24 Sep
2019
24 Sep
'19
10:38 p.m.
A note-on event can lead to clicks in synths, which is often fixed by a
fade-in algorithm.
If the attack rate for a synth's signal is 0, the synth shall not click,
and still fade in as fast as possible. What is the best algorithm for
fading in a signal, that scales also with small buffersizes, like 32
samples per buffer (at 44100 samples/s)?
For example, in zynaddsubfx, we count the ascending zero crossings and
calculate the fade-in length upon it [1]. Then, an S-curve of that
length is multiplied with the signal. Counting zero crossings prevents
clicking on lower notes, and it makes higher notes more punchy. That
algorithm works very well for a buffersize of 256 samples (at 44100
samples/s), but it fails at 32 samples (since the zero crossings of 32
samples are often not representive).
Any papers, or experience with other synths? We currently need a good
solution for (at least) zynaddsubfx and the triple oscillator in LMMS.
Links:
[1]
https://github.com/zynaddsubfx/zynaddsubfx/blob/4e36e765f677dbc689461889e50…
24 Sep
24 Sep
10:56 p.m.
Am 24.09.19 um 20:38 schrieb Johannes Lorenz:
A note-on event can lead to clicks in synths, which is
often fixed by a
fade-in algorithm.
If the attack rate for a synth's signal is 0, the synth shall not click,
Some would disagree with that. Clicking can in fact be used as a tool in
sound design. From a sound programmers perspective I'd rather like the
synth to provide different envelope curves.
Chris
25 Sep
25 Sep
12:27 a.m.
On Tue, Sep 24, 2019 at 12:38 PM Johannes Lorenz <j.spam(a)lorenz-ho.me>
wrote:
Counting zero crossings prevents
clicking on lower notes, and it makes higher notes more punchy.
There's fundamentally no such thing as a zero crossing. You might have two
samples on either side of zero, but you still don't have a sample *at*
zero, so in the general case, truncating one of them to zero and
starting/ending there is still going to give you distortion and/or noise.
Obviously there may be cases where one of them is close enough to zero for
this not to be be an issue, but it's not a general method.
Ardour applies declick fades every time the transport starts and stops. You
can read about how we do it here:
https://github.com/Ardour/ardour/blob/master/libs/ardour/ardour/disk_reader…
https://github.com/Ardour/ardour/blob/master/libs/ardour/disk_reader.cc#L14…
https://github.com/Ardour/ardour/blob/master/libs/ardour/amp.cc#L163
3:12 a.m.
On 9/24/19 4:26 PM, Paul Davis wrote:
On Tue, Sep 24, 2019 at 12:38 PM Johannes Lorenz <j.spam(a)lorenz-ho.me
<mailto:j.spam@lorenz-ho.me>> wrote:
Counting zero crossings prevents
clicking on lower notes, and it makes higher notes more punchy.
There's fundamentally no such thing as a zero crossing. You might have
two samples on either side of zero, but you still don't have a sample
*at* zero, so in the general case, truncating one of them to zero and
starting/ending there is still going to give you distortion and/or
noise. Obviously there may be cases where one of them is close enough to
zero for this not to be be an issue, but it's not a general method.
Ardour applies declick fades every time the transport starts and stops.
You can read about how we do it here:
https://github.com/Ardour/ardour/blob/master/libs/ardour/ardour/disk_reader…
https://github.com/Ardour/ardour/blob/master/libs/ardour/disk_reader.cc#L14…
https://github.com/Ardour/ardour/blob/master/libs/ardour/amp.cc#L163
Hi Paul. Thanks, I was going to mention that.
We had a similar conversation a few years ago when I was adding
anti-zipper noise to MusE, and I inquired about zero-crossing methods.
I replied that TV and amplifier electronic volumes use zero-crossing,
so why not use it?
But I conceded the very point you mention above, ie. where exactly is
the crossing point.
And it is dependent on frequency so you get a different attack result
each time.
In a TV or amp volume chip it's OK but here, meh...
So in the end, looking at the code given above, I can see I did much
the exact same thing as Ardour, using a fixed coefficient value
driven by time passage (sample rate).
Cheers.
Tim.
11:07 a.m.
On Tue, 24 Sep 2019 19:12:47 -0400
Tim <termtech(a)rogers.com> wrote:
On 9/24/19 4:26 PM, Paul Davis wrote:
Just like to add, that on a complex waveform with a strong harmonic content it's
quite likely to cross the zero point 2-3 times during each cycle of the
fundamental.
--
It wasn't me! (Well actually, it probably was)
... the hard part is not dodging what life throws at you,
but trying to catch the good bits.
On Tue, Sep 24, 2019 at 12:38 PM Johannes Lorenz <j.spam(a)lorenz-ho.me
<mailto:j.spam@lorenz-ho.me>> wrote:
Counting zero crossings prevents
clicking on lower notes, and it makes higher notes more punchy.
There's fundamentally no such thing as a zero crossing. You might have
two samples on either side of zero, but you still don't have a sample
*at* zero, so in the general case, truncating one of them to zero and
starting/ending there is still going to give you distortion and/or
noise. Obviously there may be cases where one of them is close enough to
zero for this not to be be an issue, but it's not a general method.
Ardour applies declick fades every time the transport starts and stops.
You can read about how we do it here:
https://github.com/Ardour/ardour/blob/master/libs/ardour/ardour/disk_reader…
https://github.com/Ardour/ardour/blob/master/libs/ardour/disk_reader.cc#L14…
https://github.com/Ardour/ardour/blob/master/libs/ardour/amp.cc#L163
Hi Paul. Thanks, I was going to mention that.
We had a similar conversation a few years ago when I was adding
anti-zipper noise to MusE, and I inquired about zero-crossing methods.
I replied that TV and amplifier electronic volumes use zero-crossing,
so why not use it?
But I conceded the very point you mention above, ie. where exactly is
the crossing point.
And it is dependent on frequency so you get a different attack result
each time.
In a TV or amp volume chip it's OK but here, meh...
So in the end, looking at the code given above, I can see I did much
the exact same thing as Ardour, using a fixed coefficient value
driven by time passage (sample rate).
Cheers.
Tim. 
1:29 p.m.
On Tue, Sep 24, 2019 at 02:26:53PM -0600, Paul Davis wrote:
There's fundamentally no such thing as a zero
crossing. You might have two
samples on either side of zero, but you still don't have a sample *at*
zero, so in the general case, truncating one of them to zero and
starting/ending there is still going to give you distortion and/or noise.
Obviously there may be cases where one of them is close enough to zero for
this not to be be an issue, but it's not a general method.
Even if you would have an exact zero sample, that would not avoid a
click. It would only ensure that the click has a 1/f^2 spectrum
instead of 1/f (so it would contain less HF).
From a synth I'd expect a controllable and
precisely defined fade-in,
not something that tries to be clever. If the fade-in
time has to
depend on frequency (it usually does if the frequency can change over
a large range), that should be programmable or 'voltage controlled'
just as anything else.
Ciao,
--
FA
1:41 p.m.
On Tue, Sep 24, 2019 at 08:38:19PM +0200, Johannes Lorenz wrote:
For example, in zynaddsubfx, we count the ascending
zero crossings and
calculate the fade-in length upon it [1]. Then, an S-curve of that length is
multiplied with the signal. Counting zero crossings prevents clicking on
lower notes, and it makes higher notes more punchy. That algorithm works
very well for a buffersize of 256 samples (at 44100 samples/s), but it fails
at 32 samples (since the zero crossings of 32 samples are often not
representive).
That could be a good method, but why should this (or anything a synth
does) depend on period size ? Implementation may be a bit simpler if
it does, but that's poor excuse (IMHO :-)...
Ciao,
--
FA
9:35 p.m.
On Tue, 24 Sep 2019 at 19:38, Johannes Lorenz <j.spam(a)lorenz-ho.me> wrote:
A note-on event can lead to clicks in synths, which is often fixed by a
fade-in algorithm.
No idea if it's useful in this case, or whether it duplicates anything
anyone else has said, but I use a form of the switch-and-ramp idea by
Miller Puckette to handle discontinuities in hot-swapped DSP code.
http://msp.ucsd.edu/techniques/latest/book-html/node63.html
Best wishes,
Neil
--
Neil C Smith
Artist & Technologist
www.neilcsmith.net
PraxisLIVE - hybrid visual live programming
for creatives, for programmers, for students, for tinkerers
www.praxislive.org
8 Oct
8 Oct
4:20 p.m.
On Tue, 2019-09-24 at 20:38 +0200, Johannes Lorenz wrote:
A note-on event can lead to clicks in synths, which is
often fixed by
a fade-in algorithm.
In most synthesizer tutorials (we're also talking about real hardware
analog synthesizers) they will say that to avoid the click, just add a
tiny amount to the amplitude envelope attack stage.
/Daniel
1868
days inactive
1882
days old
linux-audio-dev@lists.linuxaudio.org
10 comments
9 participants
participants (9)
-
Christopher Arndt -
Daniel Swärd -
Fons Adriaensen -
Johannes Lorenz -
lacuna_＠gmx.net -
Neil C Smith -
Paul Davis -
Tim -
Will J Godfrey