aoiandroid/supertonic-coreml-fp16

Supertonic Core ML (fp16) — on-device iOS companion

This Hugging Face model repository is maintained for on-device iOS integration around Supertone/supertonic (OpenRAIL-M).

What is published here

Path	Contents
scripts/	convert_supertonic_coreml.py, run_mac_tts.py, requirements-export.txt — download Hub ONNX + configs + voice styles; run TTS on macOS with ONNX Runtime + Core ML Execution Provider.
coreml_fp16/	Placeholder for native Core ML mlprogram / .mlpackage bundles (fp16 storage). This mirror does not currently ship converted graphs here: coremltools does not support a reliable ONNX→mlprogram path for these networks (legacy converters fail on ops such as LayerNorm). When/if native bundles exist, they should be placed under this directory with a short MANIFEST.json (see scripts/ comments).

Recommended runtime (today)

• Weights: use upstream Supertone/supertonic onnx/*.onnx + onnx/tts.json + onnx/unicode_indexer.json + voice_styles/*.json.
• Apple acceleration: load ONNX with ONNX Runtime and enable the Core ML execution provider (same approach as typical on-device iOS clients).

License

• Base model & voice styles: follow Supertone/supertonic (OpenRAIL-M) and Supertone’s terms.
• Scripts in scripts/: follow the license noted in those script files where copied from a local project.

Disclaimer

This repo is not affiliated with Supertone Inc. It exists to document and distribute companion tooling and a reserved layout for optional Core ML artifacts. Do not imply official endorsement by Supertone.