sonicLAB releases SSNN spiking neural network engine

Spiking neural networks (SNNs) use discrete spike events rather than continuous activations, which makes them a better conceptual fit for temporal and event-driven signals such as audio. sonicLAB's materials describe a leaky integrate-and-fire implementation where individual spikes become discrete sonic events, and a continuous FFT that updates connection weights in realtime, effectively making incoming audio both the training stimulus and the synthesis material. The product page frames each spike as driving per-layer audio buffers that feed eight synthesis engines, a design that pairs neuromorphic event dynamics with conventional synthesis stages.

Running a 960-neuron SNN across 32 layers with per-layer audio buffers and FFT-driven weight updates implies nontrivial CPU and threading design. Rekkerd notes a multicore engine ("3 workers + main thread") and oversampled output, which suggests concurrency and buffer management are used to keep neuron activity visible at 60 Hz while preserving audio-rate processing. The inclusion of OSC spike broadcast enables external synchronization and live-performance integration.

Public releases that embed spiking neural networks into creative tools remain uncommon compared with standard deep learning audio workflows that use continuous models, spectrogram-based models, or transformer-style architectures. SSNN represents a distinct design point that leverages event-driven neural dynamics for temporal texture and emergent rhythmic behaviour rather than sample-level predictive modelling. For practitioners interested in neuromorphic signal processing, this is a concrete, shipping implementation to examine for latency, controllability, and timbral outcomes.

sonicLAB documents features including spectral training with realtime weight morphing, custom arpeggiator patterns with quantization, mouse-over info for UI elements, and OSC spike output. Rekkerd highlights a resizable vector UI and platform plugin formats (VST3, AU). Pricing details reported by Rekkerd place SSNN at 89 GBP with a temporary launch price of 65 GBP until mid-July.

Interest from experimental musicians, sound designers, and academic groups could produce third-party demonstrations and analyses that reveal the workflow tradeoffs between SNN-driven synthesis and more conventional sample- or model-based approaches. If the plugin proves robust in DAWs and live rigs, it may surface new idioms for rhythmic generation and adaptive timbral transformation.

sonicLAB released SSNN (Spiking and Sounding Neural Network) as a live audio processor and synthesizer, with the product listed as first released in 06/2026 on sonicLAB's site. Per sonicLAB's product page, SSNN runs a leaky integrate-and-fire spiking neural network with 960 neurons across 32 layers, operating live alongside the audio engine. The same page describes a continuous FFT that writes incoming audio spectral profiles into the network's connection weights. Rekkerd's coverage adds that SSNN drives eight synthesis engines in parallel, offers a resizable vector UI, multicore processing ("3 workers + main thread"), OSC spike broadcast, and an ultra-fast event manager. Rekkerd reports availability for Windows and Mac in VST3 and AU formats and a regular price of 89 GBP, with a launch price of 65 GBP available until mid-July.

sonicLAB provides conceptual and feature-level information on the product page and Rekkerd provides an introductory hands-on summary, but neither source publishes low-level metrics such as typical CPU load, thread scheduling details, or the precise learning rule math used for weight updates. sonicLAB's site includes explanatory copy and a short quote on learning, but technical reproducibility details are not present in the scraped materials.

Observe how SSNN manages latency and stability under live input, especially when the FFT writes to synaptic weights in realtime. Monitor CPU and multithreading behaviour on typical DAW hosts and whether the OSC spike broadcast is precise enough for synchronization with external devices. Also watch for community patches, presets, or whitepapers from sonicLAB that document the leaky integrate-and-fire parameters, learning rules, and how synthesis engines map spikes to sound events.

Experimentation and benchmarking will be necessary to assess real-time reliability, controllability of emergent textures, and integration with existing toolchains. Those considering SSNN for research should treat the release as an experimental, performant plugin and plan to measure latency, determinism, and parameter stability under live conditions.