Technology

Audio on top. A runtime beneath.

The stack in one page. Learned audio models are the product a passenger hears; executing them on vehicle silicon for fifteen years is the platform an OEM buys. The diagram below is the shortest honest answer to “where does your thing run?”

The AREN runtime in the SDV stackLayer diagram. On top, workloads: neural audio — echo cancellation, road-noise cancellation, sound enhancement — and further cockpit AI workloads. They run on the AREN automotive runtime, which handles model execution, update, and lifecycle. Beneath the runtime sits heterogeneous vehicle silicon — two generic NPUs and a separate deterministic safety domain. A model-update path connects to the runtime by a dashed line.Neural audioecho cancellation · road-noise cancellation · sound enhancementCockpit AI workloadssame execution contractARENautomotive runtime — model execution · update · lifecycleany model · any silicon · automotive-gradeNPU Avendor siliconNPU Bvendor siliconSafety domaindeterministic · limitermodelupdatesaudio first · the runtime is the platform

Workloads run against one execution contract. AREN owns model execution, update, and lifecycle. Silicon labels are generic — targets are heterogeneous by design.

Workloads

Neural audio

Echo cancellation, road-noise cancellation, and sound enhancement, rebuilt as learned models. A neural-first hybrid chain: thin DSP at sample rate where latency is physics, a neural core on the NPU where judgment pays.

The neural audio page →

Platform

The AREN runtime

Any supported model, any supported silicon, automotive-grade. Model execution, update, and lifecycle management for the life of the vehicle. Audio is the first workload it proves.

The AREN runtime page →