Been messing around with an old modular synth and a neural net to pull out hidden sonic textures, think that’s something you’d dig?

Sorry, that net’s locked in my private lab, but if you want to try a similar setup I can drop a tutorial script. Just ping me the details and I’ll point you in the right direction.

I’m not allowed to hand out the net, sorry. I can sketch out the architecture though – it’s a custom neural net built on PyTorch with a gated RNN that ingests polyphonic MIDI and outputs a wavetable. I’ll send the code, but you’ll need a GPU to train it. Let me know if that’s cool.

Here’s a stripped‑down prototype you can spin up on any decent GPU. ``` import torch import torch.nn as nn class TimbreNet(nn.Module): def __init__(self, input_dim=128, hidden_dim=256, layers=4, dropout=0.2): super().__init__() self.lstm = nn.LSTM(input_dim, hidden_dim, layers, batch_first=True, dropout=dropout) self.out = nn.Linear(hidden_dim, 128) # wavetables self.activation = nn.Tanh() def forward(self, x): out, _ = self.lstm(x) out = self.out(out) return self.activation(out) model = TimbreNet() ``` Train it on a 32‑step MIDI sequence (each step is a one‑hot 128‑dim vector) and it will spit out a wavetable that you can feed into a wavetable synth. GPU: any card with at least 8 GB VRAM works fine—an RTX 2060, a 1080 Ti, or even an older GTX 1070 will do. If you can squeeze a 3090 or 4090 out of the junkyard, the training time will shrink a lot, but 8 GB is the bare minimum. Happy glitching.