[usamahz]

Adds a draft Learning Path and a new Zephyr application port for recorded-audio Silero VAD on Alif Ensemble E8 with ExecuTorch. It covers streaming context and recurrent state, released export dependencies, U85-256 export, firmware construction, programming, UART capture, and comparison of all 79 speech probabilities with the host reference.

Physical E8 deployment is unverified. No Silero firmware has been flashed or run on the board. This PR stays in draft while hardware bring-up and probability validation remain pending. The Learning Path and child pages are also marked as drafts.

Review the lessons in navigation order, then the bundled application source archive. Host dependencies use released PyTorch 2.14.0, ExecuTorch 1.5.1, and TorchAO 0.18.0. The application preserves recurrent state and includes the required CPU operators alongside the Ethos-U delegate. It processes prerecorded audio; microphone capture is outside its scope.

Validation passed: host export from the pinned public source, E8 firmware cross-build, Hugo 0.130.0 site build, repository metadata checks, all 23 shell blocks, embedded Python syntax, and source-archive integrity. The exported PTE and 79-value reference match the retained FVP-tested release artifacts byte for byte. The cross-built firmware uses 830,084 bytes of flash and 3,597,440 bytes of RAM (47.32% of the configured application region). These checks do not establish physical-board execution, probability agreement, latency, or power.

Authored with assistance from OpenAI Codex.

• I have reviewed Create a Learning Path.
• I have checked my contribution for confidential information.

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of the Creative Commons Attribution 4.0 International License.