28 Aug
2014
28 Aug
'14
9:56 a.m.
Hi fellow audio developers,
This forum is apparently mainly about audio production. But there's another side
regarding audio, and that is: how to create interesting and/or beautiful sounds in
software? Many sound generating
programs try to emulate the sounds of vintage instruments as close as possible, sometimes
with impressive results, but software has many more possibilities then electro-mechanic or
early electronic
intruments.
I try to imagine how the Hammond organ was developed. There must have been a person with
some ideas how he could generate organ-like sounds using spinning tone wheels, each
capable to generate one
sine waveform, and to combine them using drawbars. Then he implemented this idea,
listening carefully to the results, adding and removing different components. The
key-clicks, caused by bouncing
contacts, formed a serious problem, however musicians seemed to like them, and they became
part of the unique Hammond sound.
Compared to the available technical possibilities of the past, software designers nowadays
have a much easier life. A computer and a MIDI keyboard is all you need, you can try all
kinds of sound
creation, so why should you stick trying to reproduce the sounds of yore?
Maybe there are one or two eccentrics like me reading this post? In my opinion a software
musical instrument must be controllable in a simple and intuitive way. So not a
synthesizer with many knobs,
or an FM instrument with 4 operators and several envelope generators. You must be able to
control the sounds while playing. A tablet (Android or iOS) would be an ideal control
gadget. And: not only
sliders and knobs, but real-time, informative graphics.
As an example let me describe an algorithm that I implemented in a (open-source) program
CT-Farfisa. I use virtual drawbars controlling the different harmonics (additive
synthesis). The basic waveform
is not a sine, but also modelled with virtual drawbars. The basic waveform can have a duty
cycle of 1, 0.7, 0.5 etcetera. The final waveform is shortened with the same amount. The
beauty of this is
that you can control the duty cycle with the modulation wheel of the MIDI keyboard, so
it's easy to modify the sound while playing. The program has build-in patches that
have names of existing
instruments, but that's only meant as an indication: they do not sound very similar to
those instruments. This description might sound a bit complicated, but coding it is not
that difficult. Also
several attack sounds are provided, which is very important for the final result. The
program has a touch-friendly interface, runs under Linux (for easy development and
experimentation) and Android
(for playing).
It is not my aim to provide another software tool that you can download and use or not,
but to exchange ideas about sound generation. I know there are many technics, e.g. wave
guides, physical
modelling, granular synthesis, but I think that often it's difficult to control and
modify the sound while playing, in an intuitive way. By the way, did you know that Yamaha,
creator of the famous DX7
FM synth, had only 1 or 2 employees who could really program the instrument?
Wouter Boeke
31 Aug
31 Aug
12:21 p.m.
[W.Boeke]
Compared to the available technical possibilities of
the past, software
designers nowadays have a much easier life. A computer and a MIDI keyboard is
all you need, you can try all kinds of sound creation, so why should you stick
trying to reproduce the sounds of yore?
I definitely agree with the sentiment but it's not an easy task to
purely digitally create timbres as rich, complex and pleasing as
produced by analog, let alone physical (non-electronic) instruments.
For example, analog op-amp or diode saturation is quite simple to
realise and capable of producing anything from very smooth harmonic
extension to screaming distortion. The digital equivalent needs to be
oversampled, complicating things greatly.
Another example is the stretching of the partials of a string with
non-negligible stiffness. Wavetables need to be very large to create
a similar spectrum, while waveguide oscillators need a very high order
of allpass filters, again complicating things greatly.
The numerous imperfections and nonlinearities in real instruments and
the way they play all contribute similar effects, and while they may
often be inadvertent and/or subtle, they are vital to sonic interest.
You can of course eschew these and set out to find features unique to
the digital realm (and there certainly are a few that have some
appeal). Yet I think that there is a deep agreement between the way
we perceive sound and the fundamental characteristics of physical
oscillation, and that is of great importance for me personally (YMMV
of course).
Cheers,
Tim
12:56 p.m.
On Sun, 2014-08-31 at 12:21 +0200, Tim Goetze wrote:
[W.Boeke]
Wait a moment. Building an anlog synth by electronic circuits could be
harder or easier to do, it depends to the skills of the developer. For
software designers in the past hard real-time MIDI was easier to
provide, than it is to provide nowadays. There are many factors
important. Several times even galvanic isolation for the MIDI interfaces
provided by old gear and by modern USB devices was mentioned. What ever
in theory could be programmed, even the different output amplifiers of
different real synth already could provide differences for the sound,
one computer, using the same sound card for several virtual instruments
only would provide, if somebody would care about it. Not to mention
CEM/Curtis filters etc..
Understanding an ADSR with or without a graphic display shouldn't be an
issue for musicians, it might be hard for some beginners. Anyway, that
is provided since decades, even the faders of a simple ADSR already show
the envelope by the fader positions.
I suspect the OP has got less experiences with professional gear, it
abilities and what professional musicians do need.
It can't be repeated often enough. The way that was provided already in
the 80s is a vector control. Everybody can create sounds by a vector
control, absolutely no knowledge is needed. For using an ADSR there's
the need to understand very much, less how the envelope does look like,
more important is what is controlled by the envelope and that the
envelop might variate to the dynamic of the musicians playing, by the
played note number etc..
Compared to the available technical possibilities
of the past, software
designers nowadays have a much easier life. A computer and a MIDI keyboard is
all you need, you can try all kinds of sound creation, so why should you stick
trying to reproduce the sounds of yore?
I definitely agree with the sentiment but it's not an easy task to
purely digitally create timbres as rich, complex and pleasing as
produced by analog, let alone physical (non-electronic) instruments.
For example, analog op-amp or diode saturation is quite simple to
realise and capable of producing anything from very smooth harmonic
extension to screaming distortion. The digital equivalent needs to be
oversampled, complicating things greatly. [snip]
1:55 p.m.
PS: So to provide a new electric instrument even nowadays not only a
computer and a keyboard is needed. The virtual instrument might work on
some distros and on other distros it might not work. It might work with
some computers and sound cards and MIDI interfaces, but not work with
other hardware. If somebody want's to provide the complete solution it's
not just having a computer and keyboard. Even when a software developer
just cares for the studio in the box and maintains it for several
releases of just one distro, but does not care about using it with
external gear, some algorithm (user friendly or not) might need much
computer resources, could cause unwanted latency etc. pp. In the end
building a synth by a circuit layout and some semiconductor could be
easier to do, than to provide software with the guarantee that the
software will do what it should do, since a wrong MIDI interface already
could cause ground loops, assumed somebody want's to connect the virtual
synth with what external gear ever.
Making new sounds for a synth always needs a learning curve. A good GUI
isn't the solution, if a user e.g. doesn't understand how a filter does
work, the best GUI for a filter can't help. What ever synthesis is used,
there is the need to have knowledge, no new synthesis will be a solution
for this "issue". The only way are "workarounds" such as the vector
control, or anything comparable to it.
Computers can't solve higher-ranking issues caused by nature.
2:35 p.m.
Hi Tim,
As an example of attempts to recreate digitally a famous component from the past I would
take the Moog 3-stage lowpass filter. This was build using diodes where their
non-linearity was exploited in a
careful way, yielding a warm, rich sound. Numerous attempts have been made to do this
digitally, resulting in very complicated and CPU-intensive algorithms. I have the feeling
that this is not the way
to go. One better can search for methods that can easily be implemented in software and
that still sound good.
While experimenting I found that slow, small changes in the waveforms are essential, as
well as changes resulting from the excitation level. At that, the kind and value of
distortion is important.
Last but not least the basic waveform is determining, a simple pulse or sawtooth isn't
enough.
Believe me or not, but I found sounds that I enjoy very much, I can listen to a single
note or some simple chords with great satisfaction. And it would be impossible to create
them mechanically or
electronically!
The sound of a good piano can be marvelous, but it is always the same sound, you cannot
modify it. In contrast, if you get bored by a digital sound, you modify it, or choose
another. Vivat software
sounds!
Wouter
On 08/31/2014 11:21 AM, Tim Goetze wrote:
[W.Boeke]
Compared to the available technical possibilities
of the past, software
designers nowadays have a much easier life. A computer and a MIDI keyboard is
all you need, you can try all kinds of sound creation, so why should you stick
trying to reproduce the sounds of yore?
I definitely agree with the sentiment but
it's not an easy task to
purely digitally create timbres as rich, complex and pleasing as
produced by analog, let alone physical (non-electronic) instruments.
3:03 p.m.
On Sun, Aug 31, 2014 at 01:34:59PM +0100, W.Boeke wrote:
As an example of attempts to recreate digitally a
famous component
from the past I would take the Moog 3-stage lowpass filter. This was
build using diodes where their non-linearity was exploited in a
careful way, yielding a warm, rich sound. Numerous attempts have
been made to do this digitally, resulting in very complicated and
CPU-intensive algorithms.
The famous Moog 'ladder filter' has 4, not 3 stages, and doesn't use
diodes but transistor pairs as a non-linear impedance. The result is
similar to a tanh(x) function in each stage. It doesn't take very
complicated code nor excessive CPU.
Ciao,
--
FA
A world of exhaustive, reliable metadata would be an utopia.
It's also a pipe-dream, founded on self-delusion, nerd hubris
and hysterically inflated market opportunities. (Cory Doctorow)
9:13 p.m.
Sorry Fons, if you google "moog filter schematic" you will see 3 stages, where a
stage is a capacitor in series with 2 transistor emitters (which have a diode V/I
characteristic).
About that complicated code: I saw implentations with a lot of code lines, and preventing
oscillations originating from the non-linear tanh functions needed a lot of care.
On 08/31/2014 02:03 PM, Fons Adriaensen wrote:
The famous Moog 'ladder filter' has 4, not 3
stages, and doesn't use diodes but transistor pairs as a non-linear impedance. The
result is similar to a tanh(x) function in each stage. It doesn't take
very complicated code nor excessive CPU. Ciao,
9:42 p.m.
On Sun, Aug 31, 2014 at 08:13:01PM +0100, W.Boeke wrote:
Sorry Fons, if you google "moog filter
schematic" you will
see 3 stages, where a stage is a capacitor in series with 2
transistor emitters (which have a diode V/I characteristic).
All versions I've every seen have four stages, and with three
the feedback wouldn't have the right phase. If you find a
schematic with three stages please post a link.
Each stage is a capacitor in parallel with the series connection
of two emitter impedances, and driven by a current source, the
collector currents of the previous stage.
About that complicated code: I saw implentations with
a lot of
code lines, and preventing oscillations originating from the
non-linear tanh functions needed a lot of care.
I analysed this thing to dead around ten years ago, when I wrote
the MCP plugins. There is no instability problem resulting from
the non-linearity. The main problem with a straightforward digital
implementation is that it becomes a bad approximation at higher
frequencies, this requires some attention to get right. The easy
solution is to run it at a higher sample rate.
Ciao,
--
FA
A world of exhaustive, reliable metadata would be an utopia.
It's also a pipe-dream, founded on self-delusion, nerd hubris
and hysterically inflated market opportunities. (Cory Doctorow)
10 p.m.
On Sun, Aug 31, 2014 at 07:42:48PM +0000, Fons Adriaensen wrote:
Ah, the fons-filter. Extremely stable and a nice acid-y squelch. Since a tanh function
tends to unity at either end, I can't see how that *would* go unstable. You'd
need to have greater than unity gain for it to really go off.
Fons - ever considered turning your analytical skill to the MS20 and Steiner Synthacon
diode-chain Sallen-Key filter?
--
Gordonjcp MM0YEQ
On Sun, Aug 31, 2014 at 08:13:01PM +0100, W.Boeke
wrote:
Correct. Even the "18dB/octave" Roland TB303 ladder filter has four stages.
Why isn't it 24dB/octave? Because they're more honest about the real-world
performance than Moog :-D
Once you go hanging weirdass source and load impedances off your ladder the response can
vary wildly. Tim Stinchcombe has an incredible analysis which shows all sorts of odd
bumps and peaks in what you'd typically expect to be flat passband.
Sorry Fons, if you google "moog filter
schematic" you will
see 3 stages, where a stage is a capacitor in series with 2
transistor emitters (which have a diode V/I characteristic).
All versions I've every seen have four stages, and with three
the feedback wouldn't have the right phase. If you find a
schematic with three stages please post a link.
Each stage is a capacitor in parallel with the series connection
of two emitter impedances, and driven by a current source, the
collector currents of the previous stage. About that complicated code: I saw implentations
with a lot of
code lines, and preventing oscillations originating from the
non-linear tanh functions needed a lot of care.
I analysed this thing to dead around ten years ago, when I wrote
the MCP plugins. There is no instability problem resulting from
the non-linearity. The main problem with a straightforward digital
implementation is that it becomes a bad approximation at higher
frequencies, this requires some attention to get right. The easy
solution is to run it at a higher sample rate.
10:26 p.m.
On Sun, Aug 31, 2014 at 09:00:08PM +0100, gordonjcp(a)gjcp.net wrote:
Correct. Even the "18dB/octave" Roland
TB303 ladder filter has four stages.
Why isn't it 24dB/octave? Because they're more honest about the real-world
performance than Moog :-D
Or they may be some IP related issues.
Once you go hanging weirdass source and load
impedances off your ladder the
response can vary wildly. Tim Stinchcombe has an incredible analysis which
shows all sorts of odd bumps and peaks in what you'd typically expect to be
flat passband.
Yes. In a first approximation the filter isn't to difficult - each stage
is in fact nicely isolated from the following once (since emitter current
will determine collector current, and collector voltage won't affect that
very much since the bases are almost grounded for AC.
Fons - ever considered turning your analytical skill
to the MS20 and Steiner
Synthacon diode-chain Sallen-Key filter?
If you can provide a schematic I may be tempted...
Ciao,
--
FA
A world of exhaustive, reliable metadata would be an utopia.
It's also a pipe-dream, founded on self-delusion, nerd hubris
and hysterically inflated market opportunities. (Cory Doctorow)
10:43 p.m.
On Sun, Aug 31, 2014 at 08:26:18PM +0000, Fons Adriaensen wrote:
http://yusynth.net/Modular/EN/STEINERVCF/
:-D
--
Gordonjcp MM0YEQ
Fons - ever
considered turning your analytical skill to the MS20 and Steiner
Synthacon diode-chain Sallen-Key filter?
If you can provide a schematic I may be tempted...
11:21 p.m.
On Sun, Aug 31, 2014 at 09:43:36PM +0100, gordonjcp(a)gjcp.net wrote:
On Sun, Aug 31, 2014 at 08:26:18PM +0000, Fons
Adriaensen wrote:
http://yusynth.net/Modular/EN/STEINERVCF/
:-D
:-D That's going to be horrible. Everything coupled to everything
else, defeats any analysis except a quite low amplitudes.
Reminds me of a simulation I had to develop last year: aircraft
landing gear. Non-linear forces all over: the struts, friction,
sticktion, and whatever you can imagine. Had to run at a few kHz
sampling rate to get any credible results.
Ciao,
--
FA
A world of exhaustive, reliable metadata would be an utopia.
It's also a pipe-dream, founded on self-delusion, nerd hubris
and hysterically inflated market opportunities. (Cory Doctorow)
Fons -
ever considered turning your analytical skill to the MS20 and Steiner
Synthacon diode-chain Sallen-Key filter?
If you can provide a schematic I may be tempted...
1 Sep
1 Sep
12:16 a.m.
On Sunday, August 31, 2014, Fons Adriaensen <fons(a)linuxaudio.org> wrote:
--
--
A C Censi (on the road)
accensi(a)gmail.com
Accensi(a)montreal.com.br
On Sun, Aug 31, 2014 at 09:43:36PM +0100,
gordonjcp(a)gjcp.net
<javascript:;> wrote:
Original Steiner article
http://user.xmission.com/~dingebre/N-Steiner-VCF-1974.pdf
On Sun, Aug 31, 2014 at 08:26:18PM +0000, Fons
Adriaensen wrote:
> > Fons - ever considered turning your analytical skill to the MS20 and
Steiner
> > > Synthacon diode-chain Sallen-Key filter?
> >
> > If you can provide a schematic I may be tempted...
> >
>
> http://yusynth.net/Modular/EN/STEINERVCF/
>
> :-D
10:35 a.m.
So it goes like this:
The OP starts an interesting topic and illustrates this with a small example. Then
reactions appear where the example is critisized and the main topic ignored. Other
reactions give interesting
solutions to problems that were not mentioned at all. A little sadder and wiser again.
Fons, yet a final question about your Moog filter. First about complexity. Looking at the
source code in Ladspa_Moogvcf1 I see 24 multiplications and divisions per sample. With
e.g. 10 voices and 2
times oversampling this will form a heavy load for the CPU. A classic Chamberlin 4th order
filter contains 6 multiplications, and doesn't need oversampling. So an interesting
question might be: do you
hear a difference between a Moog- and a Chamberlin filter? And if the answer is yes, then
isn't there a simpler way to reach the desired result? From the literature I learned
that distorsion
characteristics inside a filter are the main contribution to the perceived sound.
Wouter Boeke
12:54 p.m.
Date: Mon, 1 Sep 2014 09:35:25 +0100
From: w.boeke(a)upcmail.nl
To: linux-audio-dev(a)lists.linuxaudio.org
Subject: Re: [LAD] Software sound
an interesting question might be: do you
hear a difference between a Moog- and a Chamberlain filter? And if the answer is yes,
then isn't there a simpler way to reach the desired result?
Yes, you definitely hear the difference between these two algorithms. The best way I could
describe the difference would be 'warmth', 'character.'
You will also see very different performance profiles for the two algorithms, the
Chamberlains are a lot more efficient.
Life is full of choices but they are all trade-offs. I might stick my neck out and suggest
that it could be possible to optimise Fons' design to factor out a few divisions. I
did the same (with help and advice from other people) on a Huovillainen filter and there
are a lot of commonalities between the two algorithms. The result on the Huovilainen
algorithm was the same sound quality, smaller CPU footprint but still not much closer to
the efficiency of the Chamberlain, almost to the point that it was not worth the effort.
Other more efficient optimisations are possible using approximation for the tanh()
function but they do affect the filter characteristics.
As a side note to this, the Moog design was patented as a four stage ladder filter using
transistors. Diverse manufacturers copied the design, the most notable that I know of was
ARP with their Odyssey synth. Moog sued them for abuse of patent, ARP backed down and
agreed to pay royalties for use of the design (in all subsequent sales of their synth).
The point I wanted to make is some people mentioned that the design used diodes rather
than transistors. Well the Moog design did not use diodes but there were several other
manufacturers who took the Moog design and manipulated it to run with diodes hence
circumventing the patent - if you have seen such designs then it is possible they were
released due to not being in violation. Pretty sure the muSoniq Sonic-5 did this and here
is another rub: muSoniq was started by an ex-Moog employee (Zumchek) and muSoniq also
later bought the rights to the Moog company after Bob pretty much went bankrupt.
Regards, nick
4:07 p.m.
Hi Nick,
Sounds interesting, but also rather vague. What was the base waveform, the cutoff freq,
the quality factor, the kind of internal filter distorsion? If the differences between
Moog and Chamberlin are
subtle, then Moog isn't worth its extra CPU cycles.
'Warmth' depends on the lower harmonics. If you are aiming for warmth then start
with a base signal with more lower harmonics, e.g. a square wave i.s.o. a pulse. Also
equalizing and a suitable
amplifier are important. And the ultimate way to get warmth is to add a small amount of
1/2 or 2/3 harmonics. In software this is easy, for real instruments it's impossible!
W.Boeke
On 09/01/2014 11:53 AM, Nick wrote:
Date: Mon, 1
Sep 2014 09:35:25 +0100
From: w.boeke(a)upcmail.nl
To: linux-audio-dev(a)lists.linuxaudio.org
Subject: Re: [LAD] Software sound
an interesting question might be: do you
hear a difference between a Moog- and a Chamberlain filter? And if the answer is yes,
then isn't there a simpler way to reach the desired result?
Yes, you definitely hear the difference between these two algorithms. The best way I
could describe the difference would be 'warmth', 'character.'
6:36 p.m.
On Mon, Sep 01, 2014 at 03:07:32PM +0100, W.Boeke wrote:
'Warmth' depends on the lower harmonics. If
you are aiming for
warmth then start with a base signal with more lower harmonics, e.g.
a square wave i.s.o. a pulse. Also equalizing and a suitable
amplifier are important. And the ultimate way to get warmth is to
add a small amount of 1/2 or 2/3 harmonics. In software this is
easy, for real instruments it's impossible!
I'm not sure why you'd say it's impossible, although I'm not sure what
you're trying to describe with "1/2" and "2/3" - maybe adding in
lower frequencies than the fundamental? I'm not sure why you'd do that and 2/3
would be inharmonic and sound shite.
If you mean 2nd and 3rd harmonics, they're *easy*.
--
Gordonjcp MM0YEQ
6:46 p.m.
Yes, I'm talking about adding lower frequencies. They don't sound shit(?), Hammond
and organ players do it all the time.
On 09/01/2014 05:36 PM, gordonjcp(a)gjcp.net wrote:
I'm not sure why you'd say it's
impossible, although I'm not sure what you're trying to describe with
"1/2" and "2/3" - maybe adding in lower frequencies than the
fundamental? I'm not sure why you'd
do that and 2/3 would be inharmonic and sound shite. If you mean 2nd and 3rd harmonics,
they're *easy*.
7:12 p.m.
Date: Mon, 1 Sep 2014 15:07:32 +0100
From: w.boeke(a)upcmail.nl
To: linux-audio-dev(a)lists.linuxaudio.org
Subject: Re: [LAD] Software sound
Hi Nick,
Sounds interesting, but also rather vague.
Vague? Yeah, I could go along with that. The thing is, how do you describe a sound quality
in words? I have heard people compare filter variants as 'fruitier' or
'squelchier' than others.
Whatever words do it for you, take them. The alternative is not that difficult: rig up a
synth that can select between Chamberlain and other filter algorithms and listen to them.
They all have really different qualities and personally I don't think it is as simple
as adding sub-harmonics.
If the differences between Moog and Chamberlin are
subtle, then Moog isn't worth its extra CPU cycles.
Again, you have to decide that for yourself, I would definitely rate Fons' and
Huovillainen's algorithms as more than worth the extra CPU but I accept that it is
subjective. There are plenty of cases where using a Chamberlain in a synth that is buried
in the mix will hide its deficiencies. Use it as a solo/lead instrument and it will be
lacking.
'Warmth' depends on the lower harmonics. If
you are aiming for warmth then start with a base signal with more lower harmonics, e.g. a
square wave i.s.o. a pulse. Also equalizing and a suitable
amplifier are important. And the ultimate way to get warmth is to add a small amount of
1/2 or 2/3 harmonics. In software this is easy, for real instruments it's impossible!
I don't think this really works, it is akin to just lowering the input signals by some
given divisor. It does not address the non-linear aspects of the filter and it does not
address how the irregularities of the filter change over the cut-off range whilst it is
being driven by a modulation signal (env/pressure/etc) - these are both aspects of the
above algorithms that are not addressed by Chamberlain filters.
Can't help you with your questions on input signals, frequencies, etc. I don't
have them and I refer back to describing a sound as to whether these parameters are
relevant. I used to be a musician playing in bands, then later started writing software. A
few years down the line I used to records stuff, look/edit the sampled waves and got to
the point where I would think 'man, you have to come and look at this sound!'
At that point I gave up on being a musician. When I stopped getting my ears around the
sound then making music lost much of its charm.....
Nick
31 Aug
31 Aug
10:52 p.m.
On 08/31/2014 08:42 PM, Fons Adriaensen wrote:
On Sun, Aug 31, 2014 at 08:13:01PM +0100, W.Boeke
wrote:
You are right: it's 4 stages.
W,Boeke
Sorry Fons, if you google "moog filter
schematic" you will
see 3 stages, where a stage is a capacitor in series with 2
transistor emitters (which have a diode V/I characteristic).
All versions I've
every seen have four stages, and with three
the feedback wouldn't have the right phase. If you find a
schematic with three stages please post a link.
3769
days inactive
3773
days old
linux-audio-dev@lists.linuxaudio.org
19 comments
7 participants
participants (7)
-
A. C. Censi -
Fons Adriaensen -
gordonjcp＠gjcp.net -
Nick -
Ralf Mardorf -
Tim Goetze -
W.Boeke