Proof of Concept: More professional AzuraCast/Liquidsoap AutoDJ Crossfading

[Moonbase59]

Many stations have expressed a certain dissatisfaction with the curent AutoDJ crossfading, which uses balanced fade-in/fade-outs. BUT: In radio, we normally don’t use fade-ins. To station owners that are real readio people the AutoDJ sounds, well, odd.

Since this would officially be a feature request, I have tried to set up a proof of concept and would like to share my findings with you.

Typical radio crossfading (desired outcome)

In radio, we usually use a crossfading curve like this (not an exact picture):

This typically has parameters like these:

• Fade in: 0 s (There are no fade-ins in radio; DJs use that in live mixes)
• Fade out: 3.5 s
• Fade out starts at: -15 dB (refers to radio standard R128 EBU)
• Max. crossing time: 4.0 s
• Width of the power computation window: 2.0 s (Liquidsoap default)

AzuraCast situation

In AzuraCast (as of May, 2023) this is currently almost impossible to achieve, since

• there is only one Crossfading duration parameter in the station profile, which is used to set
  • the fade-out time
  • the fade-in time
  • the fading duration (by multiplying with 1.5)
• writing custom crossfading functions in Edit Liquidsoap Configuration doesn’t work out, because
  • the parameters given there for the cross and cross.simple functions (i.e., fade_in=0.00) are always overwritten since AzuraCast generates an annotation for every track that overwrites the defaults (which makes sense for modified tracks, but not for all of them)
  • when setting Disable crossfading in the station profile, an annotation liq_cross_duration="0.00" is generated, overriding the values in a custom function and thus effectively disabling crossing
  • when setting Normal Mode, a default (uneditable) crossfading function is generated in the Liquidsoap configuration, which means both a custom and the built-in crossfading function are executed

Modifying AzuraCast as a proof-of-concept

The currently only chance is modifying /var/azuracast/www/src/Radio/AutoDJ/Annotations.php within the Docker container, which of course only holds until it is restarted or updated. I’m using some hardcoded values here, just for demonstration:

    public function annotateForLiquidsoap(AnnotateNextSong $event): void
    {
        $media = $event->getMedia();
        if (null === $media) {
            return;
        }

        $station = $event->getStation();
        if (!$station->getBackendTypeEnum()->isEnabled()) {
            return;
        }

        $backendConfig = $station->getBackendConfig();

        $annotations = [];
        $annotationsRaw = [
            'title' => $media->getTitle(),
            'artist' => $media->getArtist(),
            'duration' => $media->getLength(),
            'song_id' => $media->getSongId(),
            'media_id' => $media->getId(),
            'liq_amplify' => $media->getAmplify() ?? 0.0,
            //'liq_cross_duration' => $media->getFadeOverlap() ?? $backendConfig->getCrossfadeDuration(),
            'liq_cross_duration' => $media->getFadeOverlap() ?? 4.0,
            //'liq_fade_in' => $media->getFadeIn() ?? $backendConfig->getCrossfade(),
            'liq_fade_in' => $media->getFadeIn() ?? 0.0,
            //'liq_fade_out' => $media->getFadeOut() ?? $backendConfig->getCrossfade(),
            'liq_fade_out' => $media->getFadeOut() ?? 3.5,
            'liq_cue_in' => $media->getCueIn(),
            'liq_cue_out' => $media->getCueOut(),
        ];

        // Safety checks for cue lengths.
        if ($annotationsRaw['liq_cue_out'] < 0) {
            $cue_out = abs($annotationsRaw['liq_cue_out']);
            if (0.0 === $cue_out || $cue_out > $annotationsRaw['duration']) {

                $annotationsRaw['liq_cue_out'] = null;
            } else {
                $annotationsRaw['liq_cue_out'] = max(0, $annotationsRaw['duration'] - $cue_out);
            }
        }
        if ($annotationsRaw['liq_cue_out'] > $annotationsRaw['duration']) {
            $annotationsRaw['liq_cue_out'] = null;
        }
        if ($annotationsRaw['liq_cue_in'] > $annotationsRaw['duration']) {
            $annotationsRaw['liq_cue_in'] = null;
        }

        foreach ($annotationsRaw as $name => $prop) {
            if (null === $prop) {
                continue;
            }

            $prop = ConfigWriter::annotateString((string)$prop);

            // Convert Liquidsoap-specific annotations to floats.
            if ('duration' === $name || str_starts_with($name, 'liq')) {
                $prop = ConfigWriter::toFloat($prop);
            }

            if ('liq_amplify' === $name) {
                $prop .= 'dB';
            }

            $annotations[$name] = $prop;
        }

        $event->addAnnotations($annotations);
    }

The above will not lose us the ability to manually specify cue & fade in-/out points and it generates annotations like this:

2023/05/02 22:06:05 [lang:3] API nextsong - Response (200): annotate:title="Circulation",artist="Project Pitchfork",duration="235.00",song_id="74218877f7f85c47ca71586d1d70f78e",media_id="103320",liq_amplify="0.00dB",liq_cross_duration="4.00",liq_fade_in="0.00",liq_fade_out="3.50",playlist_id="118":media:Tagged/Project Pitchfork/Project Pitchfork - Akkretion (2018 album, DE)/CD 1_ Akkretion/Project Pitchfork - Circulation.mp3

So we still get an annotation for every track, but it now has the desired values.

In AzuraCast’s station settings, Crossfade Method: Normal Mode should be selected.

Then, the Liquidsoap configuration can be edited. In the input field just above AzuraCast’s def live_aware_crossfade, enter the following:

# Be sure to have ReplayGain applied before crossing.
# This only works correctly when all tracks have valid replaygain tags!
radio = amplify(1.,override="replaygain_track_gain",radio)

# Optionally, cut off silence at start and end of tracks
# Note: This affects the duration, so the annotated value can be wrong
# and the station player appear to show "prematurely ended" tracks!
# Default: Cut below -40 dB
#radio = blank.eat(radio)

# The following doesn’t work since AzuraCast annotates all tracks,
# even when Crossfading is set to Disabled,
# and the annotations override what we specify here. :-(

# Radio-like fading:
# Fade in: 0
# Fade out: 3.5s
# Fade out starts at: -15 dB (refers to radio standard R128 EBU)
# (actually, Liquidsoap’s cross function determines the crossing point)
# Max. time: 4.0s
def my_live_aware_crossfade(old, new) =
    if !to_live then
        # If going to the live show, play a simple sequence
        sequence([fade.out(old.source),fade.in(new.source)])
    else
        # Use the transition
        cross.simple(old.source, new.source, fade_in=0.00, fade_out=3.50)
    end
end

radio = cross(minimum=0.00, duration=4.00, width=2.00, my_live_aware_crossfade, radio)

The cross() minimum specifies the minimum track duration for a cross. -1 means cross always. We should use either 0.00 or a track duration in seconds here (i.e., 30.00).

duration specifies the maximum duration of the crossfade, also the number of seconds to measure at the end of the old track and the start of the new one (each) to compute the transition between tracks.

Using all this, we’re still unable to set fading curves (sin, exp, log, lin) which per default are sinusoidal. Also, we can’t set a dB level for the crossing point.

Nevertheless, Liquidsoap’s cross function does a rather "okay" job, if used with cross.simple, so probably no need to specify a level. Do not use cross.smart, because it’s not that smart for professional crossfading.

I have tested the above patches on my station for a few days, and it sounds much better.

Conclusion

More professional, radio-like crossfades for the AutoDJ are possible using AzuraCast and Liquidsoap.

How to implement in AzuraCast?

Having had a quick glance at AzuraCast code, this should be pretty easy to implement in a useful manner in a two-step approach:

Step 1

• Remove the "Crossfade Duration (Seconds)" input and the x 1.5 calculation for the total duration.
• Set feasible defaults (like above) for non-advanced mode.
• For Advanced Mode, let the station owner input:
  • Fade-in time in seconds
  • Fade-out time in seconds
  • Max. crossfade duration; check validity (must be ≥ fade-in + fade-out)
  • Width of power computation window in seconds (width parameter for cross() ), because the default nails it at 2 s. This should probably be not larger than max. crossfade duration (I’m a little unsure here) and might be beneficial in some cases.
• Keep the per-track annotations, but in /var/azuracast/www/src/Radio/AutoDJ/Annotations.php use these new backend values where appropriate (not width).
• Use the new backend values in /var/azuracast/www/src/Radio/Backend/Liquidsoap/ConfigWriter.php to generate the correct values for the crossfading scripts. (width used here.)

Done – for most normal stations. Radio stations will be happy.

Step 2 – for advanced use

We will probably never be able to set the crossing point dB level, but Liquidsoap’s cross function does a good job on that.

What we could do is set the fader curves for fade-in and fade-out. There might be cases where that would be a nice-to-have.

• In Advanced Mode, I’d use dropdown selections for both fade-in and fade-out to enable the user to select from the supported curves. Currently available are
  • sin
  • exp
  • log
  • lin
• The generated crossfading scripts would have to be modified a little to incorporate the curve selections.

Done again – for the demanding station.

---

Tried this with very different kinds of music and genres: From smooth to loud, from Easy Listening and New Age to Pop, Electronic, Rock and Heavy Metal. Also tested with and without master_me enabled.

Finding all this out, studying documentation and code, setting it up on a real station took me a few days. I hope it can be useful.

[radiobellissima]

@Moonbase59 could you guide me to reach this path? I'd like to test the change but I can't find it via ftp.
/var/azuracast/www/src/Radio/AutoDJ/Annotations.php

[Moonbase59]

Note: The above is strictly experimental, not for the faint of heart, no guarantees, you might break your station if doing something wrong and you should have a good knowledge of Linux, Docker and PHP! It will also only work until the next container restart.

I added nano to my container (for editing) in my azuracast.env:

INSTALL_PACKAGES_ON_STARTUP=nano

Then went into the container using:

./docker.sh bash

There you can then move around, edit the file above and save it. Afterwards, in the station UI, edit your Liquidsoap Configuration as desired and restart broadcasting. Voilà!

[radiobellissima]

Thanks, I'll try.

[BusterNeece]

@Moonbase59 Another perk of LS 2.2.x is that we don't actually have to send "resetting values" on annotations on a per-song basis, as by default now those annotations will all reset with new tracks. That'll let us simplify the annotation code, and will likely let your fade function work with no PHP modification.

[robbedoeske]

Works defenitely perfect!

[Moonbase59]

Note: As of Rolling Release #35a6c8c (2023-06-05 12:18), you don’t need to patch Annotations.php anymore – AzuraCast now uses Liquidsoap 2.2.x and removed now unneeded annotations. Thanks @BusterNeece for implementing this!

You still need to edit your Liquidsoap configuration, as shown above.

Note 2: Should custom fade-in and fade-out times of 0 seconds not work with your jingles, simply set them to 0.1 seconds.

[blindndangerous]

I tried adjusting my liquidsoap configuration, but got this error.
alert LogicException at /var/azuracast/www/src/Radio/Backend/Liquidsoap.php L246: At line 359, char 351-357: Error 6: Cannot apply that parameter because the function at line 359, char 0-365 has no argument labeled "protocol"!
Powered by AzuraCast • Rolling Release #35a6c8c (2023-06-05 12:18) • Docker • PHP 8.2

[Moonbase59]

I cannot see your line numbers here, of course. If using LS 2.2.x, the syntax has changed a little:

Instead of

 if !to_live then

you must use

 if to_live() then

Could that be the cause?

[blindndangerous]

That fixed that error, but now I have another one.
alert LogicException at /var/azuracast/www/src/Radio/Backend/Liquidsoap.php L246: At line 196, char 7-14: Error 5: this value has type () -> _ (inferred at line 33, char 10-20) but it should be a subtype of bool
LogicException at /var/azuracast/www/src/Radio/Backend/Liquidsoap.php L246: At line 196, char 7-14: Error 5: this value has type () -> _ (inferred at line 33, char 10-20) but it should be a subtype of bool.
I'm not sure if it's a problem with your code or if I should open a separate issue. I can paste my liquidsoap if you need, but apart from what you wrote above, I've made no other changes.
radio = amplify(1.,override="replaygain_track_gain",radio)
def my_live_aware_crossfade(old, new) =
if to_live then
# If going to the live show, play a simple sequence
sequence([fade.out(old.source),fade.in(new.source)])
else
# Use the transition
cross.simple(old.source, new.source, fade_in=0.00, fade_out=3.50)
end
end

radio = cross(minimum=0.00, duration=4.00, width=2.00, my_live_aware_crossfade, radio)

[Moonbase59]

From what you pasted, it looks like the parenthesis after to_live are missing. It should be to_live().

[blindndangerous]

Weird, that still fails with the protocol error again. I'm sure my spacing is correct, I just copied and pasted your code block from above and double checked the formatting.

[blindndangerous]

Looking again, it might be something to do with liquidsoap and rolling release as removing the code doesn't fix things.

[Moonbase59]

It’s a good idea to remove personal changes and re-check, yes. In Liquidsoap, it’s rather easy to mess things up. ;-)

[blindndangerous]

I'll wait a bit and see if more ls fixes come through. For now it's back to the way it was (meaning no changes) but is still erroring.

[LordHelmchen666]

I try with Azuracast Rolling Release #6398009 (2023-06-17 4:08)
AutoDJ setting
-> Disabled Crossfade
-> Disabled Replaygain

# This only works correctly when all tracks have valid replaygain tags!
radio = amplify(1.,override="replaygain_track_gain",radio)

# Radio-like fading:
# Fade in: 0
# Fade out: 3.5s
# Fade out starts at: -15 dB (refers to radio standard R128 EBU)
# (actually, Liquidsoap’s cross function determines the crossing point)
# Max. time: 4.0s
def my_live_aware_crossfade(old, new) =
    if to_live() then
        # If going to the live show, play a simple sequence
        sequence([fade.out(old.source),fade.in(new.source)])
    else
        # Use the transition
        cross.smart(old, new, fade_in=0.00, fade_out=3.50)
    end
end

radio = cross(minimum=0.00, duration=4.50, my_live_aware_crossfade, radio)

Seems to work correct. No loudness jumps so far.

@Moonbase59 Thanks for your very good explanations!

[Moonbase59]

@LordHelmchen666 Thanks for sharing! We all learn from each other, and help each other. That’s what it’s all about, ain’t it? :-)

[pozhiloy-enotik]

This masterpiece of engineering should achieve the desired outcome

def liquidpoop(old, new) =
   def add_transition_length(_) =
      [("liq_transition_length","5.")]
   end
   add(normalize=false,
          [ sequence([ blank(duration=3.111),
                       fade.in(duration=0.666,new.source, type="sin") ]),
            fade.out(duration=5.,old.source, type="sin") ])
end
radio = cross(minimum=0.00, duration=5.0, width=20.0, persist_override=true, liquidpoop, radio)

Took me a month to find out as this part of documentation is removed in the new docs https://www.liquidsoap.info/doc-1.4.4/cookbook.html#switch-based-transitions

[blindndangerous]

Let's give it a try. I left in this.

This only works correctly when all tracks have valid replaygain tags!

radio = amplify(1.,override="replaygain_track_gain",radio)
But replaced the rest with your code. Right above the insert telnet driven metadata.

[Moonbase59]

This masterpiece of engineering should achieve the desired outcome […]

Depends upon what "the desired outcome" is.

Your code makes every crossfade a fixed-timing sequence (with added silence). This may be undesired for jingles/sweepers/bumpers/etc. Any per-track crossfading values set in AzuraCast will be ignored. The power computation width of 20s adds to CPU usage, plus it may fail on jingles/sweepers/bumpers that are shorter.

[pozhiloy-enotik]

An example of a professional crossfade is as follows

Suggest your solutions to this problem

Not a single line of the code above was able to delay the second stream as intended. The cross duration(previously erroneously named as start_next) parameter does nothing useful to offset the second sound, but only cuts the overall duration of the fade since liquidsoap is clearly not for radio, but only for entertainment purposes. For the same reason I guess you have to raise the calculation width to 2x the duration, otherwise you'll get a tiny audible crackle 4-6s after a successful transition.

[pozhiloy-enotik]

I definitely can't insist that my code shouldn't be rewritten, the next step would be to fix blank.eat()

if there is a silence at the beginning or end of the song, the silence should be cut for a nice transition

the docs suggest using the radio = blank.eat(radio) method but not mentions the max_blank parameter which actually🙃 allows silence up to 20.0s.

The working code could have been as follows:
radio = blank.eat(max_blank=0.001, radio, at_beginning=true, threshold=-40.0, min_noise=0.2)
......
You wonder if it really works ...
Place this line above the def liquidpoop(old, new) = and get the desired result but also another audio crackle 4-6s after a transition.
Place this below cross and get nothing

[Moonbase59]

@pozhiloy-enotik Looks like you may have found a bug. The ideal place to report that would be upstream to the LS developers: https://github.com/savonet/liquidsoap/issues

Unfortunately I currently don’t have the time to re-test everything that once worked (or seemed to).

[blindndangerous]

I gave this a try, and while it works relatively well it seems to be messing up the now playing. It either gets stuck and doesn't update at all or will be one song behind. After taking it out everything went back to normal.

[Moonbase59]

Autocue v3.0.0 is out!

Features adjustable silence detection length. See Release Notes.

[bombelman]

Great ! do we need to wait for the next AzuraCast update to get this v3.0.0 ?

[Moonbase59]

Yup, but Buster is usually very fast with these. You need to be on the Rolling Releases, though.
For standalone scripts, it already works, of course.

[BusterNeece]

Updated to 3.0.0 on latest RR.

[Moonbase59]

Wow, that man is fantastic! 👍

Works nicely on my four test stations here:

2024/06/12 02:55:59 [autocue.cue_file:2] You are using autocue.cue_file version 3.0.0.
2024/06/12 02:55:59 [autocue.cue_file:2] Assure that the external "cue_file" has the same version!

...

2024/06/12 04:09:19 [autocue.cue_file:3] Now autocueing: "/var/azuracast/stations/niteradio/media/Tagged/Deep Purple/Deep Purple - The Book of Taliesyn/Deep Purple - Wring That Neck (BBC Top Gear session).mp3"
2024/06/12 04:09:19 [autocue.cue_file:3] Blank (silence) skipping active: true, set to 2.5 s
2024/06/12 04:09:19 [autocue.cue_file:3] Clipping prevention active: true
2024/06/12 04:09:19 [autocue.cue_file:3] Writing tags: false, including ReplayGain: false
2024/06/12 04:09:19 [autocue.cue_file:3] Writing metadata to /tmp/cue_file30c4c0.json
2024/06/12 04:09:22 [autocue.cue_file:3] cue_file result for "/var/azuracast/stations/niteradio/media/Tagged/Deep Purple/Deep Purple - The Book of Taliesyn/Deep Purple - Wring That Neck (BBC Top Gear session).mp3": {"duration": 282.5, "liq_cue_duration": 279.70000000000005, "liq_cue_in": 0.4, "liq_cue_out": 280.1, "liq_cross_start_next": 278.6, "liq_longtail": false, "liq_loudness": "-11.85 LUFS", "liq_loudness_range": "3.46 LU", "liq_amplify": "-6.15 dB", "liq_amplify_adjustment": "0.00 dB", "liq_reference_loudness": "-18.00 LUFS", "liq_blankskip": 2.5, "liq_blank_skipped": false, "liq_true_peak": 0.62, "liq_true_peak_db": "-4.15 dBFS"}
2024/06/12 04:09:22 [autocue.cue_file:3] Clipping prevention: Adjusted calculated replaygain_track_gain from -6.16 dB to -6.15 dB
2024/06/12 04:09:22 [autocue.cue_file:3] No fade-in duration given, using default setting (0.1 s).
2024/06/12 04:09:22 [autocue.cue_file:3] No fade-out duration given, using default setting (2.5 s).
2024/06/12 04:09:22 [autocue.cue_file:2] Given fade-out duration (2.5 s) exceeds available time, using 1.5 s.
2024/06/12 04:09:22 [autocue.cue_file:3] Metadata added/corrected for "/var/azuracast/stations/niteradio/media/Tagged/Deep Purple/Deep Purple - The Book of Taliesyn/Deep Purple - Wring That Neck (BBC Top Gear session).mp3":
2024/06/12 04:09:22 [autocue.cue_file:3] ("duration", "282.00")
2024/06/12 04:09:22 [autocue.cue_file:3] ("liq_amplify", "-6.15 dB")
2024/06/12 04:09:22 [autocue.cue_file:3] ("liq_amplify_adjustment", "0.00 dB")
2024/06/12 04:09:22 [autocue.cue_file:3] ("liq_blank_skipped", "false")
2024/06/12 04:09:22 [autocue.cue_file:3] ("liq_blankskip", "2.5")
2024/06/12 04:09:22 [autocue.cue_file:3] ("liq_cross_start_next", "278.6")
2024/06/12 04:09:22 [autocue.cue_file:3] ("liq_cue_duration", "279.70")
2024/06/12 04:09:22 [autocue.cue_file:3] ("liq_cue_in", "0.4")
2024/06/12 04:09:22 [autocue.cue_file:3] ("liq_cue_out", "280.1")
2024/06/12 04:09:22 [autocue.cue_file:3] ("liq_fade_in", "0.1")
2024/06/12 04:09:22 [autocue.cue_file:3] ("liq_fade_out", "1.5")
2024/06/12 04:09:22 [autocue.cue_file:3] ("liq_longtail", "false")
2024/06/12 04:09:22 [autocue.cue_file:3] ("liq_loudness", "-11.85 LUFS")
2024/06/12 04:09:22 [autocue.cue_file:3] ("liq_loudness_range", "3.46 LU")
2024/06/12 04:09:22 [autocue.cue_file:3] ("liq_reference_loudness", "-18.00 LUFS")
2024/06/12 04:09:22 [autocue.cue_file:3] ("liq_true_peak", "0.62")
2024/06/12 04:09:22 [autocue.cue_file:3] ("liq_true_peak_db", "-4.15 dBFS")
2024/06/12 04:09:22 [autocue.cue_file:3] ("replaygain_reference_loudness", "-18.00 LUFS")
2024/06/12 04:09:22 [autocue.cue_file:3] ("replaygain_track_gain", "-6.15 dB")

[bombelman]

love the energy here, keep it up !
great effort, it is appreciated !

[Moonbase59]

Exciting News on the Autocue!

@RM-FM and I now teamed up to bring you even better transitions, throwing both our approaches ("sustained endings" and "longtail") together into something new and real good.

Both our former approaches had their advantages and pitfalls, and we feel that collaborating and combining them into one new, homogenous work will bring even more benefits to all of us. And more listeners to your stations.

Combined with the existing blankskip handling (whose default will change from 2.5 to 5.0 seconds, to avoid false triggers), we will be able to handle the greatest variety of possible song endings with great transitions. Be it a cold end, a long fade, a trick ending, an unexpected noise, chord, riff or vocal at the end—we catch them all. Pure listening enjoyment!

I plan to have this in Autocue v3.1.x or maybe 4.x.x (depending on breaking changes).

You can always pick up the latest official release at https://github.com/Moonbase59/autocue/releases/latest.

So stay tuned! :-)

[LordHelmchen666]

Yes, using -n

[LordHelmchen666]

Here is a "better" command to pre-tag with cue_file
Prerequisite

• cue_file installed on your host system
• Backup of your files ( i only work with copies)
• Be sure you use same cue_file settings as you have in Azuracast!
• Don't just copy and paste. Understand before running what it does!

find /var/lib/docker/volumes/azuracast_station_data/_data/eisradio/media -regex '.*mp3\|.*\.wma\|.*\.MP3\|.*\.WMA' -type f -print0 | xargs -0 -P2 -n1 cue_file -kwn -b 5.0

Explanation

• /var/lib/docker/volumes/azuracast_station_data/_data/eisradio/media = Path to your media folder (you need to change it)
• -regex '.*mp3\|.*\.wma' = searching only mp3 and wma
• -type f = search only files (no directories)
• -P2 = use 2 threads (cores) if you have 16 cores available you can set it to 16, but then it will use 100% of your CPU!
• -n1 = limit xargs to parse only one file to cue_file . This important because cue_file can not handle more

[Moonbase59]

Very nicely done, and avoiding most pitfalls! Note you might want to handle prerecorded shows, podcasts, jingles, etc. differently (usually without the -b, to avoid early cue-outs).

Also, don’t forget f/--force if you need to re-analyse files that already have cue_file tags. cue_file would otherwise (wrongly, in this case) assume it’s supposed to run fast (as in playout), and just re-use tags and probably even "guess" some values.

Rule: Playout fast, pre-tagging exact!

Over here, I use mostly cue_file -fkwrb -- file.ext for songs, and cue_file -fkwr file.ext for jingles, DJ sets, recordings, etc. Jingles I also usually tag with fade-in & fade-out set to 0.1s.

[LordHelmchen666]

The command find /var/lib/docker/volumes/azuracast_station_data/_data/eisradio/media -regex '.*mp3\|.*\.wma' -type f -print0 | xargs -0 -P8 -n1 cue_file -fkwn -b 5.0 now "only" needed 4h30m to tag 38.000 files.

[Moonbase59]

Very nice! Thanks for the feedback!

[BornKillaz]

@Moonbase59, thank you for your explanations. I will just run the cue_file on pre-recorded shows when they are uploaded by the DJ's. I'm already running rsgain (the project built on loudgain).

But, the default cue_file settings are the same that ship with AzuraCast?

Thank you.

[Moonbase59]

@BornKillaz I’m not affiliated with rsgain and didn’t even know they forked mine! 😀 I still use loudgain.

Anyway, nice to see people building on what I wrote. Maybe it has its benefits, and I have heard of people using it.

Yes, the default settings of cue_file are the same the Liquidsoap integration uses (and thus AzuraCast). Note that the blankskip (-b/--blankskip) parameters have changed and will change again in AutoCue 4.x:

• before v3.0.0 — boolean (true/false), no definable minimum silence duration
• v3.0.0+ — optional float (0.00–60.00), zero means "switch off", default 2.5 seconds
• v4.0.0+ — optional float (0.00–60.00), zero means "switch off", default 5.0 seconds

So if you are using cue_file 3.x now, you might already use -b 5.0 to have it use 5 s instead of the 2.5 s used when you only use -b/--blankskip.

Old tags already written to files in pre-v3 versions will be read and handled gracefully, replacing true with the current default.

[BornKillaz]

@Moonbase59, since there is talk in these pre-recorded shows, I'm setting -b 10.0 just to be safe.

$ /usr/local/bin/cue_file /var/www/myfile.mp3 -k -w -b 10.0

Thank you.

[Moonbase59]

You could also simply not use the -b switch at all. It will then disable blank skipping for this track. So no guesswork involved and you’re on the safe side!

[RM-FM]

@Moonbase59 Yet unreleased 4.0.0 with new Sustained Ending Algorithm is now live on all RadioMonster.FM Channels!

[RM-FM]

@Moonbase59 Since the "dB" fix yesterday, files with liq_amplify in annotation are also handled correctly. I can't give a final feedback yet, but the relevant endings have already been retained for many songs so far. The transitions sound smooth and not hasty.

[wegGehamstert]

Did you changed something in the setttings?

[Moonbase59]

Most often used cue_file parameters are these:

• -k/--noclip — clipping prevention, as in loudgain
• -b [BLANKSKIP]/--blankskip [BLANKSKIP] — where [BLANKSKIP] is the optional number of seconds a silent part must have before it is used to cue-out the track at the silence’s beginning.
• -w/--write — tags to file except ReplayGain2 data
• -r/--replaygain — add ReplayGain2 tags if also using -w
• -f/--force — force a re-analysis (important if you changed things and want to write tags again!)
• -n/--nice — use nice on Linux/MacOS to avoid it blocking too much CPU

I mostly use cue_file -fwrb -- "My Long Prerecorded Show.flac" when quickly tagging a file. (Switches that have no parameters can be grouped. The usual Linux rules apply, i.e., use -- if needed.)

[Moonbase59]

So you’re fine with using only cue_file—one for all! ;-)
The 1/100 LU/dB difference one sometimes sees is just a rounding error, since I force display to two decimals. No one would hear a difference. And—as stated—the peak calculation is done differently by different programs. I use the most complicated and aggressive analysis, so you get the real (true) peak that might happen in a playout chain.

If tagging, you should be sure Mutagen is installed on the machine you’re using cue_file on. It makes available much more file types, especially Apple M4A and ALAC files. Might come in handy. cue_file --help will show you if Mutagen is installed, and what file types it can write to safely.

[LordHelmchen666]

@Moonbase59 Would it be a problem to have files tagged generated from different software (loudgain, rsgain, cue_file)? So I could pre-tag downloaded files (syndications) only with cue_file.

I am also with BornKillaz executing only one command would be nice.

[Moonbase59]

@LordHelmchen666 As I said, no problem if you don’t need the album RG. I made cue_file use nearly (as far as possible) the same logic and filters as loudgain, and even the -k/--noclip switch, so you can get by. Or did I misunderstand your question?

loudgain has some other features, though, like special options for tags (upper-/lowercase), ID3v1 tag removal, etc. It depends. Personally, I use loudgain for my music collection, and would use cue_file for station-only data, if it were not the same. ;-)

[LordHelmchen666]

That's exactly what i want to know. And i use it this way now. Thanks!
I use rsgain now (because it can skip tagging if replaygain tags already exists, what speeds up a lot) on my local music library.
I copy them to my Azuracast Server and there i use cue_file to tag them with liq* tags. cue_file -kwn -b 5.0
All syndication files (news, prerecorded shows, etc.) i only use cue_file for liq* and replaygain cue_file -kwrn

[Moonbase59]

…and don’t forget the -f… ;-)

[BornKillaz]

@Moonbase59 ,

I've encountered an unexpected issue where a pre-recorded show of approximately 34 minutes is being ended by AutoCue at around the 23-minute mark. Analyzing the wave spectrum, despite the lack of proper compression and track gain adjustments, reveals a section with lower volume. However, this section doesn't appear to be significantly lower in volume...

And the corresponding Liquidsoap log:
`2024/06/13 22:00:01 [autocue.cue_file:3] ("liq_amplify", "-5.58 dB")
2024/06/13 22:00:01 [autocue.cue_file:3] ("liq_amplify_adjustment", "0.00 dB")
2024/06/13 22:00:01 [autocue.cue_file:3] ("liq_blank_skipped", "false")
2024/06/13 22:00:01 [autocue.cue_file:3] ("liq_blankskip", "10.00")
2024/06/13 22:00:01 [autocue.cue_file:3] ("liq_cross_start_next", "1398.")
2024/06/13 22:00:01 [autocue.cue_file:3] ("liq_cue_duration", "1400.50")
2024/06/13 22:00:01 [autocue.cue_file:3] ("liq_cue_in", "0.00")
2024/06/13 22:00:01 [autocue.cue_file:3] ("liq_cue_out", "1400.5")
2024/06/13 22:00:01 [autocue.cue_file:3] ("liq_fade_in", "0.1")
2024/06/13 22:00:01 [autocue.cue_file:3] ("liq_fade_out", "2.5")
2024/06/13 22:00:01 [autocue.cue_file:3] ("liq_longtail", "true")
2024/06/13 22:00:01 [autocue.cue_file:3] ("liq_loudness", "-12.42 LUFS")
2024/06/13 22:00:01 [autocue.cue_file:3] ("liq_loudness_range", "7.85 LU")
2024/06/13 22:00:01 [autocue.cue_file:3] ("liq_reference_loudness", "-18.00 LUFS")
2024/06/13 22:00:01 [autocue.cue_file:3] ("liq_true_peak", "1.284000")
2024/06/13 22:00:01 [autocue.cue_file:3] ("liq_true_peak_db", "2.17 dBFS")
2024/06/13 22:00:01 [autocue.cue_file:3] ("replaygain_reference_loudness", "-18.00 LUFS")
2024/06/13 22:00:01 [autocue.cue_file:3] ("replaygain_track_gain", "-5.58 dB")
(...)
2024/06/13 22:00:02 [show_meta:3] ("duration", "2072.92")
2024/06/13 22:00:02 [show_meta:3] ("liq_amplify", "-5.58 dB")
2024/06/13 22:00:02 [show_meta:3] ("liq_amplify_adjustment", "0.00 dB")
2024/06/13 22:00:02 [show_meta:3] ("liq_autocue", "cue_file")
2024/06/13 22:00:02 [show_meta:3] ("liq_blank_skipped", "false")
2024/06/13 22:00:02 [show_meta:3] ("liq_blankskip", "10.00")
2024/06/13 22:00:02 [show_meta:3] ("liq_cross_duration", "2.5")
2024/06/13 22:00:02 [show_meta:3] ("liq_cross_start_next", "1398.")
2024/06/13 22:00:02 [show_meta:3] ("liq_cue_duration", "1400.50")
2024/06/13 22:00:02 [show_meta:3] ("liq_cue_in", "0.")
2024/06/13 22:00:02 [show_meta:3] ("liq_cue_out", "1400.5")
2024/06/13 22:00:02 [show_meta:3] ("liq_fade_in", "0.1")
2024/06/13 22:00:02 [show_meta:3] ("liq_fade_out", "2.5")
2024/06/13 22:00:02 [show_meta:3] ("liq_fade_out_delay", "0.")
2024/06/13 22:00:02 [show_meta:3] ("liq_longtail", "true")
2024/06/13 22:00:02 [show_meta:3] ("liq_loudness", "-12.42 LUFS")
2024/06/13 22:00:02 [show_meta:3] ("liq_loudness_range", "7.85 LU")
2024/06/13 22:00:02 [show_meta:3] ("liq_reference_loudness", "-18.00 LUFS")
2024/06/13 22:00:02 [show_meta:3] ("liq_true_peak", "1.284000")
2024/06/13 22:00:02 [show_meta:3] ("liq_true_peak_db", "2.17 dBFS")
2024/06/13 22:00:02 [show_meta:3] ("replaygain_reference_loudness", "-18.00 LUFS")
2024/06/13 22:00:02 [show_meta:3] ("replaygain_track_gain", "-5.58 dB")
2024/06/13 22:00:02 [show_meta:3] Now playing: [REDACTED]
2024/06/13 22:00:02 [cross:3] Analysis: -20.317081dB / -22.772868dB (2.48s / 2.48s)
2024/06/13 22:00:02 [cross:3] Computing crossfade transition over reconciled duration with 2.48s ending and 2.48s starting buffered data.

Notice:
2024/06/13 22:00:02 [show_meta:3] ("liq_cue_duration", "1400.50")
and
2024/06/13 22:00:02 [show_meta:3] ("liq_cue_out", "1400.5")

To prevent this issue in the future, I plan to set a cue_out via the AzuraCast API after cue_file pre-processing. However, I would like to confirm if this behavior is expected.

Additionally, is it possible to set custom cue_file tags directly on the files? Ideally, I would like to configure the cue_out within the cue_file parameters during the cue_file pre-processing, avoiding the need to use the AzuraCast API later.

Thank you for your assistance.

[Moonbase59]

Measure the loudness at 1400.5, and you’ll see it did the right thing. An amplitude display in an audio editor is often misleading—we’re measuring loudness as perceived by the human ear, and the offsets are not based on "what you see", but the integrated track loudness. So the above looks perfectly alright to me, considering the track’s program loudness of -12.42 LUFS.

To avoid that exact situation happening, I always recommend to not use blankskip on prerecorded shows, DJ sets, podcasts, news, jingles, dry sweepers and liners, etc.!

[BornKillaz]

Thank you for your enlightening answers @Moonbase59

From now on, I will not use the blankskip on these files.

However, I did notice this line:

2024/06/13 22:00:01 [autocue.cue_file:3] ("liq_blank_skipped", "false")

Doesn't that means that there will be no blank skipping in this this file?

Thank you.

[Moonbase59]

If you have set the tag (and not overridden it by an annotation): No, I mis-read.

If you set liq_blankskip=false, it won’t do any blank skipping. liq_blank_skipped just shows the result of the operation (i.e., has it or not?).

[BornKillaz]

@Moonbase59, is this the correct command to NOT skip blanks? I'm in doubt if it can be 0.00 or just 0...

$ cue_file MyFile.mp3 -frw -b 0.00

For the issue above:

• I've previously parsed that file with cue_file -b 10.0

• I have the default liquidsoap config on AzuraCast

Thank you.

[Moonbase59]

Yo can use -b 0 or leave the -b out completely.

[Moonbase59]

A lot of work, day and night, but we did it:

Autocue v4.0.0 is out!

• Release & Release Notes at: https://github.com/Moonbase59/autocue/releases/tag/v4.0.0

Thanks to @RM-FM & @JohnnyC1951 for their contributions! (And to me for many, many days and nights spent coding, testing, talking and supporting… 😁 )

[BornKillaz]

Very nice @Moonbase59 I can barely wait to try this. Thank you.

[Moonbase59]

2024-06-15 - v4.0.1

• Bug fix: Catch index error in sustained ending calculation if we’re already at the end of the track. This could happen in very rare cases, if no earlier overlay start than the end of the track could be found.
• Added song that exposes this situation to the test set:
  • Camel - The Great Marsh

Please upgrade, just in case.

[BornKillaz]

Is it safe to use this cue_file standalone version on files to be played on the current AzuraCast Rolling Release version?

[Moonbase59]

"This cue_file version" >= 3.0.0? If so, yes, absolutely.
When changing things, I always try to keep things as backwards-compatible as possible.
Good question, though!

[wegGehamstert]

Hey, i want to know how the fading is reacting to songs wich are forced to play at a certain time. will there be a "hard" cut oder is there still a good fading?

i'm talking about this setting in the playlist configuration. As far as i can remember there was a very weird behaviour somewhere around the 2.0 version. Could somebody test it please so i can make sure, that AutoCue would be a option for me?

[Moonbase59]

I don’t know the exact way AzuraCast does the interrupt, but I guess it should work. At least my top-of-the-hour time announcement works fine. I mean if it makes a request, and that is prepared by Liquidsoap and we see it, it will be autocueing it correctly.

[wegGehamstert]

Okey, for the blankskip, do i need to activate it trough custom liquidsoap code or how am i doing it?
settings.autocue.cue_file.blankskip := true # skip silence in tracks dosent let me save anymore

[Moonbase59]

Right. Because it has changed. Back in version 3.0.0..

[sander816]

just updated, but as soon as i enable "Enable AutoCue (Beta)", liquidsoap doesn't start anymore.
There is nothing in the LS boxes(everything commented out with #)

No idea why it says Nonexistent file or ill-formed URI
or
Couldn't read playlist: request resolution failed

disabling autocue and everything is playing again

2024/06/15 16:16:57 [playlist_antennehits:3] Reloading playlist with URI /var/azuracast/stations/maasduinen_radio/playlists/playlist_antennehits.m3u.
2024/06/15 16:16:57 [request:3] Nonexistent file or ill-formed URI "/var/azuracast/stations/maasduinen_radio/playlists/playlist_antennehits.m3u"!
2024/06/15 16:16:57 [playlist_antennehits:3] Couldn't read playlist: request resolution failed.
2024/06/15 16:16:57 [request.dynamic:3] Fetch failed: empty.
2024/06/15 16:17:08 [main:3] Shutdown signal received.
2024/06/15 16:17:08 [threads:3] Main loop exited
2024/06/15 16:17:08 [main:3] Shutdown started!
2024/06/15 16:17:08 [threads:3] Waiting for main threads to terminate...
2024/06/15 16:17:08 [local_1:3] Closing connection...
2024/06/15 16:17:08 [clock.main:3] Streaming loop stopped.
2024/06/15 16:17:08 [threads:3] Main threads terminated.
2024/06/15 16:17:08 [threads:3] Shutting down scheduler...
2024/06/15 16:17:08 [threads:3] Scheduler shut down.
2024/06/15 16:17:08 [main:3] Cleaning downloaded files...
2024/06/15 16:17:08 [main:3] Freeing memory...
2024/06/15 16:17:08 >>> LOG END

[Moonbase59]

Could you search for LOG START and then post the three lines with [autocue.cue_file:2] as shown below?

2024/06/15 13:34:58 >>> LOG START

...

2024/06/15 13:34:58 [autocue.cue_file:2] You are using autocue.cue_file version 4.0.1.
2024/06/15 13:34:58 [autocue.cue_file:2] The external "cue_file" is version 4.0.1
2024/06/15 13:34:58 [autocue.cue_file:2] Setting `settings.autocue.target_cross_duration` to 2.5 s, from `settings.autocue.cue_file.fade_out`.

If autocue was shutting down your LS due to a mis-configuration, it would look more like this:

[autocue.cue_file:1] ERROR: autocue.cue_file v4.0.1 doesn’t match external "cue_file" v3.0.0!
[autocue.cue_file:1] Autocue NOT ACTIVATED!
[autocue.cue_file:1] Shutting down...

So I think this has nothing to do with autocue per se.

[Moonbase59]

Btw, if it looks like a bug (your LS shutting down), better make an issue so Buster and I can more easily follow it. We both follow the issues since this is relatively new technology we’re testing in the Rolling Releases. Thanks.

[sander816]

today i wanted to try to post the log, but its working now, no idea what went wrong

[Moonbase59]

Autocue v4.0.2 is out

See Release Notes.

[LordHelmchen666]

Do i need to newly pre-tag files with every update?

[Moonbase59]

Not necessarily, if you’re happy with the results already. I would advise to do so if we have breaking changes (like the blankskip change from bool to float lately). These are noted in the changelog and/or release notes.

In case a default changes, like from 4.0.2 to 4.0.3, it’s a good idea to take a few testfiles, evaluate the changes, and decide then. A good tool is for instance Audacity, you can switch the selection input to seconds start & end, and just enter the values cue_file got you on the test file.

[darellrazk]

@Moonbase59, I really appreciate the frequent updates on Autocue. However, I cannot afford the downtime (estimated 30 minutes of high CPU usage on a 16 vCPU setup) on my station. Therefore, version 4.0.2 will be my last update for a while. I truly appreciate all the effort you put into this masterpiece.

[bombelman]

@Moonbase59, I really appreciate the frequent updates on Autocue. However, I cannot afford the downtime (estimated 30 minutes of high CPU usage on a 16 vCPU setup) on my station. Therefore, version 4.0.2 will be my last update for a while. I truly appreciate all the effort you put into this masterpiece.

just curious, how many songs and how many playlists ?

[Moonbase59]

@bombelman ~160,000 tracks (central on NFS-mounted NAS), 4 stations with ~70 playlists ea. here. Only about 20 of these are pre-tagged (jingles, a few test songs) so I can always test the worst case.

(170698 songs shown because some are in more than one playlist)

@darellrazk No problem, 4.0.3 was mainly a "community convenience" update, it only changed the default of -x/--extra and settings.autocue.cue_file.overlay_longtail from -15.0 LU to -12.0 LU. You can do this manually in the v4.0.2 settings and on the CLI.

For comparison: All stations here start rather fast (like without Autocue) if you disable Always Write Playlists to Liquidsoap. When enabled, it takes about 8 minutes, on a 4 CPU 4 GB VM, before the load comes down again to ~1–2.

[sander816]

I dont know what i am doing wrong, but I have a playlist of 10 files with jingles, which i assigned jingle mode too. I removed all cue points, but still some jingles are cut at the end, before they are finished. Op do 20 jun 2024 om 09:20 schreef Matthias C. Hormann < ***@***.***>:

…

@bombelman <https://github.com/bombelman> 160,000 tracks (central on NFS-mounted NAS), 4 stations with ~70 playlists ea. here. Only about 20 of these are pre-tagged (jingles, a few test songs) so I can always test the worst case. @darellrazk <https://github.com/darellrazk> No problem, 4.0.3 was mainly a "community convenience" update, it only changed the default of -x/ --extra and settings.autocue.cue_file.overlay_longtail from -15.0 LU to -12.0 LU. You can do this manually in the v4.0.2 settings and on the CLI. For comparison: All stations here start rather fast (like without Autocue) if you disable *Always Write Playlists to Liquidsoap*. When enabled, it takes about 8 minutes, on a 4 CPU 4 GB VM, before the load comes down again to ~1–2. Auswahl_427.png (view on web) <https://github.com/AzuraCast/AzuraCast/assets/3706922/a8253891-6128-4a83-a2a9-fdf20e10d405> — Reply to this email directly, view it on GitHub <#6252 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AIODLJP5ADDPRKAPEF64EWLZIJ7CDAVCNFSM6AAAAAAXTUOU76VHI2DSMVQWIX3LMV43SRDJONRXK43TNFXW4Q3PNVWWK3TUHM4TQMRVGM4DE> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[Moonbase59]

@sander816 Without having much more details about your setup, it’s impossible to diagnose this.

First idea that comes to mind: "removed all cue points"—is it possible your jingles are just faded out using the defaults (typically 2.5s)? You’d need at least some fade settings for the jingles, like in the Visual Cue Editor or Advanced tab:

I typically just set the fades short for jingles, and leave the rest to Autocue:

This will override the default settings for the jingles (they will still be used for songs).

[sander816]

Ok, i thought when jingle modus was applied, the files weren't touched by any cue points or any fade in/out

[Moonbase59]

Better know than assume. It’s all in the presentation, the docs, the FAQ. "Jingle Mode" just auto-deactivates the blank skipping mode, so not to cut something short because of pauses.

[Moonbase59]

New FAQ for Autocue

I added a FAQ – Frequently Asked Questions section to my repo that answers the most often asked community questions.

[BornKillaz]

@Moonbase59 , I have 1 specific playlist that I would like to set liq_fade_out = 5

p = playlist(prefix='autocue:annotate:liq_cue_out=5.00:', '/path/to/playlist.m3u')

The question is, where should I place this command? Do I need to use AzuraCast Manual AutoDJ Mode?

Thank you.

[BornKillaz]

But is it possible to add this to some of LS box in order to get this specific fade out time for one playlist only?

p = playlist(prefix='autocue:annotate:liq_fade_out=5.00:', '/path/to/playlist.m3u')

In the previous post I mentioned cue_out, but I meant fade_out.

Basically I want that this specific playlist overlaps the 2 songs a little bit more.

[Moonbase59]

If I were you, and it were only a few tracks, I’d skip the pain to do all that manually and lose AzuraCast’s advanced features in doing so.

Since we don’t (yet?) have playlist-based Autocue settings in AzuraCast, just go to the playlist, click on the number of entries, and it’ll jump to Music Files, with a pre-made search of playlist:playlistname, showing just the files in that playlist.

Then just clik the Edit button for each, and set the 5 second fade-out in the Edit Media Advanced tab:

---

For the more technically inclined: There is a difference between tagging liq_fade_out=5.00 and using the AzuraCast track editor:

• Using the AzuraCast track editor (or Visual Cue Editor) overrides file tags and Autocue will try to "fulfill your command" by adjusting other parameters to see that a 5.0 second fade can actually be achieved.
• Using tags will only result in a 5 s fade-out if the analyzed ending duration is longer than 5 s, otherwise the fade-out will be shortened, as per usual Autocue calculations.

This is intended behaviour because I follow the Autocue priority chain (low to high): Defaults → tags → annotations, and AzuraCast uses annotations in the track editor.

[BornKillaz]

Thank you for the suggestion, but this playlist has more than 500 tracks, so it would be a time consuming task. I can automate a processes that does that with AzuraCast API, but even so... maybe in the future there can be some option to set different parameters for different playlists.

[RM-FM]

Thank you for the suggestion, but this playlist has more than 500 tracks, so it would be a time consuming task. I can automate a processes that does that with AzuraCast API, but even so... maybe in the future there can be some option to set different parameters for different playlists.

I started to use the media CSV export/import feature and Modern CSV for mass changes like this. Modern CSV is able to filter, sort and change the CSV data just like in Excel but without the need of any tranformations. Just download the media metdadata, edit it on the fly and upload it again - without the risk of messing up something due to data import into speadsheet applications like Excel.

[BornKillaz]

Thank you for the suggestion, but this playlist has more than 500 tracks, so it would be a time consuming task. I can automate a processes that does that with AzuraCast API, but even so... maybe in the future there can be some option to set different parameters for different playlists.

I started to use the media CSV export/import feature and Modern CSV for mass changes like this. Modern CSV is able to filter, sort and change the CSV data just like in Excel but without the need of any tranformations. Just download the media metdadata, edit it on the fly and upload it again - without the risk of messing up something due to data import into speadsheet applications like Excel.

Cool idea! I haven't thought about that! I may even automate the process... Thank you.

[sander816]

I cannot really figure out why my jingles are still cut before they end. I set fade in and out to 0.1, but they get cut before they are played full length

Other problem that i have is that after some days playing the playing song( at its end) song abrupt stops for a few seconds, no audio, then resumes and fades to the next song.
Where dan i check what is going wrong?
The vps has 4xcpu 3.2 ghz and 8gb mem