoap. Although I'm not a programmer, I still realize that my solution is very ugly. But the result is quite amazing. My script checks every second during the last seconds of a music file between sequence or crossfade .I wrote the script it in Liquidsoap 2.1.3 and 2.1.4.

After upgrading to 2.2.4 it stopped working. I tried to diagnose the reason and I think it's because now all sequenced sources are merged in cross, even with sequence(merge=false, [a.source, b.source]).

I call the transition script in a while loop like this:

duration = ref(src)
while !duration > 0 do
src := cross(minimum=0., duration=float(!duration), width=(float(!duration) + 1.), transition, !src)
duration := !duration - 1
end

Because I cannot stop the while loop from the transition function, I have to stop the cross function in a different way, for example when the a.source is a jingle. I do that with sequence(merge=true, [a.source, b.source]). Since the upgrade to Liquidsoap 2.2.4 merge=true seems to happen always, or I I wrong?

I hope my issue is clear. I am not a native speaker. Anyway, thanks for offering Liquidsoap. I use it 24/7 on a stream and every Sunday for my broadcast via medium wave.

[gAlleb]

Hi!

As far as I can tell the !x notation for getting the value of a reference is now deprecated. You should write x() instead.

Also since that time a lot of new stuff has been implemented in regards to crossfade. See https://www.liquidsoap.info/blog/2024-02-07-liquidsoap-v2.2.4/

Also If you can upgrade to latest Rolling Release (https://github.com/savonet/liquidsoap/releases) you could take FULL advantage of new autocue protocol and/or use autocue2 option by @Moonbase59 here https://github.com/Moonbase59/autocue with some additional features. This would make possible for your to compute cue and/or fade values on the fly.

As for the time being new Rolling Releases should be as stable as ordinary ones.

[JohanJNL]

Thanks for your reply.

I know the !x notation is deprecated, but as long as I keep switching between 2.1.4 and newer releases, it's easier because it still works.

I don't see a options to cross at different times with the new features, so for now I continue using my script with the ugly while loop. I'm happy with the crossfades and it works in 2.1.4. I hope I can upgrade in near future.

[gAlleb]

Maybe you could try single_track=false instead of merge=false in 2.2.4.

[JohanJNL]

No, this doesn't solve the issue. First I tried it only with single_track=false, later together with merge=false. Both tests have the same result that I can't do a new cross with a lower duration and width.

[toots]

Hi @JohanJNL ! I'm not sure what the script is expected to do but I'm pretty sure it might have been working by chance!

What exactly are you trying to implement?

[JohanJNL]

My script calls cross at 8 seconds, 7 seconds ... until 1 second before the end of each track. Every second it tries if a nice crossfade (defined in the transition function) is possible.

In case of a cross (lets say at 5 seconds before the end of each track) sources are merged and cross has nothing to do during the last 4 loops in the while loop. That somehow works until 2.1.4.

In the seconds where no nice crossfades are possible, I call cross with sequence(merge=false, [a.source, b.source]. So in the example of the 5 seconds crossfade, the 8th, 7th and 6th second are sequenced and not merged.

Only jingles are an exception, those I call with sequence(merge=true, [a.source, b.source]. They are in the first loop merged and sequenced.

When I use Liquidsoap-2.2.4 my script works the same as if I would call it only with sequence(merge=true, [a.source, b.source]

[JohanJNL]

I am now running my script on version 2.2.3 for the first time. On this version my script still works fine. It just like in version 2.1.4 with the way of crossfading I like.

I also found a little issue in 2.1.4 and other versions with the width variable in cross. When I use for example integer values width=1. width=2., I receive the correct levels from b.source at 1 and 2 seconds. When I use for example width=1.5 or width=2.5 I don't receive levels from b.source at 1.5 or 2.5 seconds, but levels from a completely different moment in b.source.