Forum index > Feature request

EssenceFm as a sampler?

DX100
Hi,

Do you at Kodamo ever think about making this amazing machine be able to be used as a sampler?
It would just be beyond incredible...FM and samples with this kind user interface...one can dream..
looneytoonz
It might be cool just to have transient samples that could be combined with 6-op steady-states (i.e. Linear arithmetic, like a Roland D-50).
KODAMO
That would be fun. But it creates some technical challenges, usually in any device the memory can be either small and fast (that's what we have in the EssenceFM), or big and slow.
Samples need the 2nd option, which is OK for playback but doing real-time FM synthesis with them isn't easy without affecting performance.

Also needs a totally reworked GUI for sample editing
DX100
Even a very small amount of sample space would be totally awesome, maybe making it really lofi would help cpu, like 8 bit or even lower?
The Essences excellent ui + samples would be a match made in heaven.
DX100
Personally I think you should re-release essence but as a sampler, the ui would suit so well, with its performances and patches,
if you added filters and few other things it would be one of the best hardware samplers ever.
Strider
+1000
KODAMO
What specifications/features would make it good enough as a hardware sampler ?
We noticed some demand for modern hardware samplers, but usually people end up wanting something almost as powerful as a computer, at a lower cost, which is impossible.
Colin Muir Dorward
I think because of the way patch/performance setup works, you could get away with a very simple sampler (rampler?) voice. Only things that come to mind are a few extra filter-types, and a well-planned data/sample pool management system. Fine control of tuning and a wide re-pitch range would get me as far as I'd like to go! I wouldn't be looking for time/pitch stretch algos, or any beat-slicing in a machine that was also my go-to FM synth.

Of course if you re-built the entire OS and developed an entirely new sampler product, then I think that's a different conversation.
DX100
A few things.. the way I'd use samples.

Using audio transients, bit like D50, maybe combined with FM in a patch.

Short wave samples looped, like vocal/acoustic sounds, maybe combined with FM in a patch.
bi-directional and normal looping would be a necessity if there is a very limited memory.

Using the round robin feature with the above ideas to get more sonic variation, cycling thru different samples.

LoFi drum kits built with lots of short samples, 1 band EQ would help here, existing filter would be good for sounds that are too bright, individual outs could be used though and mixed externally.

Way to backup samples and reload.
deastman
My favorite example was using Seer Systems Reality, the early software synth by Dave Smith. It could load any arbitrary, long user sample into an operator. FM modulation between multiple samples in a complex algorithm can produce some really interesting results. One time I loaded the Windows 98 (95?) startup sound into several slots and had them interact with each other, which resulted in an amazingly long, evolving ambient patch. Brian Eno would have been proud! I suppose something like that is beyond the technical limitations of EssenceFM, but it would be nice...
dstengle
I don't do much want to blend or edit samples. Simply being able to load and play multisamples would be awesome to take advantage of all of that multi timbrality. I have a modx and I love having a big sound library but I'm replacing it with the essence fm because the workflow is just too painful. I have other multi timbral romplers but it would be great to have it all in one small package.
Teomi
I will add my thoughts as I have been using all kinds of hardware samplers in the past and now I'm mostly using Software-based samplers. For me, the most musical hardware sampler was the Ensoniq ASR-10. When you think about it, there is a lot of similarities in approach between the EFM and the Ensoniq design. So I think that if there could be a special memory buffer reserved for samples, maybe 128MB that would be enough for creative sampling.

Some thoughts about possible features:

- Drum samples and note mapping similar to how the EFM is currently handling it. The ability to layer samples with FM would be very powerful.
- Basic looping functions with access to the filter/LFO and manipulating playback speed and direction.
- something similar to Ensoniq's Transwave technolgoy which is a different take on wavetable synthesis - splitting a sample to 128 cycles that can be individually played and scanned using a modulator.
- Resampling through the fx or through the main output
- Ability to use the voice sequencer with loaded samples.

Something tells me this is not completely out of scope with what is possible with teh EFM but of course I may be wrong.
Just thought I would offer some musical features :)
Failed Muso
I'm not so keen on the idea of adding sampling to the EssenceFM. It's an out and out FM synth, and I wouldn't want anything to detract from that. That said, when Yamaha combined AFM and AWM on the SY/TG77 and SY99, that was incredible, but it was designed as such from the ground up. So I'd definitely be interested in a new instrument that combines sampling and FM, but I don't think the EFM is it šŸ™‚

Also worth mentioning that the SY22/35's and TG33's way of blending AWM & FM using vectoring was pretty cool. A more up to date and powerful implementation of that might be cool?
Petajaja
Imo, the best thing would be to have the sampler as an alternate firmware so that it wouldn't have to share resources but I also think this is a huge thing to ask of Kodamo, if it were possible though, it'd had such incredible value to it as an instrument.

I also think if it had similar capabilities to something like the ASR10, it'd be perfect. In software, TAL sampler is my favorite and again, something in that vein would be amazing
Soarer
Also not too keen on this here. I think the EFM shines at being an FM synth and it's capable of making such a great variety of sounds already. It would probably make things messy and use up resources. I'd much rather see some wavetable features perhaps with audio file import.
Benis67
Regarding the possibility of using the samples, for certain timbres a single MegaByte used for example by D-50 could be enough.
The latter is famous for its "bell-like" and "blowing" tones, all obtained thanks to 8 looped samples of 2048 Samples (Spect1, Spect2, .. Spect8) which are interpolated over the entire range of the keyboard. They are sound spectra obtained by FFT in which all the first harmonics have been eliminated (excellent work by the legendary Eric Persing).
DJ Fresh
Plus one for sampled transients the d50 had this šŸ˜Ž
KODAMO
Transients are fairly easy to create with FM synthesis, eg. breath noise for a flute or the extra rich harmonics on a violin attack, also it has the huge advantage over samples that you can modulate it depending on velocity or anything else.
If you really want to use sampled attacks you can use any waveform slot (which are 512-sample long) and use a very short envelope to prevent it from looping.
korhanerel
People, please do not try to turn the EFM into a workstation. It is a great FM synth and let it stay that way. If you need a sampler, buy a sampler. Or get an iPad, which has amazing sampler apps for it.
Jazz Paladin
Getting ready to buy mine in two weeks or so, I wouldnā€™t want it to sample everything under the sun, but for the type of music I do and associate synths with, I think some piano (not a whole lot of velocity layers needed) would be nice, even if lo-fi as suggested above, as would some real snare, hi hat, kick and cymbal. Again not a ton of layers needed, just to add the usual consistently heavy strikes for EDM

I do think a nice sequencer with 480 ppqn would be nice too, as I plan on using an old Yamaha qy75 to sequence. Would be cool to have things integrated.
gearhound22
A simple sampler would be so appreciated! Hope something can be done in the future šŸ˜Š
Colin Muir Dorward
Hijacking this thread a little, but is there even one rack-format sampler in production today?
Keith

Hijacking this thread a little, but is there even one rack-format sampler in production today?


Yes the Waldorf Iridium is a rack format sampler, at least I rack mounted mine with a VESA monitor mount. I believe a rack kit is also available now. It is a wonderful synth but lacks the UI genius of the EssenceFM IMHO.

There are many simple module sized hardware samplers though and some are beautifully executed. They offer portability and/or immediacy in exchange for reduced features and less interface sophistication compared with modern software samplers. Some of the best samplers now appear to store something like a windowed FFT with control sequences to represent the changing levels and phase along with a short ā€œnoiseā€ sample for the transient. There are huge benefits to this but it also must be complex if you wish to input modulation changes in real time. I donā€™t know if they are synthesizing a filter bank for decoding but it seems very desirable to do that and also difficult to get right in software. Most samplers donā€™t have this latency-reducing optimization, but they pay a price in the flexibility of routing modulations because of latency if I am observing correctly.

I love the FS1R. It has a different way of creating transients: those eight additional noise operators. It sounds like witchcraft when you hear about it but I am finding it easy to use that feature to get beautiful and realistic sounds. The EssenceFM is even more powerful in the same way because the transients can be layered in one or more additional voices. The FS1R is not really more powerful in this respect. The noise operators are specialized for creating transients though and there is some magic in providing such a specialized feature and priming the users with the suggestion that they should use it in a particular way.

The best additive synthesizers seem to be like samplers also, but give access to the spectral envelopes directly. They still have to solve the problem of storing transients as samples or providing some way to synthesize them.

One way to view FM is as an efficient approximation to additive synthesis. I have a desire to write a program to create EssenceFM patches from samples. I donā€™t know how important it would be to other people but it seems not too difficult with current computers. You can imagine a greedy algorithm that makes a 6-operator approximation and then iterates on the residual error. If the patch could be programmed from a DAW plugin it would seem to create a capability that nothing else currently hasā€¦ because the FM voices of the patch could then be edited and customized. None of this would be so interesting except that the EssenceFM has made FM editing so much easier in a form where many voices can be produced.

I wonder if something with this flavor could meet the needs everyone has for realistic sampled sounds that can be used in new ways. It feels like a fun and useful project but without huge impact for the EssenceFM. But a future synth could then provide some means of manipulating the time line of the individual voices at the patch level. That seems exciting to me.

The format capabilities of the FS1R are another exciting feature. The FSEQ facility (multichannel mod sequences) in the FS1R is what makes formant voices work. They are at the FS1Rā€™s equivalent of the patch level and then affect the spectrum of the voice operators in a coordinated way. Those FSEQ modulations in the FS1R appear to have been created by reverse-engineering samples. It might have been done by hand or by an algorithm in the FS1R. It makes me wonder if it would be possible in the EssenceFM to provide waveforms from the system store of editable waveforms to a patch for use in the voice mod matrix. That would get us well into FS1R territory, and if a sample to patch creator were possible from the DAW then the patch could be manipulated in satisfying ways.

Those are just some stray thoughts and I hope it is ok to put them into this conversation.

Keith
KODAMO
Generating EssenceFM sounds from samples would be super-awesome šŸ‘
It may be possible to start from a totally random voice which generates slightly randomized children at each iteration. You need an algorithm to give a score to a voice depending on how close it sounds to the original. If the scoring system works well, you may end up with a good approximation of your input sound with maybe thousands or millions iterations. Even if it's not working as expected, it might be a very inspiring and creative tool.
Keith
Iā€™ll try this. When I get home in a couple of days Iā€™ll look at it carefully and try to figure out the mechanics.
hwk
for this, aphex twins/dave griffiths "midimutant" comes to mind.
quote from the magpi article: "...you give it a sound, and it aims to recreate it on the Yamaha TX7"
some material can be found here: https://fo.am/activities/midimutant/
but as far as i know, the promised release of the source code never happened.
the essence FM would need a thorough sysex programmability for this though...
KODAMO
Sysex documentation is available here: https://kodamo.org/efm_sysex.php
Colin Muir Dorward
"...you give it a sound, and it aims to recreate it on the Yamaha TX7"... or EFM

I don't understand how this can work without somehow changing into a speedier domain. This is way outside my knowledge, but seems like you'd need to run everything at superspeed for it to work properly. Imagine a long pad sound which lasts one second, if it takes 1x10^6 iterations to "learn", then you're committing to 278 hrs of processing time. OK so how about 1000 iterations? That's still close to 20 mins. This is based on the assumption that every iteration needs to be produced once, in real time. Maybe I'm missing something here.

But if you had an EFM, TX7, etc model running on a fast computer, you could run it in superspeed/supersonic and reduce that timeframe. I guess I just can't see how using a HW device helps here.
hwk
of course, this is in no way meant as a realtime process - considering the limitations imposed by midi/sysex communication as a serial protocol, that would be plain impossible anyway.
but the whole idea and research behind it is a quite fascinating take on the analysis/resynthesis branch of research.
and sometimes the deviations and unexpected intermediary steps occuring along the way might be even more interesting than the sounds generated to mimick the "original".
(e.g. google deep dream).
@kodamo: thanks for reminding me to look again at the sysex specs. they seem to have evolved considerably since i looked last time šŸ˜€
Colin Muir Dorward
Deepdream or any of these other AI-powered image-generation tools can take as fast as they want to look at each iteration. But for an AI to listen to listen to the output of a synth, you're stuck with realtime playback from that synth, so each iterative "listen" takes as long as your sound is. You can't tell EFM to playback at x1000 speed, and even if you could, you'd find another bottleneck in the band-limited analog signal path to your AI.
So.... I still don't understand how this is a realistic option using HW.

If you did it in software, you can write code to enable x1000 speed playback, then it can work.

Or... maybe it's possible with HW because you can get the job done in just 10 or 100 listens? I don't know much about AI!

Great convo, thanks for the brainstorm!
Keith
There are a few ways to do it. Audio from the EFM is the cleanest, but using the DX7 subset, audio can also be rendered in software as an optimization, borrowing source code from one of the existing open source FM synths maybe. There is more project complexity to deal with the audio and MIDI interface than there is for the parameter matchingā€”if we take a simplified approach of modeling one spectral peak at a time with a short operator stack.

There is independent value in having a software module for EFM sysex so Iā€™ll give the audio interface approach a shot first. It means having to be smart (and limited) about the parameter search at first. I agree that even in the failure case something good can be done with the parts. Iā€™m also thinking it is better to start by approximating a static complex spectrum instead of trying for moving/changing peaks at the beginning. Iā€™ll definitely look at the prior work that I hadnā€™t known about.
hwk
@colin: deep dream is fast, but training the model behind it is not. that relies on many, many iterations of feeding training material and twisting the results.
(if you want a quick run at training a simple model recognizing audio try: https://teachablemachine.withgoogle.com/train/audio
its kind of fun...
so for each "real world" sound you want the EFM to mimic, you have to train a model first, extract relevant features, label them and feed them to a model of the EFM, which has been specifically crafted to understand those features.

the approach aphex twin/dave g. took is different (if my understanding is correct): extract features of a "real world" sound, start with *any* sound on the EFM, compare and use a genetic algorithm with a fitness function to gradually tweak the EFM sound until its features match as best as possible with the features of the "real world" sound.
this takes (lots of) time - on the other hand i donĀ“t understand, why it even should be realtime?
but maybe i am misunderstanding the whole point (i am not so much interested in having the EFM play samples).

Reply