Any news about a new yamaha workstation at name 2016 ??

Hi --

It’ll be interesting to see what turns up!

Yamaha do have a new tone generation chip (which includes the effects processing, BTW). The chip is designed into the new PSR-S770 and S970 arrangers. This level of chip is usually designed into the mid- and upper-range Yamaha products including the Motif, MOX, Tyros and Clavinova product lines. I suspect that it will be deployed in the new workstation (when it is released).

Gobs more info and speculation at: http://sandsoftwaresound.net/new-workstation-namm-2016/

All the best—pj

http://sandsoftwaresound.net/

Thanks pjd, very interesting indeed

Do you have any clue what the Yamaha “alien synthesizer” rumoured in the web could be about?

http://createdigitalmusic.com/2015/06/new-yamaha-synths-will-include-totally-alien-new-technology-says-interview/

The SWP70 is basically “only” AWM2 brought to the next level, if I get it right? And btw, do you know what’s inside the Reface keyboards? (DSP wise)

Hi Faldin --

I haven’t seen any “alien technology” in the patent filings. :-) I read that article, too, and kind of shook my head…

I wish I knew what is inside of the Refaces. The CP and YC are most likely some kind of AWM engine, but the DX and CS—that would be most interesting.

Sometimes I wonder if the VCF/VCA part of AWM could be strapped to FM algorithms or emulated oscillators and, voila, FM and virtual analog. Fun to think about, anyway.

All the best—pj

Markus Fuller has opened up a Reface YC but I don’t think he identified any of the key Yamaha chips.

https://www.youtube.com/watch?v=wy0lpXWAv68

Hey, hey, thanks for the link!

If you watch the video, the DSP chips are hidden under the heat sink (that big metal plate). Not enough other information to infer what’s hiding there.

One very interesting point is the ARM FM3 chip. Yamaha have typically used a Renesas H8 or SH-4 SOC for microcontroller applications like this. This is the first time that I’ve seen an ARM microcontroller!

All the best, pj

The processor on the Reface series is an ARM Cortex-M3 https://en.wikipedia.org/wiki/ARM_Cortex-M#Cortex-M3 -
an embedded Microcontroller unit (MCU) version of ARMv7 architecture, probably a Toshiba TX03.

I’d think Yamaha probably needed better development tools and compilers for their proprietary OS software, as the Reface does not use Embedded Linux, unlike the Motif XS/XF and MOX/MOXF.

The Reface series are most likely based on the “Spectral Component Modeling” (SCM) synth engine, first used in the CP1/CP5 and CP4/40 series.

As explained by Yamaha’s Athan Billias, SCM actually means two separate concepts:
1) “spectral modeling” - where the sound is analyzed with Fourier transforms and broken down to essential harmonics each representing some distinct spectral component, which are stored using parametric data or a large number of basic sample layers; and
2) “component modeling” - where these individual spectral components are re-combined with a kind of additive synthesis, and then shaped using filters/envelopes and effect processors (which can be physically modeled using Virtual Component Modeling (VCM)) as in standard subtractive sytnehis, to recreate the original sound.
PianoWorld Forum - Spectral Component Modeling #post2148653

The tone generator chip is probably the latest SWP70, http://sandsoftwaresound.net/tag/swp70/

Even though the SCM engine can still use traditional multilayered samples (as in acoustic piano on the CP-series), just like the older AWM2 engine, and implements the traditional substractive synthesis which shapes the sound with resonant filters controlled by LFO and envelope generators and effect processing units, it can also generate the sound “on-the-fly” using “spectral modeling” of individual harmonics - this can be used to convincingly emulate analog, FM, acoustic and electric piano, and organ.

This is NOT like the traditional “physical modeling” of the oscillator source, which uses mathematical models to recreate the properties of physical objects that generate sound in the real world, or “virtual modeling” of analog oscillators.

SCM was developed by a team led by Toshifumi Kunimoto, who is also the engineer behind Yamaha’s VCM technology of the 2000s and VP1/VL1/AN1 synthesis of the 1990s.

http://www.yamahaproaudio.com/global/en/about/interview/vcm/
Yamaha Commerial Audio - 2004/10/05 An interview with a pioneer of the VCM technology

More on this here:
KorgForums - Yamaha Reface post#652582
KorgForums - Yamaha Reface post#653334

Watching this theard with pointed ears!

Interesting comments- so the Montage could be about SCM synthesis?

However, despite reading the links and Yamaha “explanations” etc., it’s still not really clear to me what SCM actually is. Does it mean a Fourier decomposition of (sampled?) sounds, and then the sound is resynthesized again as a kind of additive synthesis of the Fourier transforms, possibly applying some modeling (amplitude, filter,...?) to those “components”? And sometimes (as in the CP acoustic piano sounds) only “conventional” samples are used as “spectral” components? For the latter, I wonder whats the difference between the existing (up to 8 elements) Motif voices…

Plus, I wonder what a “large number of spectral components” means. Would this not require a Tone Generator with large polyphony? Or is the “large number” just eight? ;-)

Hi Faldin --

You’re probably not going to get those answers here. :-)

I attached some starting points below. I suggest searching on “spectral modeling synthesis” to find the early publications and then dive into the US patent database. The Stanford patent covering the early work by Serra was assigned to Yamaha in 1996 (?).

The math is enough to choke a team of horses. After all of this, you may _still_ not know specifically what Yamaha is doing. I will say that Yamaha patents that cover parts of the new tone generator IC refer to 256 tone generation channels. Perhaps all of the polyphony is going into SCM. Only Yamaha knows for sure.

All the best—pj

My blog: http://sandsoftwaresound.net/

“A System for Sound Analysis/Transformation/Synthesis based on a Deterministic plus Stochastic Decomposition”, Xavier Serra, Ph. D. Dissertation, Stanford University, written by Xavier Serra, October, 1989.

U.S. Pat. No. 5,029,509, “Musical Synthesizer Combining Deterministic and Stochastic Waveforms”, Serra, et al., Assigned to Yamaha, July, 1996.

U.S. Patent No. 5,536,902, “Method of and apparatus for analyzing and synthesizing a sound by extracting and controlling a sound parameter”

U.S. Patent No. 8,389,845, “Apparatus for generating musical tone signals based on mixing different decay waveform data “, Kunimoto , et al., Assigned to Yamaha, March 5, 2013

“Spectral Modeling Synthesis,” Xavier Serra and Julius O. Smith, CCRMA, Stanford University

Faldin - 19 December 2015 04:15 PM

Interesting comments- so the Montage could be about SCM synthesis?

I believe so.

Does it mean a Fourier decomposition of (sampled?) sounds, and then the sound is resynthesized again as a kind of additive synthesis of the Fourier transforms, possibly applying some modeling (amplitude, filter,...?) to those “components”?

It seems like additive/subtractive hybrid:
a) first, the actual oscillator is modeled from “spectral components” using additive synthesis, then
b) the sound from the oscillator (VCO) goes down the standard subtractive synthesis path, with AEG to VCA and FEG to filter (VCF) through LFOs, and finally through effect processing busses.

This essentially makes SCM backward compatible with sample-based AWM2 systhesis and the large Yamana sample library, but future products could use more of these “spectral models”.

In the CP series, only EPs were modeled with “spectral components”; I believe the reface AN/DX also use SCM tone generator to model analog/FM synthesis, even though Yamaha is not advertising SCM in these products.

Plus, I wonder what a “large number of spectral components” means.

Yamaha US said in one of the posts that “SCM uses granular parametric data”. It’s probably parametric data for FFT window functions, so the actual digital sound will be generated parametrically in realtime, either using the RAM buffer or directly on-chip.

This is also reinforced by this statement: “There are continuous spectral changes across the entire range of velocities which are not based on sample switching, but windowed spectral data”.

Such approach shouldn’t use many hardware voices/notes, however different “spectral components” probably do use separate Elements, since these contain filter and envelope information as well as key/velocity mappings.

This all is just a wild guess though. Current SCM products are not user-programmable - there is no on-board editing and no detailed programming reference manual - so the implementation details are hidden from the end user.

There may be relevant metadata encoded within the PCM elements themselves that enables the system to reconstruct the articulations of specific instruments, as well as a simpler but still effective benefit of being able to offer oscillator level timbre shifting of both basic and complex static waveforms (like deriving PWM from a square,varying from triangle to saw, wavesequencing type effects with complex single cycle waves etc).

Instead of spec sheet marketing features like “Nine different types of modelling!”, which is for sure a good collection of soundmaking options, but that doesn’t necessarily quite get there with some emulations and is more an ‘either/or’ or layering only option (and some models are computationally expensive), a single unified scheme that does what sampling is good at, but also adds a more nuanced, granular and effective way to recomposite the natural dynamic behaviour of specific instruments (helped by the per-sample metadata), could also cover a lot of other synthesis ground by way of applying the same techniques across the board and giving front panel access to those parameters (other than the sample-level metadata itself).

Yamaha’s definitions are consistent with terminology used by Stanford’s CCRMA (Center for Computer Research in Music and Acoustics) https://ccrma.stanford.edu/

CCRMA developed and patented a practical realisation of digital FM synthesis and virtual acoustic physical modelling synthesis, and Yamaha had a license from them for both of these, so it’s only natural they are heavily influenced by CCRMA research.

CCRMA uses “spectral modeling” as an umbrella term for formant (vocoder), additive, FM, and “sinusoidal modeling” synthesis types; the latter includes “parametric spectral modeling” and “sine"/"sine+noise"/"sine+noise+transients" synthesis.

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

See also “Historical Overview of Audio Spectral Modeling”, a 2014 presentation by Prof. Julius Smith of CCRMA, pages 39-57:
https://ccrma.stanford.edu/~jos/pdf/SMS.pdf#page=39
This is an updated version of “History and Practice of Digital Sound Synthesis”, a 2006 presentation which talks about both spectral modeling synthesis on pages 25-48 and physical modeling synthesis on pages 49-84:
http://www.aes.org/technical/heyser/downloads/AES121heyser-Smith.pdf#page=49
Prof. Smith also hosts his full online book “Physical Audio Signal Processing for Virtual Musical Instruments and Audio Effects”, which can be found here:
https://ccrma.stanford.edu/~jos/pasp/
You can also look into “Spectral Modeling Synthesis: Past and Present”, a presentation by Prof. Xavier Sierra http://www.dtic.upf.edu/~xserra/ who developed his Ph.D. thesis in Stanford’s CCRMA and is a co-author of Prof. Smith on several papers relating to spectral modeling.

So when Yamaha say “parametric spectral data” and “spectral windows” etc., I hold that it actually means parametric data created by analyzing the original sound with Fast Fourier Transform to separate it into individual spectral components. It’s basically a form of very effecient lossless compression, allowing the synthesis engine to exactly recreat the original sound at playback time using window functions based on this spectral data.

I would expect this approach would reduce the storage/memory requirements to represent a “sound,” so there should be more sounds available. Depending on Yamaha’s implementation, this could mean either more voices, better quality voices consisting of more/higher quality samples, or both.

As for price, I would expect Yamaha to keep the price comparable to a Motif XF when first released, but to provide more capability for that price.

Of course, all of this is pure speculation.

NAMM is only 3.5 weeks away, so it won’t be long before we get to know more about what Yamaha has been up to.
Maybe a teaser will be launched shortly to add some fuel to the fire of speculation.

We are talking about very recent trends - there were only a few research papers on “Spectral Modeling Synthesis” up until now, and I’ve found no sound demos.

However if the reface DX (FM emulation) and reface AN (analog emulation) are any indication, the SCM (Spectral Component Modeling) engine is capable of very faithful emulation of real-world synth engines.

Here is what I can think of.

Pros:

1) Spectral components (harmonics) are stored as parametric data (i.e. as a series of parameters to some pre-defined mathematical function), so the original sound can be restored exactly as it was sampled (or mathematically modelled) with floating-point precision, given enough components (harmonics) are detected during FFT pre-processing;

2) Velocity can have a continuous effect on the timbre - there are no separate velocity layers and the transitions are smooth;

3) Storage requirements are greatly reduced, maybe 1/1000 to 1/1,000,000 of the original uncompressed sample;

4) Greatly reduced computational complexity, comparing to physical modeling (where high processing performance requirements have a direct negative effect on polyphony): any complex sound can be made of finite number of spectral components which use standard cheap Spectral Component Synthesis hardware;

5) Many-dimensional synthesis control can be possible (as opposed to just pitch, velocity and aftertouch in subtractive sample-based synthesis), much like physical modeling synthesis (VL) or articulation modeling (SA) but without the high performance and storage requirements

6) Less effect processing units are needed, since some of the effects can be emulated with additional spectral components;

7) Good price/performance - hardware should be inexpensive to implement, yet it would have very high quality sound.

Cons:
1) None of the benefits above apply to legacy sample-based programs created for AWM synthesis model;

2) Not user-programmable - designing your own programs requires very specific professional skills and proprietary Yamaha software, most likely not possible with onboard user interface;

3) User cannot create their own models ("instruments") beyond what is available in the synth OS - unlike physical modeling which is fully parametric;

4) Spectral components that make up sound models ("musical instruments") are self-contained and not extensible - onboard editing of programs is probably restricted to adjusting some pre-set parameters and/or standard continuous controllers as allowed by the designers of the model.