You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

supertonic-onnx16

Hybrid FP16 / FP32 ONNX repackaging of Supertone/supertonic: duration_predictor, text_encoder, and vocoder use fp16 internal weights (keep_io_types=True). vector_estimator remains fp32 because a fully-fp16 export fails ONNX Runtime type checks (Cast / float mismatch).

Layout

onnx16/
  duration_predictor.onnx
  text_encoder.onnx
  vector_estimator.onnx
  vocoder.onnx
  tts.json
  unicode_indexer.json
voice_styles/
  *.json
manifest.json

Usage

Load with ONNX Runtime (CPU or Core ML EP on Apple platforms). Contract matches upstream onnx/ graphs (text_ids, text_mask, etc.).

How this was built

See scripts/build_hub_onnx16_staging.py in your local project (Models/supertone-supertonic-coreml/): fp32 ONNX → onnxconverter_common.float16.convert_float_to_float16(..., keep_io_types=True).

License

Follow Supertone/supertonic (OpenRAIL-M). This mirror is not official Supertone content.

macOS sample (hybrid fp16 ONNX)

Benchmark text (local INPUT/text/1_text_eng.txt):

The spring park was full of cherry blossoms, and the visitors were taking pictures and enjoying the walk.

samples/1_text_eng_fp16.wav

Synthesized from local hub_onnx16_staging/onnx16/ (hybrid fp16: vector_estimator remains fp32). Voice F1, steps=20, ONNX Runtime + Core ML EP on macOS.

Downloads last month: -; Downloads are not tracked for this model. How to track

Model tree for aoiandroid/supertonic-onnx16

Base model

Supertone/supertonic

Quantized

(6)

this model