Overview
AirTen is our CPU-based neural network inference engine. It runs without external dependencies like ONNX or ML frameworks, providing low-latency inference suitable for real-time audio processing.
Characteristics
- CPU-only: No GPU or external inference runtimes
- No framework dependencies: Self-contained, no PyTorch or TensorFlow
- Deterministic latency: Suitable for real-time audio processing
- Low memory overhead: Minimal footprint for embedded systems
Model Support
AirTen executes our model binaries (.aicmodel files). Supported models and specifications are documented on our Models Guide.
- Sub-frame latency on modern CPUs
- Minimal memory footprint
Performance varies with CPU architecture and model complexity.
