anira

Real-time neural inference for audio

anira bridges modern neural network architectures with the hard real-time constraints of audio processing. It runs inference off the audio thread, manages a static thread pool for parallel execution, and ships with benchmarking tools to measure and tune latency end-to-end.

Features

  • Real-time-safe with deterministic runtimes
  • Static thread pool for parallel inference
  • Supports LibTorch, ONNX Runtime, TensorFlow Lite
  • Stateful and stateless models
  • Cross-platform: macOS · Linux · Windows
  • Built-in benchmarking tools