The Holy Grail ? (second reprise)

#61

haha, are you aware who @SmashedTransistors is? i guess so...

#62

Yepp. I sneaked around this code for a while, without understanding the point, but with ST`s posting yesterday I grabbed it.

#63

ouch, that's a very old piece of code...
This is amplitude modulation of a bandlimited signal. The amplitude modulation by a pair of cosines - locked to the pitch - translates the spectrum of the bandlimited signal. Much like AM for radio.

I think that the FS1R is rather based on parts of

http://www.google.com/patents/US5610354

But is difficult to tell as the FS1R is far from an open source system...

#64

The trautonium even included a magnetic wire multi tap delay !

Many instruments of this era used RLC formant filter banks. (for example, the Givelet/Coupleux organ).
Some others such as the Ondes Martenot rather used special transducers - "La Palme", "Le Metallique" to get these kind of results.

#65

But, getting back to the actual (sub)subject: Pray tell me which smiley's are masculine an which are not so that I may never ever, ever make such a mistake. Is this a masculine one maybe: :heart_eyes: ?

#66

If I may:

Formant = overtone shaping is as old as synthesis itself. In what we tend to consider classical synthesis, which is actually subtractive synthesis, the only real compromises made to emulate the influence of formants on a sound are:
1 - adding one resonant peak per available filter
2 - adding key follow so that high sounds may sound bright when low sound sound dull.

In reality shaping the propper overtones is much more complex. Every instrument, including the human voice, has a set of non-altering overtones that shape that instruments basic character. That is also why everybody has his own voice character (and I do not sound like Caruso).

The major question then is: Can you afford to add extra oscillators to create all the dynamic or static overtones you need to get the desired emulation? That is why additive synthesis is, at least in theory, the most powerful synthesis method of all but also the most diffcult to buidl and handle in practice.

If you cannot afford to add oscilators at will another approach must however be used. In stead of using dedicated extra oscillators per formant comb filtering is used to accentuate certain harmonics that are already present in a complex waveform. So this is actually one step below additive synthesis nad therefore a much more practical proposition as vocoders and string/choir ensembles have shown.

The reference to the Trautonium is in so far spot on that the later version that Oskar Sala developed already could be used as a sort of early additive synthesizer (although the same already applies to the classic pipe and Hammond organ) but also had an adjustable bank of non-dynamic filters that could (sort of) emulate fixed formants.

Circling back to the subject at last : Our main associations with vocal sounds are based on the fact that their formants change in a rather pronounced way when we change the shape of our vocal tract. Which brings us to the formant shapig in the FS1R: It uses an algorythm emulating multiple time dynamic filters to accentuate certain frequency ranges to do a very similar thing.

Without searching for the exact patent: In the 90s it became possible to program algorythms that could do complex filtering processes fast enough in real time. So software became fast enough to use all these old insights to actually make your synth sound in real time like somebody is using his mouth to shape the formants of an electronic sound. By the way: Such high processing speeds where also the reason why physical modeling, the original subject of this thread, also moved into focus at about the same time.

Further advances in DSP speed have led to realtime fast formant synthesis systems that can actually create the overtones themselves in stead of using a filter based approach but I expect the FSR1 to be too early for that. It is however possible that bits of Karplus Strong modeling where already involved (https://en.wikipedia.org/wiki/Karplus%E2%80%93Strong_string_synthesis).

Anyway: Never has a synth sounded as much as Peter Frampton doing his talkbox guitar solo on Show Me The Way.

#67

And it had a "touch-based user interface" :wink: (Okay, not multi-touch...)

Thanks for the patent link, too. Always good for expanding one's horizon. I hope I'll get the idea. Usually I do.

#68

There is many algorithms for formant synthesis.
Substractive synthesis is the most obvious one as it can be related to quasi physical modelling (LPC).

Granular synthesis/ FOF synthesis are rather related to the autotune Lent algorithm. An algorithm that allows separated control of pitch and formant frequencies.

Whatever method you use, I think that what is the most important for expressive instrument or voice synthesis is articulation.

After the FS1R Yamaha developped an Formant shaping extension card for the Motif synths that sold exclusively to Japan... not a great success because of poor articulation database and poor integration into the synth.

Their great success is Vocaloid.. Yeah... it's synthetic J-POP :scream:... They developed the Vocaloid software based on a diphone database combined to a Lent algorithm.
Xavier Serra provided them some extensive knowledge and skills about audio database and processing.

#69

I agree. :slight_smile:

#70

I had no idea about that, interesting to read. I only remember it (fondly) as being an FM synth with Formants bolted onto it. The formant sequencing was pretty neat as well cause you could modulate start and stop points etc. I think that's how they get the radio tuning effect he shows in the demo (modulation mapped to start point etc). Might have to download the official manual from Yamaha and have a look, just out of curiosity. It's years now since I owned mine, and I really wish I'd kept it, but I used to buy a synth, then use it while saving more money so that when something else interesting arrived, I could sell it and try something else. Never had the luxury of owning everything at once, which is a real bummer.

Now then, Mr. Brasse, that smiley is perfectly fine and masculine! I even used it in my VHS thread, so no worries on that one. It's that other one that concerns me, the one Smashed Transistors appears to have taken a liking to in the quote below:

Some J-POP is actually quite cool, like this one, though it's a mix of synthetic and guitars. Full-screen it chances are you'll really enjoy it (or at least her when she loses her temper) :heart_eyes::stuck_out_tongue_winking_eye:

That said, she's not the sort of girl a guy would want to have hanging around his synth cave, not with a temper like that. I mean you could come home and find your rackmounts smashed, knobs everywhere, your Axoloti in the fishtank, and your patch cables used as tiebacks!

#71

You can play with
Help->library->community->tiar-osc->CosCosAM
It implements a slight variant of the algorithm you found on musicdsp.

Maybe, i'll try to combine this with @lokki's Formant Settings object...

The main advantage of this kind of methods over filters is that there is less issues with hi resonance saturations and self-oscillations.

#72

that would be cool indeed!!

#73

I've tried something here:
Help->library->community->tiar-osc->CosCosAM2

It's not really a formant sequence, but its quite fun to play with.

#74

Thank you, very interesting unit. As i am not heading for a mini vocaloid, first thing I´ve done, was cranking up th PM. Makes already some beautifull noises, but destroyes the formantic structure. Self PM does`nt, even through a distortion plus chorus. Feedback amount seems to be frequency/formant dependant.

#75

This might have made a good challenge, synths that can sing a complete sentence with no samples allowed.
WIll take a look at that once I get my Linux installation sorted, very curious to see what you made!

#76

Well, I think that LPC à la Speak and Spell has already been experimented... I think it is the best candidate because LPC analysis already exists.

#77

The "CosCos" oscillator tends to produce clicks in monophonic mode because note changes create phase discontinuities :frowning:

Help->library->community->tiar->osc->SyncPM2_2

is quite similar to the previous patch but it uses different oscillators.
Those are synched (trapezoid windowed) 2 operator phase modulation oscillators.

Those are glitchless :slight_smile:

#78

Right, there are no glitches anymore, but I cannot see a difference in the oscillator code. :astonished: As there are 13 audio rate sine calls in the patch I looked for a parabol approximation and found this . http://www.coranac.com/2009/07/sines/ . S3 version is fastest, but haven`t seen this asm instructions here. S4 version is still 30% faster (from the diagramm).

#79

The patches have almost the same structures. The oscillators differ.

Both "SyncPM2" and "CosCos" oscillators use sines, but they use them in different ways:

  • CosCos generates a pair of cosines phase locked on harmonics (centered on the formant pitch).
  • SynPM2 generates a phase modulated sine wave synched to an internal master oscillator.

In both objects, I use sine2t[].
sine2t[] is a 4096 point sine table embedded in the Axoloti SRAM (1 cycle access).
I don't think that polynomial approximations will be faster.

Note that Chebychev polynomial waveshaping is a very nice way to distort waveforms :wink:

#80

I accidentally opened your second coscos example instead of the SyncPM2, therefore the misunderstanding. It`s not easy, to keep track with your output. :grin: (random emoticon would be a nice feature )