AI voices in Kroma run fully on the device using:

- Supertonic 3 text-to-speech by Supertone Inc.
  Inference code adapted from the Supertonic examples — License: MIT (Copyright (c) 2025 Supertone Inc.)
  https://github.com/supertone-inc/supertonic
  Model weights — License: OpenRAIL-M — https://huggingface.co/Supertone/supertonic-3/blob/main/LICENSE
  The OpenRAIL-M license includes use-based restrictions (for example, no use to deceive, defame,
  harass or impersonate people without consent, or for unlawful purposes). By using the AI voices
  you agree to follow those restrictions.
  Voice styles (voices.bin) are converted from the official voice_styles of Supertonic 3.
  The int8 model files are downloaded once from
  https://huggingface.co/csukuangfj2/sherpa-onnx-supertonic-3-tts-int8-2026-05-11 and checked by SHA-256.
- ONNX Runtime Web — License: MIT (see lib/ort/THIRD-PARTY-NOTICES.txt)
