Formant synthesis

[Big Tick]

Is Big Tick still alive? Rhino seems quiet interesting...

Yeah the virus hasn't got me yet. Mostly working on 2getheraudios stuff these days though.

[GRUMP]

Big Tick - Angelina
https://www.kvraudio.com/product/angelina-by-bigtick

THX that you reminded me of BigTick. I have heard a Demo of Rhino some Months ago, loved the Sounds but was too busy and just loaded the Demo and tried it myself. That little Cutie is a perfect Companion for some of my Creations and a great Place to dive deeper into FM.

[GRUMP]

Is Big Tick still alive? Rhino seems quiet interesting...

Yeah the virus hasn't got me yet. Mostly working on 2getheraudios stuff these days though.

Oh - have seen it too late. Good to know - apologize my Question

My Ex-Girlfriend always liked Rhinos and so do I, too now. It´s of unbelievable Value. All those Patches. Just like a whole Library full of Books. <3 Oldschool FM!°

I just had to search for Tim. FM has lost a Genius with him!

All the Best 4U!

[cryophonik]

On a related note, I owned a Yamaha FS1R for years (but had absolutely no idea what I was doing when it came to sound design with it). It was a hybrid FM/Formant Shaping synth, and I seem to recall some discussions/arguments years ago that it was not actually considered a "formant synth", in the normal sense, but never really understood why. Can anyone enlighten or correct me?

[Kwurqx]

On a related note, I owned a Yamaha FS1R for years (but had absolutely no idea what I was doing when it came to sound design with it). It was a hybrid FM/Formant Shaping synth, and I seem to recall some discussions/arguments years ago that it was not actually considered a "formant synth", in the normal sense, but never really understood why. Can anyone enlighten or correct me?

Not all to familiar with the FS1R, but going by Wikipedia and Sound On Sound articles, it is a "limited" implementation of formant synthesis.

https://en.wikipedia.org/wiki/Yamaha_FS1R

The FS1R uses the "formant sequence", this being a series (128 or 512) of "frames" that define the level and frequency of each of up to eight 'voiced' (pitched) and eight 'unvoiced' (un-pitched) formant generators over time. The number of frames limits the typical length of a sequence to a few seconds, though this length and pitch can be varied in real-time with few or no artefacts. Applying a formant sequence to a sound allows a complex, evolving sound to be programmed in a relatively short time. 90 formant sequences were supplied with the FS1R.

The Wikipedia article refers to a Sound On Sound article with more details on FS1R Formant Synthesis.
https://web.archive.org/web/20150606100 ... hafs1r.htm

And some additional info on the FS1R approach in another Sound On Sound Article on Formant Synthesis.
https://www.soundonsound.com/techniques ... -synthesis

[Kwurqx]

Angelina seems interesting too, but I have no 32Bit Option.

Don't know if your (64-bit?) DAW supports (bit) bridging? E.g. Reaper does. But there are third party options too.

Some examples.

JBridge
https://jstuff.wordpress.com/jbridge/

BridgeWize
https://ddmf.eu/bridgewize/

NetVST (free)
http://netvst.org/wiki/doku.php

TBridge (free)
https://totomusic.github.io/TBridge/indexEn.html
viewtopic.php?t=541616

[Kwurqx]

Recently Stone Voices (from Ambient Reverb and PolyGAS granular synth fame) released MOSS. It is an FFT based EQ. To help set all of the hundreds of individual amplitudes, it can import a spectrum (from a text file).

It is an implementation of a tool I described earlier: it is in effect a huge series of ultra narrow bandwidth bandpass filters where the width (Q) per bandpass filter is just 1 frequency/harmonic. This could very well be used for Formant synthesis / Vocoder effects.

Stone Voices - Moss
http://stone-voices.ru/vst/moss_equalizer?lang=en
viewtopic.php?f=6&t=544728

https://youtu.be/d5EuGqcwWE0

[Kwurqx]

Is Big Tick still alive?

Yeah the virus hasn't got me yet. Mostly working on 2getheraudios stuff these days though.

When calling the (still pretty unique) Angelina to memory earlier in this thread....I also suggested (hoped) that Angelina might get a reissue via 2getheraudio...

Is this under consideration?

[GRUMP]

Is Big Tick still alive?

Yeah the virus hasn't got me yet. Mostly working on 2getheraudios stuff these days though.

When calling the (still pretty unique) Angelina to memory earlier in this thread....I also suggested (hoped) that Angelina might get a reissue via 2getheraudio...

Is this under consideration?

I have some Problems with that Name. She reminds me of someone I dont want on my PC - but 32 Bit is too skinny anyway

MOSS is already installed. I vistit the BPB every Morning

I´ll hopefully find some Time for it - after I have generated some Motown-AIR, layered it with some of my celestial Unison Multisamples from DUNE and pimped that Bunch with some FM-Cheese from my new Rhino

Constructive Contribution to the Discussion from my Side: I have just anylyzed the Effect of a Vowel Filter (The Orb) on a Pad-Chord. Settings (default):

Orb.JPG

Spectrogram - dry:

SPCTR_dry.JPG

Spectrogram - 100% Orb:

SPCTR_Orb.JPG

Audio Example (sry, no Webspace):

The Orb_Example.zip

Result: The Formants in the Mids and lower Highs remain as before. Deep + High Frequencys are cut away (Band-Limiter). The Audio has a stronger Vowel Character with the Orb, but it is obvious that the Formants have already been there before and the Job could also have been done with an EQ - what does not mean, that "The Orb" has no Value - but as you can hopefully hear and see in the Screenshot from Zebra the Voice-Characteristics have been generated with different Options. Turn the Unison off and the Sound is Garbage, Turn the Notch-Filter off and it looses some of its Character - that is mnaybe not a mathematical "Aaah", but maybe because of that even more interesting.

Zebra:

zebra.JPG

[cybilopsin]

So lets talk about the granular-formant connection. A very familiar example of a synth that uses this connection is the "formant/grain complex" oscillators on some Virus synths. They use a granular streams and wavetables to produce sounds with controllable formants.

Terminology can get a little tricky with granular synthesis. A "grain" is a tiny burst of sound, almost like a very short pluck sound. Basically its a waveform with an amplitude envelope that lasts 100ms or less in duration (archetypically). In the most familiar incarnation of granular, the waveforms that make up the different grains are little chunks of an audio sample - either the same little bit for every grain (unchanging grain waveform) or a different little bit for each grain (changing grain waveform). However I will be analyzing granular synthesis that uses different sources for the "grain waveform", or waveform inside each grain. Each grain has its own tiny amplitude envelope, the grain envelope, like a micro- version of a traditional synth envelope. Because grain envelopes are repeated so quickly in a single note-event, or even overlap, they are usually heard not as a modulation of the sounds envelope but rather as part of the timbre of the sound. Now granular synths generally output a more or less continuous "stream" of individual grains - the can overlap or have gaps between them. The "granular stream". The grain stream can have its own envelope, which functions more like a traditional synth envelope, fading the sound in and out.

Ok so here's the connection. If you have a syncronous granular stream, meaning the frequency of new grains being created (sometimes called the density or grain rate) is regular and not random, assuming the grain rate is sufficiently fast enough (30 Hz or above), you will hear one continuous tone, even though that tone is created by a succession of short grains. The fundamental pitch of that tone comes from the grain rate, NOT the grain pitch (the pitch of the grain waveforms). If there are 60 grains per second, the pitch will sound as 60 Hz - regardless of what frequencies are contained in each grain (they could be noise, bits of a sample, an analog waveform, whatever).

If the pitch of the grain stream or note-event is only affected by the grain rate and not by the pitch of the individual grains, what effect does changing the pitch of the grains have? Simply put, it changed the formants of the sound. Specifically, for every frequency contained in the grain waveform, the tone of the overall note will appear to have a formant at that frequency, as if there was a bandpass filter on the sound.

For example, lets say the grain waveform is a sine wave. In other words, each grain is a burst of sine, with an envelope, and the entire sound is created by a stream of many of these grains per second. The frequency of the sine waves inside each grain - the grain pitch - does not change the apparent pitch of the overall tone, but it changes the formant. Say each grain contained a sine wave at 700 Hz, and there are 60 new grains generated every second. The resulting sound will have a pitch (fundamental) of 60 Hz with a formant at 700Hz. Since 700 Hz is not a harmonic of 60, there will be no amount of 700 Hz energy in the output. But the harmonics around 700 Hz - eg 660, 720 - will be amplified compared to the other harmonics, just as if there was a bandpass filter centered at 700 Hz.

But this relation works for any waveform inside the grains, not only sine waves. Lets say I the grain waveform is now a square wave at 700 Hz and the grain rate is 60 Hz (60 grains per second). The pitch will remain 60 Hz. However, now there will be formants around 700 Hz, 2100 Hz, 3500 Hz, 4900 Hz, 6300 Hz, 7700 Hz, etc. Thats because those frequencies are found in a 700 Hz square wave (the odd harmonics of 700 Hz). Because a square wave's harmonics get quieter as you go up, the lower formants will be the loudest.

Also the characteristics of the grain envelope have specific effects on the shape and bandwidth of the resulting formants, too complicated to get into here.

TL;DR - a syncronous granular stream turns the frequencies are contained in the grain waveform into formants of the sound's timbre. Basically the pitch of the individual grains becomes the formant of the overall granular stream. Granular synthesis allows fine control of formants completely independent from pitch.

[cybilopsin]

P.S. if the previous information is adequately digested it implies the following:

If I put a single unchanging waveform - lets say some digital wavetable - into my 60 Hz granular sycronous stream. Every grain reads the same wavecycle as an oscillator. Pitch will be 60 Hz. But if I slowly modulate the *grain pitch* - the pitch of the wavetable oscillators inside the individual grains - I will hear what appears to be the same set of formants sliding together around the frequency spectrum, a sound similar to the Virus synth "f-shift" function. (And note in addition that this modulation could be described as shifting or, more accurately, scaling the formant frequencies). I do not know of a way to shift rather than scale the frequencies unless Frequency-Domain Reconstruction is involved or unless the waveform is changing, because the formants correspond with the frequencies contained in the wave generated by each grain. This application is reminiscent of oscillator hard-sync but has more control over the sound due to the availability of changing the grain-envelope parameters.

[cybilopsin]

I think in a way what GRUMP is saying with regard to the inadequacy of a "mathematical" chart is that the real music lies in a micro-composition of modulating formant frequencies using interesting amounts and control of the specific formants in Hz involved. This theoretical stuff is ultimately secondhand knowledge. To know, one must do

[GRUMP]

I think in a way what GRUMP is saying with regard to the inadequacy of a "mathematical" chart is that the real music lies in a micro-composition of modulating formant frequencies using interesting amounts and control of the specific formants in Hz involved. This theoretical stuff is ultimately secondhand knowledge. To know, one must do

Unfortunately we´d still play with Woodsticks unless some "Physiacians" had made some Oberservations and Inventions... But yes - on the other Hand - there is a Gap between Users and "Philosophers". And as long as we have no Access and like the Sound in the End we may for sure call all that a Theory. No Sound, no good. Simple but true

By the Way Granular Synthesis is common already, but doesn´t yet express in corresponding Parameters. The Philosopher in me would ask "why?" here!

[GRUMP]

So lets talk about the granular-formant connection.[...]

Nice to hear from you, especially in such a instructive Way! Read it - understood something - and ended up with Questions as back then in School

It is difficult to discuss that Topic here. I only have basic Knowledge in "Sound Theory" and my English is not the best, too.

At first I´d ask if the Approach that you describe really exists and if it´s really called "Granular Synthesis". As far as I can see Grains are just the Elements of some Kind of "Frequency Synthesis" here, that is quiet difficult to differ from Additive Synthesis.

And secondly I want to state that the Result of the Operation that you descrice is a Single Cycle Wavetable. What would be the Advantage and Difference of this Approach towards Additive / Wavetable Synthesis? The Problem would still be the same - the Y-Axis, the Time-Dimension [...] with all Implications that are so difficult to operationalize and configure.

We can quiet easily generate SC-WT today and even more complex Wavetables. We can layer, detune and modulate them. We can morph and sequence additive Spectra and what ever. MSoundfactory is a wonderful Playground if you feel like making such Experiments.

But all those Approaches to conquer the Y-Axis failed yet. Yamaha tried it, Wolfgang Palm tried it and probably some others, too - and we all know the Reason why the FS1R for Example was a failure. I have a PPG Phonem here - and I can tell you why this intersting Approach is absolutely not the right Tool if you are out for (at least somehow) musical Results.

At the End of the Day we all face limited Ressources and we have to chose the most practicable Way to generate our Results. And the World of Music has many Examples for pleasurable VOX/Formant Results, that have mostly been made without Granular Magic and probably often "just somehow".

To make the long Story short: My Focus are VOX Pads and my Results are getting better every Day. The Methods I use are mostly simple compared to what I could do - just for one Reason: Quality. They are simply more pleasurable for me - personally. And I have learned that complex doesn´t necessarily mean "better". It usually just means that you´ll just spend more Time on something.

Developing a Theory is the first and Experiments are the second Step of scientific Working. And when it´s about Music (and not Computer-Speech or Signal Processing Theory) they are still and will always be the best Way to go - just because the Goal/Target can´t be calculated (so my Theory!). In othe Words: I´d strongly prefer an inductive over a deductive Approach here...

I´ll search for a Virus Formant OSC Demo now... I still have a Version A standing around here (reactivated last Week), so it´s interesting to see how it has developed. EDIT: Heard all that Years ago already. Yamaha is still ahead and Leads are Off-Topic for me :/

If you want to and find the Time we can maybe start some Experiments together? I have a lack of Theory here and would really like to get in Touch with People who share my Interests...

Regards!

[cybilopsin]

At first I´d ask if the Approach that you describe really exists and if it´s really called "Granular Synthesis".

It is easy to implement usibg any granular engine with a normal level of functionality. All you need is:
* Function to import a soundfile as grain "source table"
* Separate control over the grain pitch (this directly controls formant frequency)
* Separate control over the grain rate or density (control's the notes actual pitch aka fundamental)
* Separate control over the grain envelope (grain attack and decay settings, usually around 1-5ms, directly control formant bandwidth).

This, like the waveform-bandpass technique I mentioned earlier, is simply a basic recipe for creating a harmonic sound with a controllable formant frequency and bandwidth parameter. The actual implementation should involve modulating all of these settings in musical ways, and layering several layers with the same fundamental pitch abd different formant settings. I just think all that is a different topic- it goes without saying that this is just another tool.

I believe these techniques are the basis of the engine that the Virus formant/grain complex oscillators use. That is a well known way to make formant sounds in psytrance especially.

And secondly I want to state that the Result of the Operation that you descrice is a Single Cycle Wavetable

Yes, technically that is true of any granular synthesis, **assuming you don't modulate any parameters with automation/knobtwiddling/envelopes/etc**. The advantages of this method are that several useful parameters for modulation become available:

You can twiddle the grain pitch to directly modulate the formant frequency: double the grain pitxh and formant frequency increases by one octave. Quadruple the grain pitch and your formant freq increases by 2 octaves - and so on.

You can keep grain pitch and everythibg else constant while changing the wavetable for a different wavetable.

Because the individual frequencies in the wavetable transform into individual formant frequencies, each Single Cycle Wavetable functions as an entire set of complex parallel filters. Change one wavecycle for another, and its like you have reconfigured hundreds of filters to create a totally different sound.

That is why I bother describing this technique in detail. You are only asking if this is useful because you haven't fully comprehended the implications of what I have been saying. This is a recipe for creating endlessly varied formant sounds and being able to twist them in extremely powerful ways. Of course the same results can be done with carefully editing wavetables, but this method turns ANY wavetable into a "formanty vocal" wavetable. Don't you understand? With this method you can pick a wavetable at random from your collection and hear it morph from its "normal" sound into a formant-vocal type sound with full use control of the interpolation between them. That's worth something.

For instance, take FM synthesis. Its just as powerful in terms of the sounds it can make. But the control over those sounds ks very difficult- it is well known that the FM parameters affect the resulting formants in a very indirect, nonintuitive way. Fm doesn't give you a direct "formant frequency" control. Everyone knows how you have to "tune in" formant sounds with FM. Granular takes all the guess work out because you have parameters that directly move the formants up and down, while the wavetables you select change the shape of the formants.

Wheb I get the chance I'll demonstrate that using this method is very simple by writing a ridiculously short Csound program that generates formant sounds of enormous complexity and variety from a few simple lines of code. And the same can be replicated in almost any granular Vst if you can load a sine wave as the sound table.