# Intel Kokoro TTS Model Staging

This folder is the provider-local staging area for the Intel Kokoro TTS benchmark.

Runtime layout:
- `config.json` is staged on first run from `mweinbach1/ai-pc-benchmarks-tts-intel-openvino`, with `hexgrad/Kokoro-82M` as fallback.
- `kokoro-v1_0.pth` is staged on first run from `mweinbach1/ai-pc-benchmarks-tts-intel-openvino`, with `hexgrad/Kokoro-82M` as fallback.
- `voices/` is populated lazily on first run from `mweinbach1/ai-pc-benchmarks-tts-intel-openvino`, with upstream fallback for requested voice files such as `af_heart.pt`.
- `npu_static/` may hold provider-local OpenVINO exports when the benchmark is run with `--compiled-model-dir workloads\TTS\Intel\model`.

The checked-in provider contract intentionally keeps this folder lightweight. The benchmark materializes the real model payloads here so subsequent runs stay provider-local and can run offline after staging.

These files hold source payloads and optional OpenVINO IRs, not machine-specific compiled binaries. OpenVINO still compiles for the active Intel device locally at runtime, and any generated compiler cache stays machine-local.