SSNN from sonicLAB is a live audio processor built around a spiking neural network running 960 neurons across 32 layers, where each spike from those neurons drives eight synthesis engines in real time.
The architecture sets it apart from AI-powered audio tools built around pre-trained models. SSNN does not load a model or draw on offline training data. Instead, it listens to incoming audio, captures the spectrum profile in real time, and writes it directly into the network’s connection weights. The same signal that teaches the network also fills the per-layer audio buffers the synth engines operate on, meaning the output reflects what the plugin is currently hearing rather than what was encoded before it shipped.
Eight synthesis engines respond to spike activity. Generators include a Pulse Gen and Modal and Synaptic FM oscillators, with their behavior driven by neural network output. Audio processors (Granular, WaveShaping, FM, CombFilter, and TapeDelay) operate on a per-neuron basis, meaning each of the 960 neurons can trigger its own processing independently. The texture of the output tracks the spectral and dynamic content of the input in real time.
Users can control network behavior across a range from deterministic to stochastic, adjusting how predictably the neurons respond to incoming material. A quantization mode constrains spike timing to specific intervals, introducing rhythmic regularity into what would otherwise be a continuous and variable firing pattern. The arpeggiator maps the 32-layer network structure onto musical scales and chord progressions, giving the network’s architectural data a pitched, melodic form.
OSC spike data can be broadcast externally. The included NNnotes utility plugin converts those OSC messages to MIDI notes, allowing the network’s output to route into external instruments or other DAW tracks. A 60Hz visual display shows neuron activity in real time.
The plugin uses a multi-core engine and produces oversampled output. sonicLAB recommends a processor equivalent to or better than the Apple M1, which reflects the computational demand of running 960 neurons at audio-rate resolution.
SSNN targets sound designers working in experimental and generative contexts, particularly those looking for a synthesis instrument whose behavior shifts continuously with the audio feeding into it. The combination of live network learning, per-neuron audio processing, and OSC output puts it closer to a feedback-driven modular patch than a conventional software synthesizer.
Features
- 960 neurons across 32 layers.
- Eight synthesis engines: Pulse Gen, Modal FM, and Synaptic FM oscillators (generators); Granular, WaveShaping, FM, CombFilter, and TapeDelay (per-neuron audio processors).
- Real-time audio spectrum learning written directly to network connection weights, no pre-trained model.
- Deterministic to stochastic control over neural network behavior, including stochastic injection support.
- Quantization mode constrains neuron firing to specific timing intervals.
- 32-layer arpeggiator maps network architecture to musical scales and chord progressions.
- OSC spike broadcast output.
- NNnotes utility plugin included, converts OSC messages to MIDI notes.
- 60Hz real-time neuron activity visualizer.
- Multi-core processing engine with oversampled output.
- Fully resizable vector UI.
Pricing and Availability
SSNN is available now for £89 GBP (intro price of £65 GBP until mid-July 2026). Formats: VST3, AU. Platforms: macOS (universal binary), Windows. A multi-core processor equivalent to or better than Apple M1 is recommended.
YouTube.com/watch?v=akx37fk3jFs
YouTube.com/watch?v=OlPgveuldJ4
YouTube.com/watch?v=aDl00p3Kp6c
Read More