// the find
christian-vorhemus/azure-percept-py
Library to access the sensors of Azure Percept in Python
Python library wrapping the Azure Percept DK's audio and vision hardware — lets you record audio, run ONNX models on the Myriad VPU, grab camera frames, and record video from Python without touching the underlying C/USB stack directly. Audience is the small set of people who actually own Azure Percept DK hardware and want to write Python against it.
The VPU inference pipeline handles model conversion from ONNX to blob format inline, so you don't need a separate OpenVINO toolchain step. Multi-source camera input (module camera + USB cameras simultaneously) is a genuinely useful feature for edge ML applications. The audio API's polling-based readiness check (`is_ready()`) is the right pattern for hardware that requires network authentication before use — doesn't paper over the latency. Pre-built aarch64 shared libraries mean users skip a painful native build on the device itself.
Azure Percept was retired by Microsoft in March 2023 — the hardware is end-of-life and the authentication endpoint it phones home to may stop working at any time, making this library a dead end. Pre-built `.so` files committed directly to the repo with no reproducible build means you're trusting binaries you can't audit or rebuild. Python 3.6–3.10 support only; Python 3.11+ is explicitly unsupported and there's no indication anyone will fix that. Eight stars and two forks says the community never materialized, so don't expect help when the USB enumeration breaks.