Neural audio for the software-defined vehicle

The cabin is the first place a car should be intelligent.

Echo cancellation, road-noise cancellation, engine sound — rebuilt as learned models running on the vehicle’s own silicon, executed by an automotive-grade runtime.

Read the thesis The AREN runtime

ECEcho cancellation
RNCRoad-noise cancellation
ASESound enhancement
ARENAutomotive runtime

Deterministic where it must be. Learned where it pays. The network aims — the DSP fires.

The operating principle behind every system we ship

01What we build

Three claims, each held to evidence.

Neural audio, from a production audio team.

In-cabin acoustics — echo cancellation, road-noise cancellation, sound enhancement — rebuilt as learned models on the vehicle's NPU, by engineers who have shipped production audio for ten years.

Neural audio →

A hybrid chain that keeps its promises.

A thin DSP layer executes at sample rate behind a hard safety limiter; the neural core adapts, predicts, and suppresses at frame rate. Reviewable by a safety team, adaptive in the field.

The architecture →

An execution layer beneath, not a model on top.

Learned workloads need automotive-grade execution, update, and lifecycle management on heterogeneous silicon. Audio is the first workload — the AREN runtime is the platform it proves.

The AREN runtime →

02Architecture

One chain. Two disciplines.

Where a millisecond matters, the system stays deterministic. Where judgment matters, it is learned. Concept level by design — the details are patent work, not marketing.