Announcing the launch of ULTRA

[ultra_audio]

Ableton Log:
"2026-02-08T18:32:45.119858: info: VST3: couldn't get controller state of Ultra: not implemented"
error in Ableton Live indicates a failed communication between the plugin's processor and controller, often caused by buggy VST3 implementation (frequently using the JUCE framework). Fixes include updating the problematic plugin

Hmmm I'll PM you...

[teilo]

Depending on the sound, it might be better to turn the attacks off. If it's introducing a delay in one channel, it would likely exist in the source sample also.

You can get release behaviour. Set the loop to section in the middle of the sound. When you release it, will exit the loop and play the cycles at the end

I think my mistake was using the automatic mode to process the samples. All things considered it would probably be better to process each sample manually.

Thank you for the hint about release behavior. That makes perfect sense.

[ultra_audio]

Depending on the sound, it might be better to turn the attacks off. If it's introducing a delay in one channel, it would likely exist in the source sample also.

You can get release behaviour. Set the loop to section in the middle of the sound. When you release it, will exit the loop and play the cycles at the end

I think my mistake was using the automatic mode to process the samples. All things considered it would probably be better to process each sample manually.

Thank you for the hint about release behavior. That makes perfect sense.

Auto mode is great for a lot of types of sounds, but it can't do everything. Things with strong repeating cycles work best in auto mode.

[Tj Shredder]

I tried a collection of clavecin samples and the result was a bit weird. The sustain sound was kind of wobbly. I think the reason was that it has slightly detuned strings, resulting in unclear pitches. It seems it did not create pitch curves, which probably could correct that. Maybe I did something wrong while importing?
I am flashed about that whole idea of turning a sample into a wavetable, simply by taking the zero crossing complete cycles, and correct them accordingly in pitch and amplitude. This is new, simple and effective at the same time…

[teilo]

Depending on the sound, it might be better to turn the attacks off. If it's introducing a delay in one channel, it would likely exist in the source sample also.

You can get release behaviour. Set the loop to section in the middle of the sound. When you release it, will exit the loop and play the cycles at the end

I think my mistake was using the automatic mode to process the samples. All things considered it would probably be better to process each sample manually.

Thank you for the hint about release behavior. That makes perfect sense.

Auto mode is great for a lot of types of sounds, but it can't do everything. Things with strong repeating cycles work best in auto mode.

Working with this more now, I can see some improvements that would be helpful in editing attacks. Auto-snap to zero crossovers would be super helpful to avoid clicks when editing attacks. Not sure if there would be some way to do phase alignment similar to the "Align Phases" functions on cycles on the Attacks themselves.

[Tj Shredder]

I tried a collection of clavecin samples and the result was a bit weird.

To answer myself, yes, I had to tick the pitch to do it correctly.
Now I have the problem, the only way to define the tail seem per threshold. I would like to move the tail markers by hand or simply shorten the sample while I finetune the import. The easiest would be to have a a choice for the tail. Either threshold or time...

[teilo]

How does one export a patch so that it can be shared? Not all of us want to sell our patches, but so far I have found no way to do this.

[liquidsound]

Maybe the Share button in your User library?

[ultra_audio]

How does one export a patch so that it can be shared? Not all of us want to sell our patches, but so far I have found no way to do this.

you can right click the header. we are going to make a file based sharing system at some point but that works for now

[Tj Shredder]

Playing around with the automatic mode, I often get pretty glitchy results, also it seems that a noisy recording confuses the detection.
I was thinking how this works, or maybe could work. Obviously I got huge jumps in the pitch envelope with high pitched decaying sounds. Listening to the original I heard the low noise floor getting louder than the pitch.

How would i do it? Assuming we only want pitched material like that clavecin, I would first get the perceived pitch of the sample. I would stick to it as it determines the range of one cycle. Then I would cut it at zero crossings to get clean cycles and correct the pitch curve according to the slightly varying size of the cycles. Cut it up until the end or to the now active pitch detection within the sample has a drop of more than an octave. Certainly there is more noise than pitched sound there.

I normalize the slices and adjust the levels according the original amplitude envelope. Now I can eliminate slices according to their similarity. Instead of simply telling it how many slices it should keep, I would rather have a similarity threshold to reduce it. If the similarity between far away slices is close enough I can eliminate those in between. As each slice has its own pitch and we glue them at zero crossings this should (theoretically) keep it smooth.

This would also need a third kind of envelope. The speed it runs through the cycles. some cycles need to be played multiple times, the original fluctuation would be still in the (smoothed) pitch envelope. The speed of the advancement in the wavetable would also be a welcome modulation target or extra parameter. I could lengthen very short decays easily...

The noise parts could actually be put back in with the noise oscillator of Ultra. Another method to get rid of the noise would be to split the sound into two frequency bands, all above the perceived pitch and all below consider as noise who's shape can be applied to the noise oscillator.

That way Ultra could completely replace a sampler with far more options to tweak the sound and much less space resources. Way fewer samples would be necessary to get realistic results...

I don't know how close this is to what is actually implemented. I was just contemplating how this idea could work...

[Download SOphist]

Playing around with the automatic mode, I often get pretty glitchy results, also it seems that a noisy recording confuses the detection.

me too. I started all enthusiastic with this synth, focusing (like most of us, I guess) the attention on the ultra wave OSC but couldn´t really get nothing useful out of it and my short attention span is getting tired.
tried to recreate pianos, acoustic guitar, harpsichords,... if there is not a strong repeating pattern for the OSC to cycle through all I got was very unnatural, glitchy sounds, full of artefacts. I´m mediocre at synth programming, so maybe that´s that. eager to be proven wrong, though.
OTOH one can easily get better results with Icarus (maybe I shout have demo´d it...), Myth and Sumu.
too bad as the concept is quite interesting, genuinely new and the GUI is just stunning. will wait to see how this one develops.

[Download SOphist]

- double post -

[Tj Shredder]

I wouldn’t call my results unusable, I also think the basic idea is brilliant and much more interesting in terms of tweakability of the result.
As I layed out, the process can probably be improved. I also thought for filtering out noisy parts, one could use a fourier transform on the complete sample. Not fft, but taking all. That would certainly get a pretty precise pitch estimation and as its a non-realtime process its independent of the length of the sample.
If we also get a better control than a threshold for defining the tail, we are close to what I dream of… A sound starting point within a synthesizer that sounds authentic to its sampled source…

[teilo]

Maybe the Share button in your User library?

That's not an export. That's sharing a patch using their cloud service, and it's limited to 10 people.

[teilo]

Playing around with the automatic mode, I often get pretty glitchy results, also it seems that a noisy recording confuses the detection.

me too. I started all enthusiastic with this synth, focusing (like most of us, I guess) the attention on the ultra wave OSC but couldn´t really get nothing useful out of it and my short attention span is getting tired.
tried to recreate pianos, acoustic guitar, harpsichords,... if there is not a strong repeating pattern for the OSC to cycle through all I got was very unnatural, glitchy sounds, full of artefacts. I´m mediocre at synth programming, so maybe that´s that. eager to be proven wrong, though.
OTOH one can easily get better results with Icarus (maybe I shout have demo´d it...), Myth and Sumu.
too bad as the concept is quite interesting, genuinely new and the GUI is just stunning. will wait to see how this one develops.

There are lots of ways to work around this. First, use Automatic mode to import the sample, but don't accept every cycle. Set a fixed number of samples, and have it distribute them evenly. Next edit the result, eliminating samples you don't actually need. Finally, select all the samples and use the cycle functions to phase-align them. This part is very important and usually eliminates most of the glitchiness you are experiencing.

If you still don't like the results and need to pull back in more cycles, edit the corresponding table again, and re-import your original sample back in (notice the Import section on the right. Just drag-and-drop). You can now select additional cycles from the sample, and add them where you wish. Don't forget, if necessary, to phase align everything again if you do this.

Also note the filter in the cycle editor. A bit of low-pass on your cycles goes along way toward eliminating noise.

The biggest weakness at present is the transition from attack to your cycles.You will often get clicks because there is no smoothing applied from attack to your cycles. You will need to pay close attention to the crossover from attack to cycles and make sure your attack sample ends at an appropriate point on both channels. You can adjust this by zooming way into the attack and deleting small segments. Finally, you can import your sample back into the patch and add a new cycle from the end of your attack to the very beginning of your cycles, making sure to leave space between that new cycle and the remaining cycles. This will allow the engine to transition smoothly from that last sample from the end of your attack to the remaining cycles.

Pay close attention to the volume levels of the attacks, and of your cycles. It's often necessary to adjust this in your amplitude curve so that you get consistent volume from the attack to the cycles.