I've just started using NAM in my own music and it's seriously impressive
it looks like the RTneural implementation is using a pretty standard wavenet setup that works on a sample-by-sample level so it should be pretty easy to port over once it's ready, basically just a "more specific" neural node, it might even already be possible to use the regular neural node for it if you know how to import the weights
@clevername27 No sorry, I think I ended up removing the SN network entirely
@HISEnberg NEATBrain doesn't use any crossfading shenanigans, it just loops a single cycle at its fundamental frequency to simulate a wavetable synth
@d-healey I just realized a pretty decent workaround is to assign relevant GUI parameters to CC messages, that way the end user can also right-click and bind/unbind whatever they want for automation AND if you stick to conventional CC (like assigning Attack to CC73), most workstations will label them automatically.
@d-healey Like I said it's not a HUGE requirement, I like to end some songs with a long fade guitar chord to transition into some other sections or whatever, I can always just max the Sustain and use volume automation before the amp :)
So you would have to compile Rhapsody with
HISE_MACROS_ARE_PLUGIN_PARAMETERS=1 HISE_NUM_MACROS=8 (or whatever)I've implemented this for a particular project, but I tried to enable the use-case of Rhapsody (or other Full-Instrument players) with this too so let me know if there are some issues.
@d-healey have you gotten around to this? ran into this issue again today
@iamlamprey ah just found this comment at the top of the cpp file:
/*//
This is an automatically generated GUI class created by the Introjucer!
Be careful when adding custom code to these files, as only the code within
the "//[xyz]" and "//[/xyz]" sections will be retained when the file is loaded
and re-saved.
*/
edit: appending the code there compiles, but crashes HISE when I add an envelope. it's not too important of a feature so i don't really care to test any more ¯_(ツ)_/¯
@orange I only see these:
addAndMakeVisible (decaySlider = new HiSlider ("Decay"));
decaySlider->setRange (1, 20000, 1);
The rest of the results seem to be related to Scriptnode, dsp unit tests or otherwise commented out
@orange Just tried modifying it and it doesn't seem to change anything (limit still capped at 20000), plus I'm not sure if source code changes will propagate down to the exported Rhapsody Library, since I think that depends on the HISE settings that originally compiled it.
Unless I'm wrong? @d-healey
Is this possible? Can't find a preprocessor or anything for it, ideally I would avoid ScriptNode because it's a Rhapsody library and some networks can be a bit weird in it
@d-healey are sub 0.5hz LFOs supported in Rhapsody? also should the default not just be 0.01hz? Is there any reason why we should need the preprocessor flag?
Another error related to this line:
var importedSamples = Sampler1.asSampler().importSamples([path], false);
Exception thrown: read access violation. std::_Atomic_address_as<long,std::_Atomic_padded<int> >(...) returned 0xFFFFFFFFFFFFFFFF.
This Atomic file is part of VS2022 so I'm in uncharted waters now I think 
okay nope, still crashing, now with a different jassert:
// if this hits, you're using multiple writer threads. consider the ScopedMultiWriteLock instead
jassert(lock.writer == thisId || lock.writer == i);
@Christoph-Hart is this a drive-related error? its an old dying HDD so I'm assuming it's related
Okay I've moved Sampler1.setAttribute(Sampler1.RRGroupAmount, NUM_ROUNDROBINS); out of the internal function so it's only being called once, silly me 
@dannytaurus Yeh I pasted that hastily, it's part of a larger function but includes the important parts.
I just debugged it and got a seemingly unrelated error so I'm thinking the problem is actually due to this line:
Sampler1.setAttribute(Sampler1.RRGroupAmount, NUM_ROUNDROBINS);
I'll keep digging
I have a backend script that has to import some 1600 (small) samples, set various sample attributes (root, velocity, RRgroup etc) then save the sampleMap but every time I run it it crashes HISE.
What are the best practices for batch importing like this? Tried:
killAllVoicesAndCall which made it worsesamples = FileSystem.findFiles(samplesFolder, "*.wav", false);
for (i=0; i<samples.length; i++)
{
// Get sample name as string
prefix = "{PROJECT_FOLDER}" + samplesFolder.toString(1) + "/";
name = samples[i].toString(3);
path = prefix + name;
// Set the RR Group Amount & Import Samples
Sampler1.setAttribute(Sampler1.RRGroupAmount, NUM_ROUNDROBINS);
var importedSamples = Sampler1.asSampler().importSamples([path], false);
for (s in importedSamples)
{
sT = s.get(Sampler.FileName);
idx = sT.indexOf("root") + 4;
subString = sT.substring(idx, sT.length);
idxEnd = subString.indexOf("_");
subString = subString.substring(0, idxEnd);
rootNote = Math.round(subString);
//s.set(Sampler.LoKey, 64); // doesn't work
// Low Key Fix
for (x = 3; x < 5; x++)
{
s.set(x, lowKey);
}
s.set(Sampler.Root, rootNote);
s.set(Sampler.HiKey, highKey);
// etc
Sampler1.asSampler().saveCurrentSampleMap(sampleMapName);
}
@iamlamprey Okay a simple workaround is to just call another inline function inside the Loris one:
inline function adjustHarmonicGain(partial, freq, target, tolerance, multiplier)
{
// Adjusts the gains of any partials that lie within the tolerance of a target frequency
local spectralDistance = freq > target ? freq - target : target - freq;
if (spectralDistance < tolerance)
partial.gain *= multiplier;
}
inline function myCoolFunction(obj)
{
adjustHarmonicGain(obj, obj.frequency, 440.0, 50.0, 0.0); // works
}
lorisManager.processCustom(file, myCoolFunction);
Yeah that works for regular functions but Loris is a bit different:
// works
function myCoolFunction(obj)
{
// even Console.print(obj.frequency) here crashes HISE
obj.frequency = 440.0; // works
}
// doesn't work
function myLessCoolFunction(obj, f)
{
obj.frequency = f;
}
// obj is already defined by a separate Loris process:
lorisManager.analyse(file, f0); // this is our obj, we don't need to declare it or pass it to the function
// works
lorisManager.processCustom(file, myCoolFunction); // no parameters, just the function name
// breaks
lorisManager.processCustom(file, myLessCoolFunction(obj, 440.0));
// standard lambdas also break
lorisManager.processCustom(file, function [obj, 440.0](myLessCoolFunction)); // "can't capture anonymous expression"
It's not a huge deal, I can just make a few variants of the function ¯_(ツ)_/¯
Unless I'm not shaping the data object correctly for the filter?
// using time in seconds (default)
// pitch shift works (example from the Loris docs)
var pitchShiftData = [[0.0, 0], [0.2, 0], [0.4, 300], [0.4, -300], [0.6, 0]];
lorisManager.process(file, "shiftPitch", data);
// doesn't work, just mutes the output
var filterData = [[0.0, 0], [0.2, 0], [0.4, 1], [0.4, .5], [0.6, 0]];
lorisManager.process(file, "applyFilter", filterData);
// also doesn't work, same result
var filterData = [[0.0, 0], [0.2, 0], [0.4, 20000], [0.4, 10000], [0.6, 0]];
lorisManager.process(file, "applyFilter", filterData);
Edit - here's the source code for it:
if (command == ProcessIds::applyFilter)
{
juce::ScopedValueSetter<TimeDomainType> svs(options.currentDomainType, TimeDomainType::Frequency);
auto env = createEnvelopeFromJSON(data);
for (auto& l : list) // for each partial
{
for (auto& p : *l) // i assume this is the envelope
{
for (auto& b : p) // for each breakpoint in the envelope
{
auto freq = breakpoint_getFrequency(&b); // "returns the frequency of the specified breakpoint"
auto gain = linearEnvelope_valueAt(env, freq); // "returns the interpolated value of this linearEnvelope at the specified time"
gain *= breakpoint_getAmplitude(&b); // "returns the absolute amplitude of the specified breakpoint"
breakpoint_setAmplitude(&b, gain); // "assigns a new (absolute) amplitude to the specified breakpoint"
}
}
}
destroyLinearEnvelope(env);
return true;
}
Shouldn't the linearEnvelope_valueAt function be taking a time value, not a frequency? Or is that being handled by the ScopedValueSetter?