# The built-in voice pack

Every clip under this folder was rendered by `scripts/voice-pack.py` with
[Kokoro-82M](https://huggingface.co/hexgrad/Kokoro-82M) (weights and voice
packs under the Apache License 2.0), run through
[kokoro-onnx](https://github.com/thewh1teagle/kokoro-onnx) (MIT). The
voices are Kokoro's `af_heart` (Heart), `bm_lewis` (Lewis) and `am_michael`
(Michael).

Each clip was checked by a speech recogniser (Whisper small.en) before it
was kept; `scripts/voice-pack-report.json` lists the few plain words it
could not confirm (homophones such as "hymn", and some names).

The audio is served as part of phonetic; neither model runs in the product.

Apache License 2.0: https://www.apache.org/licenses/LICENSE-2.0