Pre-converted Whisper speech-to-text models for Dictate's on-device (offline) transcription (issue #104).
These are mirrors of the sherpa-onnx ONNX exports by @csukuangfj, derived from OpenAI Whisper. OpenAI Whisper is MIT-licensed and sherpa-onnx is Apache-2.0, so redistribution is permitted; they are hosted here so the app downloads from a stable, project-controlled source.
All int8-quantised. Multilingual: tiny / base / small (99 languages). English-only: the *.en variants. Base is the recommended default.