bass_boost

[Sarabanga]

bass_boost does not work as written in comments.
It is clear from the description that it should work like "Frequency below which sound is considered as bass", It is like "low shelf" filter,
but we have a completely different sound.

[smimram]

Thanks for the report, it seems to be due to quite a bad bug. I have minimized to this:

s = sine()
s' = amplify(0.001, s)
s = add(normalize=false, [s, s'])

output(s)

We hear almost nothing wereas we should hear one sine (or more precisely 1.001 sine).

@toots any idea? Looks like a bug with the sharing of frames...

[toots]

@Sarabanga is the bug happening on main or in 2.0.2? main has been in a state of flux, with some heavy changes already committed and others pending which I haven't had time to complete yet.

[Sarabanga]

@toots , 2.0.2.
liquidsoap_2.0.2-ubuntu-focal-1_amd64.deb
Ubuntu 20.04

[toots]

Thank you. We'll have a look then.

[smimram]

For future reference, here is a more full-fledged test:

#!../../src/liquidsoap ../../libs/stdlib.liq

%include "test.liq"

s = sine()
s = add(normalize=false, [s, amplify(0., s)])
s = rms(s)
rms = s.rms
output(s)

def on_done () =
  print("RMS: #{rms()}")
  if rms() > 0.5 then test.pass() else test.fail() end
  shutdown()
end

thread.run(delay=2., on_done)

[smimram]

For info 5418311 is passing the test. Let's bissect.

[smimram]

The outcome is, as expected: 7057092 is the culprit!...

[toots]

Ha! But the original reporter is also pointing to 2.0.2. Could we have two different issues here?

[smimram]

Ah right, it could, my test is passing with 2.0.2...

[smimram]

Ok so here you go, bass boost with increasing gain in 2.0.2. The effect starts to get really noticeable starting from 0 dB, you can hear that there is mostly no distortion in the higher frequencies:

out.mp4

So, it seems to work in 2.0.2 for me.

For info, the script I used is

def gain()
  -20. + time.up()
end

s = single("annotate:liq_cue_in=50.:em.mp3")
s = cue_cut(s)
s = bass_boost(gain=gain, s)
s = mux_video(video=blank(), s)
s = mksafe(s)

s = video.add_text.native({"Bass boost:\n#{string_of_float(decimal_places=0, gain())} dB"}, s, speed=0, size=100, y=260)
output.audio_video(s)
enc = %ffmpeg(format="mp4",
%audio(codec="libmp3lame", q=4),
%video(codec="libx264", preset="fast", crf=20))
output.file(enc, "/tmp/out.mp4", s)

[toots]

@Sarabanga how does @smimram's test compares to you?

[Sarabanga]

It sounds good.
my current sound is like a beaver singing from a toilet bowl.

[toots]

Ha 😅

Can you share a script to reproduce?

[Sarabanga]

All by default.

radio = amplify(1.,override="replay_gain",radio)
radio1 = sky(radio)
radio2 = nrj(radio)
radio3 = bass_boost(radio1)

Listen here
http://kulverstukas.nethub.fi:8000/

[Sarabanga]

beaver comes after 6db
It is not "low shelf"

out.mp4

[toots]

Thanks for checking on this, @Sarabanga. Sam and myself cannot really hear it on our systems and also with headsets but I think I have a lead:

It looks like the signal starts clipping after a certain threshold, which might be responsible for the artifact that you are hearing.

My guess is that, in order to do. automated audio processing like that, you'd need to make sure that you have enough room in your signal on each track, which seems tricky. As such, you'd probably need to at least:

• Normalize to -6dB
• Apply bass_boost to +6dB
• Renormalize after the filter.

Another approach could be to apply the bass_boost before the sky and nrj which, typically, already push your loudness to the max, leaving very little room for additional boost.

I'm not sure that this is actually an issue with us so I'm converting this to a discussion. Thanks for reporting, tho, this is interesting info!

[Sarabanga]

thanks for the advice!