Liquidsoap and Stereo Tool on a RPI3B+ : Optimization, tips, advices...

[nordseele]

Hi,
I'm trying to process the output of Liquid Soap with Stereo Tool on a Raspberry Pi 3B+

I've installed the 32bits cmd line version of Stereo Tool and my script is based on an example seen on the Liquid Soap workshop 1.0.

I have no issues with Liquidsoap 2.0 alone, but with Stereo Tool it fails most of the time directly with some "PANIC" error messages, or it just hangs or skips to the blank() fallback every second until the program suddenly exits.

I'm wondering if anyone has successfully used the same kind of setup, and if so, would share some advices on how to optimize the whole thing ? For example, I'm not sure wether I should set jackd to RT priority or not... Also , should I try to use a RT kernel on the RPI ? Perhaps an install on top of a Patchbox OS ? Or do you think the Raspi 3B+ is just too weak to run Liquid Soap , process sound through Stereo Tool, encode, decode, manage Beets, etc at the same time ? I've not started to stream yet, I'm only testing with a direct output to Alsa.

I know there are other really cool ways to process sound within liq or using FFmpeg or Ladspa, but for the sake of simplicity , I would have liked to start with some presets. Just wondering if the Pi3B is a viable solution for that.

Thanks for any help :)

Hardware and software details:

• Rpi 3B+
• Rasbian OS Buster 32bits
• WM8731 DAC
• Liquidsoap 2.0
• StereoTool 9.23

Here's the script I'm using:

#!/usr/bin/env liquidsoap

BEET = "/home/pi/.local/bin/beet"

def beets(arg="") =
  fun() -> list.map(fun(item) -> request.create(item),
     process.read.lines(
      "#{BEET} random -f '$path' #{arg}"
    )
  )
end

def transition(a,b)
  # If old or new source is not music, no fade
  if a.metadata["source_tag"] != "music" or a.metadata["source_tag"] != "music" then
    sequence([a.source, b.source])
  else
    # Else, apply the standard smart transition
    cross.smart(a, b)
  end
end

# A function to add a source_tag metadata to a source:
def source_tag(s,tag) =
  def f(_)
    [("source_tag",(tag:string))]
  end
  map_metadata(id=tag,insert_missing=true,f,s)
end

# Tag our sources
default = blank()
music = request.dynamic.list(beets("length:..03:30"))
music = source_tag(music, "music")
music = amplify(1.,override="replaygain_track_gain", music)
music = skip_blank(music, max_blank=5., threshold=-50.)

radio = music
radio = cross(duration=5., transition, radio)

st = pipe(process='/usr/bin/stereo_tool - - -q', radio)

s = fallback(track_sensitive=false, [st, default])
output.alsa(s, device="pcm.liquidsoap")

And this is the kind of output I get from the terminal:

[mp3 @ 0x6758b2a0] Estimating duration from bitrate, this may be inaccurate
2021/07/28 19:29:58 [request.dynamic.list_0:3] Prepared "/home/pi/anda/music/jazz/Roland Kirk - Blacknuss - 1972/01 Ain't No Sunshine.mp3" (RID 0).
2021/07/28 19:29:59 [request.dynamic.list_0:2] #get_frame didn't add exactly one break!
2021/07/28 19:29:59 [threads:2] Queue generic queue #1 crashed with exception File "source.ml", line 746, characters 12-18: Assertion failed
pi@radio:~ $ ./radio.liq
no more csLADSPA plugins
2021/07/28 19:36:41 >>> LOG START
2021/07/28 19:36:23 [main:3] Liquidsoap 2.0.0+git@a697602c
2021/07/28 19:36:23 [main:3] Using: bytes=[distributed with OCaml 4.02 or above] pcre=7.5.0 sedlex=2.3 menhirLib=20210419 curl=0.9.1 dtools=0.4.4 duppy=0.9.2 cry=0.6.5 mm=0.7.1 xmlplaylist=0.1.5 ogg=0.7.0 ogg.decoder=0.7.0 vorbis=0.8.0 vorbis.decoder=0.8.0 opus=0.2.0 opus.decoder=0.2.0 mad=0.5.0 dynlink=[distributed with Ocaml] lame=0.3.4 gstreamer=0.3.1 ffmpeg-avutil=97c1fc4 ffmpeg-avcodec=97c1fc4 ffmpeg-avdevice=97c1fc4 ffmpeg-av=97c1fc4 ffmpeg-avfilter=97c1fc4 ffmpeg-swresample=97c1fc4 ffmpeg-swscale=97c1fc4 bjack=0.1.6 alsa=0.3.0 ao=0.2.3 samplerate=0.1.5 taglib=0.3.6 ssl=0.5.9 camomile=1.0.2 faad=0.5.0 ladspa=0.2.0
2021/07/28 19:36:23 [main:3] 
2021/07/28 19:36:23 [main:3] DISCLAIMER: This version of Liquidsoap has been
2021/07/28 19:36:23 [main:3] compiled from a snapshot of the development code.
2021/07/28 19:36:23 [main:3] As such, it should not be used in production
2021/07/28 19:36:23 [main:3] unless you know what you are doing!
2021/07/28 19:36:23 [main:3] 
2021/07/28 19:36:23 [main:3] We are, however, very interested in any feedback
2021/07/28 19:36:23 [main:3] about our development code and committed to fix
2021/07/28 19:36:23 [main:3] issues as soon as possible.
2021/07/28 19:36:23 [main:3] 
2021/07/28 19:36:23 [main:3] If you are interested in collaborating to
2021/07/28 19:36:23 [main:3] the development of Liquidsoap, feel free to
2021/07/28 19:36:23 [main:3] drop us a mail at <savonet-devl@lists.sf.net>
2021/07/28 19:36:23 [main:3] or to join the slack chat at <http://slack.liquidsoap.info>.
2021/07/28 19:36:23 [main:3] 
2021/07/28 19:36:23 [main:3] Please send any bug report or feature request
2021/07/28 19:36:23 [main:3] at <https://github.com/savonet/liquidsoap/issues>.
2021/07/28 19:36:23 [main:3] 
2021/07/28 19:36:23 [main:3] We hope you enjoy this snapshot build of Liquidsoap!
2021/07/28 19:36:23 [main:3] 
2021/07/28 19:36:39 [dynamic.loader:3] Could not find dynamic module for fdkaac encoder.
2021/07/28 19:36:39 [gstreamer.loader:3] Loaded GStreamer 1.14.4 0
2021/07/28 19:36:39 [clock:3] Using builtin (low-precision) implementation for latency control
2021/07/28 19:36:41 [lang.deprecated:2] WARNING: "skip_blank" is deprecated and will be removed in future version. Please use "blank.skip" instead.
2021/07/28 19:36:41 [frame:3] Using 44100Hz audio, 25Hz video, 44100Hz main.
2021/07/28 19:36:41 [frame:3] Video frame size set to: 1280x720
2021/07/28 19:36:41 [frame:3] Frame size must be a multiple of 1764 ticks = 1764 audio samples = 1 video samples.
2021/07/28 19:36:41 [frame:3] Targeting 'frame.duration': 0.04s = 1764 audio samples = 1764 ticks.
2021/07/28 19:36:41 [frame:3] Frames last 0.04s = 1764 audio samples = 1 video samples = 1764 ticks.
2021/07/28 19:36:41 [sandbox:3] Sandboxing disabled
2021/07/28 19:36:41 [video.converter:3] Using preferred video converter: ffmpeg.
2021/07/28 19:36:41 [audio.converter:3] Using samplerate converter: ffmpeg.
2021/07/28 19:36:41 [clock.alsa:3] Streaming loop starts in auto-sync mode
2021/07/28 19:36:41 [clock.alsa:3] Delegating synchronisation to CPU clock
2021/07/28 19:36:41 [switch_0:3] Switch to blank_1.
2021/07/28 19:36:41 [alsa_out(pcm.liquidsoap):3] Using ALSA 1.1.8.
2021/07/28 19:36:41 [alsa_out(pcm.liquidsoap):3] channels=2, samplerate=44100Hz, buffer size=60211B, frame size=4B, periods=2077
2021/07/28 19:36:42 [clock.alsa:3] Delegating synchronisation to active sources
2021/07/28 19:36:52 [request.dynamic.list_0:3] Prepared "/home/pi/anda/music/Various Artists - Strange Breaks & Mr Thing Vol. 2 - 2008/15 Hot City Bump Band - It's Just Begun.mp3" (RID 0).
[mp3float @ 0x67512000] Could not update timestamps for skipped samples.
2021/07/28 19:36:57 [switch_0:3] Switch to pipe_0 with transition.
2021/07/28 19:37:02 [switch_0:3] Switch to blank_1 with transition.
2021/07/28 19:37:04 [switch_0:3] Switch to pipe_0 with transition.
2021/07/28 19:37:06 [switch_0:3] Switch to blank_1 with transition.
2021/07/28 19:37:07 [switch_0:3] Switch to pipe_0 with transition.
2021/07/28 19:37:11 [switch_0:3] Switch to blank_1 with transition.
2021/07/28 19:37:13 [switch_0:3] Switch to pipe_0 with transition.
2021/07/28 19:37:18 [blank.detect_0:2] #get_frame didn't add exactly one break!
2021/07/28 19:37:18 [threads:2] Queue non-blocking queue #2 crashed with exception File "source.ml", line 746, characters 12-18: Assertion failed
2021/07/28 19:37:18 [threads:2] Raised at Process_handler.run.handler in file "tools/process_handler.ml", line 299, characters 16-23
2021/07/28 19:37:18 [threads:2] Called from Duppy.Task.t_of_task.(fun) in file "src/duppy.ml", line 148, characters 50-66
2021/07/28 19:37:18 [threads:2] Called from Duppy.exec in file "src/duppy.ml", line 275, characters 12-21
2021/07/28 19:37:18 [threads:2] Called from Duppy.queue.run in file "src/duppy.ml", line 307, characters 7-24
2021/07/28 19:37:18 [threads:2] Called from Duppy.queue in file "src/duppy.ml", line 357, characters 8-12
2021/07/28 19:37:18 [threads:2] Re-raised at Duppy.queue in file "src/duppy.ml", line 361, characters 8-17
2021/07/28 19:37:18 [threads:2] Called from Tutils.create.(fun).process in file "tools/tutils.ml", line 173, characters 12-15
2021/07/28 19:37:18 [threads:2] 
2021/07/28 19:37:18 [threads:1] PANIC: Liquidsoap has crashed, exiting.,
2021/07/28 19:37:18 [threads:1] Please report at: savonet-users@lists.sf.net
pi@radio:~ $ 

[toots]

Hi! This appears to be a bug so I'm converting this as one, thanks for reporting.

Regarding pip with stereotool, I believe most users do disable the mechanism that closes and restarts the process on each track and adjust the replay_delay parameter.

I just updated the doc for it:

 * replay_delay : float? (default: null)
     Replay track marks and metadata from the input source on the output
     after a given delay. If `null` (default) close and flush the process on
     each track and metadata to get an exact timing. This parameter is
     typically used when integrating with `stereotool`.

[nordseele]

Alright, thanks Toots ! :)
I'll work on the rest of my script and I'll leave the output processing for later until this gets fixed. Or I'll simply use the Ladspa or FFmpeg treatment in the meantime.
Yes, I found the replay_delay in one of the examples of the book after posting this question. I'll read the documentation in more depth.

Thanks again.

[rrolla]

Hi! This appears to be a bug so I'm converting this as one, thanks for reporting.

Regarding pip with stereotool, I believe most users do disable the mechanism that closes and restarts the process on each track and adjust the replay_delay parameter.

I just updated the doc for it:

 * replay_delay : float? (default: null)
     Replay track marks and metadata from the input source on the output
     after a given delay. If `null` (default) close and flush the process on
     each track and metadata to get an exact timing. This parameter is
     typically used when integrating with `stereotool`.

This is something new to me, before there was advice to use merge_tracks #885 (comment)

[toots]

Ha I see. Yes, both advice relate to avoiding a restart of the process I believe.

[KevanGP]

Does anyone know how to get rid of the 5 second delay I'm getting when Stereo Tool is on through Liquidsoap?
I asked on the Stereo Tool forums, and they say it's Liquidsoap's doing.

[DroidU]

I tried it with a working but not full-feature license key, and I also tried it by deleting the license_key variable. It fails in both cases. I also took a closer look at the Icecast2 server log.
[2023-04-07 08:37:26] WARN source/get_next_buffer Disconnecting source due to socket timeout
This disconnection is not logged by Liquidsoap. So it seems that things get stuck somewhere, because after connecting, no data goes to Icecast2. And because of this, it drops the connection.
It might be due to Debian 11. But even Debian 10 is used by a lot of people.
If the bug persists in a newer Liquidsoap version, I may send the VirtualBox vdi file to the above email address.

[toots]

Thanks for pushing back. I tracked it down to a bug in the new muxer code. Should be fixed in main and rolling-release-v2.2.x now (both are currently building)

[DroidU]

I also experimented with encoding to mp2 (twolame) stream without stereotool. It works perfectly in stereo, but if I encode the same in mono, it stops/gets stuck in the same way as in the case of stereotool.
output.icecast(%ffmpeg(format="mp2",%audio(codec="mp2",samplerate=44100,b="128k",ac=1)),host="127.0.0.1",port=8000,mount="/trans.mp3",user="source",password="hackme",icy_metadata="true",mean(mksafe(s)))

CTRL+C to stop doesn't work with stereotool and in this case either.

[toots]

I can confirm that this use-case is also fixed now.

[DroidU]

I tried 471bd7c.
Stream playback is now near perfect in all cases.
However, the stream converted to ogg/vorbis has a very slight skip when changing songs, with "Unknown" appearing instead of the song title. The tiny skip itself may not occur with all players, but it certainly does with VLC.
Also an ogg problem is that if the stream being processed is ogg, all encoder channels display "Unknown" as the song title. And the OPUS and OggFLAC song title display is blank.

[moleculezz]

I'm on x86_64. What files from the Stereo Tool generic plugin do I have to copy to the system?

[hansvanzutphen]

Just libStereoTool_64.so.

[toots]

Hi @hansvanzutphen . I just sent an email to support@thimeo.com with what seems to be an issue with the dynamic plugin erroneously reporting invalid keys. Is there a better way to report issues? Thanks!