• Fix situation where supplied JSON data could still overwrite cue_file results even if a file’s liq_cue_file tag was set to true, meaning the cue_file results should be preferred in any case.
• For freshly generated/downloaded news or time announcements, this could lead to erroneous cut-offs or loudness jumps in case AzuraCast’s media checker hadn’t yet picked up the new data. It would then overwrite the actual file data with stored values from the database.

• New liq_cue_file handling, allows to ignore overrides for cue_file data if true. This is mainly for fast-changing files like news or time, for which LS/AzuraCast might not yet have updated the metadata.
  • not set — default behaviour (metadata can override cue_file results)
  • false — don’t autocue (still use metadata if present)
  • true — cue_file results override metadata (special use cases)

For a more thorough explanation, see the FAQ

• Make duration non-overridable by existing metadata or annotations.
  • The real file duration is not a tag, but determined otherwise.
  • If such a tag is found, it is typically unusable, incorrect or in a different format.
  • AzuraCast might annotate the duration from an earlier version of the file, because the media scan is only done once in a while. With frequently changing files of the same name, like syndicated news or time announcements, this could be problematic.

• Fixed a situation where cue_file would read a string duration from ffprobe for audio stream #0, which led to an error in the AzuraCast log like this:

  2024/07/02 16:03:27 [autocue:2] Error while processing autocue: error(kind="json",message="Parsing error: json value cannot be parsed as type {duration: float, _}",positions="at autocue.liq, line 814 char 6 - line 826 char 10")
  

• cue_file now outputs the same duration for both tag reading and full analysis mode.
• cue_file will not anymore write a duration tag to files. That was a newly introduced bug in v4.0.4.

• Allow to override results via JSON even after having done a fresh analysis (automatic or forced), for ultimate flexibility when using cue_file for pre-processing. You can now add fades, ramp or hook points, or do other calculations and feed the results into cue_file for tagging. Use with care, because some values are dependent on others. In any case, cue_file will ever only write tags beginning with liq_ and (if requested) replaygain_.
• Prevent some strange errors that could happen when piping something into cue_file and JSON input was not stdin. We now use ffmpeg -nostdin to prevent it reading input that was meant for cue_file.
• Don’t write all liq_* tags (could have side effects), but only those known (see cue_file --help for current list).
• Streamlined tag conversion code a little.

• Changed default of -x/--extra and settings.autocue.cue_file.overlay_longtail from -15.0 LU to -12.0 LU, requested by @RM-FM and the community. Together with the -d/--drop default change from 60.0 to 40.0, this makes for a "tighter" playout and doesn’t lose too much of long or sustained endings.
• Updated README and Presentation as well, to reflect the new default.

• Allow recognition of values like -8.33dB (without blank before "dB") in autocue.cue_file.liq as well (it’s already in cue_file). Note: Metadata with units should be used with a blank in between value and unit.
• Streamline "sustained endings" code.
• Add some diagnostics output to cue_file on the console to easier find and fix problems:

  Overlay: -23.50 LUFS, Longtail: -38.50 LUFS, Measured end avg: -48.40 LUFS, Drop: 29.09%
  Overlay times: 336.50/348.50/0.00 s (normal/sustained/longtail), using: 348.50 s.
  Cue out time: 353.00 s
  

  Overlay times shown as 0.00 mean a calculation has not been done or wasn’t needed, like longtail above. Diagnostics are output to stderr, so don’t interfere with the JSON on stdout.
• Changed default for "sustained endings" loudness drop from 60.0% to 40.0%. This was a wish from the community and makes for a little tighter/denser song overlap.
• Modify to your liking by using cue_file’s -d/--drop parameter or settings.autocue.cue_file.sustained_loudness_drop := 40.0 in Liquidsoap/AzuraCast. A value of 0.0 switches the "sustained endings" feature completely off (behaviour like in older versions).

• 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

• Even better transitions, using both "longtail" logic and the new "sustained endings" feature by @RM-FM, which we could even improve on, in a collaborative work. We are really happy with the results!

• Combined with the existing blankskip handling (whose default is now 5.0 seconds instead of 2.5 s, to avoid false triggers), we are 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!

• If you’re interested, here’s part of our testing set. Try it out for yourself!
  Songs with difficult endings:

  • Beatles, The - Strawberry Fields Forever
  • Ben Folds Five - Underground
  • Black, Mary - Columbus
  • Darkness on Demand - Quicksand
  • Def Leppard - Let It Go
  • Electric Light Orchestra - Don't Bring Me Down
  • ella_henderson_rudimental_-_alibi_feat._rudimental
  • Global Deejays - Get Up (feat. Technotronic)
  • J.B.O. - Ein Fest
  • Nirvana - Something in the Way _ Endless, Nameless
  • Pink Floyd - Goodbye Cruel World
  • Queen - Bohemian Rhapsody
  • radiomonster.fm_-_dropin_01
  • R.E.M. - Losing My Religion
  • robbie_williams_-_angels
  • Stürmer, Christina - Ich lebe
  • test-5-15-5-15-5s
  • testfiles.txt
  • TLC - Waterfalls
  • Toto - Africa
  • Vega, Suzanne - Tom's Diner (vocals only version)
  • Walker, Tom - Leave a Light On
  • Who, The - Won't Get Fooled Again
  • Wonder, Stevie - Another Star

• New "sanity check" that (also) checks if your external cue_file and the Liquidsoap code have matching versions (a suggestion by John Chewter (@JohnnyC1951)), and shuts down otherwise. After your settings, simply use this code:

  # Your settings go here...
  
  # Check Autocue setup, print result, shutdown if problems
  # The check results will also be in the log.
  # Returns a bool: true=ok, false=error. We ignore that here.
  # Set `print=true` for standalone scripts, `false` for AzuraCast.
  ignore(check_autocue_setup(shutdown=true, print=true))
  

• Always use the check_autocue_setup function after your settings, in both standalone scripts and AzuraCast. It will not only check versions, but also set the crossfading duration to a correct value, based on your fade-out setting.

• New informational boolean value liq_sustained_ending in cue_file result, file tags, and Liquidsoap metadata. It shows whether the "sustained endings" feature has been used to determine the "start next song" overlay point.

• Analysed timings shown on console, in case you manually tag a file or just want to check the results. This output don’t affect normal operation, it is done to stderr:

  Start next times: 257.50/261.30/0.00 s (normal/sustained/longtail), using: 261.30 s.
  Cue out time: 263.20 s
  

• Adjustable minimum silence length for blank skipping, i.e., cueing out early when silence is found in the track, for instance from a long pause and "hidden tracks".
• A much requested feature many have been waiting for (including myself)!
• This new function adds almost no extra CPU load.
• Longer silent parts and "hidden tracks" will be detected accurately, without many false triggers, and the track cued out early, with a perfect transition.
• Long tail handling, overlay point detection (next song start) and fading stay active as before—they just work from the new cue-out point.
• The new cue-out point will be set at the beginning of the specified and detected silent part. We don’t want to produce "dead air", after all.
• cue_file still supports the -b/--blankskip option, which will use a default of 2.5 seconds minimum silence duration.
• Note: If using only -b/--blankskip without a value, you should use it as the last parameter and then add a -- before the filename, to signify "end of parameters". This is standard syntax and you probably know it already from other programs.
  Example:

  $ cue_file -kfwrb -- "Nirvana - Something in the Way _ Endless, Nameless.mp3"
  

• You can add the desired duration after -b/--blankskip in seconds, it is a new optional parameter.
  Example:

  $ cue_file -k -f -w -r -b 5.0 "Nirvana - Something in the Way _ Endless, Nameless.mp3"
  

• In Liquidsoap/AzuraCast, you can use this setting which defaults to zero (0.00) and means "disabled":

  settings.autocue.cue_file.blankskip := 0.0
  

  Don’t forget to change this setting if you used true or false before!

• I recommend not to use this feature on jingles, dry sweepers or liners, advertisements and podcast episodes. Especially spoken text can contain some pauses which might trigger an early cue-out.
• Nobody wants a podcast episode to end in the middle, or even risk losing ad revenue!
• You can easily prevent this and turn off blank skipping
  • by tagging a file with the liq_blankskip tag set to 0.00,
  • by not using cue_file’s -b/--blankskip option when writing tags,
  • by prefixing annotate:liq_blankskip=0.00 on playlists.
  • An annotation has precedence over a file tag.
• In some special cases we automatically turn off blank skipping:
  • SAM Broadcaster: Song category is not "Song" (S). For all other categories like News (N), Jingles (J), Ads (A), etc. We look for the songtype tag here. This is useful if you use a common music library, or have files that have been tagged using SAM categories.
  • AzuraCast: Hide Metadata from Listeners ("Jingle Mode") is selected for a playlist. This sets a jingle_mode=true annotation we honor.

• liq_blankskip is now a float, not a boolean anymore!
• Re-tagging recommended! Sorry for that, but I hope the new functionality will outweigh the effort.
• Both cue_file and autocue.cue_file.liq will handle the "old" tags gracefully, using 0.00 for former false and your setting or the default of 2.50 s for former true.
• 0.00 (zero) now means "blankskip disabled".

• Make JSON override switchable (settings.autocue.cue_file.use_json_metadata). Defaults to true.
• Minor code cleanup (thanks @vitoyucepi).

• JSON override tags for cue_file in temp file: Allows passing annotate/database overrides to cue_file, to reduce re-analysis runs even more.

• Prepare for third-party pre-processing software:
  • A JSON file (or stdin) can now be used to set or override any of the known tags (see help). Values from JSON override values read from the audio file’s tags.
  • This can be used, for example, to set values from a database (like AzuraCast’s cue and fade data from their Visual Cue Editor).
  • cue_file might still decide a re-analysis being necessary, so it’s wise to re-consolidate the data after cue_file has returned its results.
• More robust variable reading from tags or JSON:
  • Booleans can be bool or string
  • Typechecking on tags with unit suffixes, especially liq_true_peak, which was a "dBFS"-suffixed string in v1.2.3 and now is a linear float.
• Added liq_fade_in & liq_fade_out to known tags. We don’t use these, but pre-processors might want to set them.
• More advance example in test_autocue.cue_file.liq that shows how annotations can be used to play 15-second snippets of songs.

• Fix ffmpeg erroneously treating .ogg files with cover image as video. The .ogg filetype is now handled by Mutagen.

• Show autocue.cue_file version number in Liquidsoap log on startup, to ease bug hunting. Your cue_file should use the same version.

The log entry looks like this:

2024/06/05 16:16:03 [autocue.cue_file:2] You are using autocue.cue_file version 2.0.2.
2024/06/05 16:16:03 [autocue.cue_file:2] Assure that the external "cue_file" has the same version!

• Fix small bug when reading already stored liq_true_peak that contained a dBFS value from v1.2.3.

Version 2.0.0 is out! It offers the chance to write ReplayGain data to audio files, in addition to the Liquidsoap/Autocue tags.

This is useful for station owners that can’t or don’t want to use audio tracks that have been pre-replaygained. You can now add ReplayGain data to songs and thus ensure the analysis won’t be done over and over again.

This is an additional option to the existing write_tags feature and will only work when write_tags is enabled. Based on our file analysis, it will save (and overwrite) ReplayGain Track data, such as the gain value, the reference loudness, integrated loudness and true peak info.

The logic is inside cue_file, so it can as well be used for pre-processing your files without the need of other tools. Autocue & ReplayGain—all in one!

The cue_file help now shows the file formats it supports, and the tags it knows about. cue_file will automatically detect if Mutagen is installed and offer you a whopping 20 more filetypes to work with!

• liq_true_peak (in dBFS) has been renamed to liq_true_peak_db.
• A new liq_true_peak has been added that stores the peak value as expected and needed by ReplayGain.
• Both cue_file and autocue.cue_file.liq must be updated to v2.0.0 for everything to work correctly.
• It is advisable to install Mutagen on the system running Liquidsoap, or in the AzuraCast Docker container. Doing this gets you more file and tag types you can work with, and safer/more compatible tagging. Mutagen can usually be installed with pip3 install mutagen (preferred), or sudo apt install python3-mutagen (often older versions, distros don’t update that often).

First public version with a numbering scheme. We use SemVer:

• MAJOR.MINOR.PATCH
• MAJOR increments when API-breaking changes are released.
• MINOR increments with new functionality compatible with the existing API.
• PATCH increments when API-compatible bugfixes and minor changes are done.

The version numbers of autocue.cue_file and the external cue_file binary should usually be the same.

Introducing:

• Mutagen (Pyhon tagging library) should be installed and allows writing tags to many file and tagging formats.
• ffmpeg is now only used for tagging where we know it’s safe.
• Much better error checking:
  • Writing tags only to "known good" file types.
  • Gracefully skip writing to write-protected files/folders/drives.
  • cue_file has got better CLI params checking and much improved help.

• First released built-in implementation in AzuraCast Rolling Release #caeea9d.

In preparation for later merge with master branch:

• Rename autocue2.liq to autocue.cue_file.liq.
• Rename test_local.liq to test_autocue.cue_file.liq.
• Use same cue_file as in master branch, update autocue.cue_file.liq.
• Add settings to test_autocue.cue_file.liq so autocue does the right thing (hopefully).

• Started a CHANGELOG.md, so interested parties can follow.
• Streamlined autocue2 code, removed unnecessary stuff.
• Change and solidify metadata handling. It now ensures that:
  • Metadata seen in both autocue2: protocol and enable_autocue2_metadata() mode.
  • All "processing flags" always seen, like
    • liq_autocue2
    • liq_blankskip
    • jingle_mode
  • replaygain_track_gain & liq_amplify always unified when settings.autocue2.unify_loudness_correction is true. ReplayGain, if it exists, takes precedence.
  • All possible combinations of enable_autocue2_metadata(), autocue2: protocol, and annotate: actually work correctly. It is still considered a user error if enable_autocue2_metadata() is used in conjunction with autocue2:, since this would invoke processing twice.
  • Priorities are always respected, so the user can override automatically calculated values using tags in the files, annotations or AzuraCast’s Visual Cue Editor, for example. Priorities are, from lowest to highest:
    • Calculated values from cue_file/autocue2.
    • liq_* and replaygain_track_gain as tags in the audio file.
    • Annotated settings like liq_blankskip=false. These include all settings coming from AzuraCast’s Visual Cue Editor and Advanced Edit Media settings.

You can now do really wild things, like enabling autocue2 processing globally, and still create something like a "sampling" playlist which plays 30s-snippets of a list of songs. And even with the simplest crossfade command possible!

enable_autocue2_metadata()

...

radio = playlist(
  prefix='annotate:liq_cue_in=30.,liq_cue_out=60.,liq_cross_duration=2.5:',
  'Classic Rock.m3u'
)

radio = crossfade(radio, fade_in=0.1, fade_out=2.5)

Or you can exclude your video files from processing, for example:

video_playlist_night = playlist(
  id="video_playlist_night",
  prefix="annotate:liq_autocue2=false:",
  mode="randomize", reload_mode="watch", videos_night_folder)

The sky is the limit!