Anira Web

Demos

• Simple Gain Stereo
  The minimal end-to-end Anira Web setup — a stereo gain plugin running through the default WASM-side ONNX Runtime, and the foundation every other demo builds on.
• Streaming Gain Stereo
  Delivers gain as a second streamable tensor for sample-accurate updates, driven by a Web Audio LFO through a custom audio worklet.
• JS Callback
  The simplest way to move inference out of WASM and into JavaScript — plugs in the built-in JSBackendBase as a passthrough hook on the inference worker.
• ONNX Runtime Web Backend
  Runs the same stereo gain model through Anira Web's ONNXRuntimeWebBackend, swapping the WASM engine for onnxruntime-web on the JS side.
• JS Copying
  A user-written custom inference backend — subclasses JSBackendBase and ships its own custom inference worker, providing the template for wrapping any JS-side runtime.
• PrePostProcessors
  Routes pre- and post-processing through JS via JSPrePostProcessor, demonstrated by clamping the gain slider to [0, 1] from a JavaScript callback firing on every block.
• Steerable-NAFX (CNN)
  A CNN-based neural guitar amp whose JSPrePostProcessor assembles each block's input as a sliding window prepended with 132 samples of past context for the convolutions to look back at.
• GuitarLSTM (HybridNN)
  A hybrid LSTM/dense guitar amp whose JSPrePostProcessor packs each block into a batched [BUFFER_SIZE, 1, 150] tensor of overlapping 150-sample context windows run in parallel.
• RAVE: Funk & Djembe
  Two RAVE neural synthesis models — funk drums and djembe — running in parallel. A fade slider blends between them; a mix slider sets the dry/wet balance with latency-aligned bypass.

Third-party licenses

This site bundles third-party open-source software. See licenses for the full attribution list.