# GPT-SoVITS v2ProPlus — ONNX (Sokuji Local Native TTS)

Converted from the base pretrained checkpoints of
[GPT-SoVITS](https://github.com/RVC-Boss/GPT-SoVITS) (MIT) —
[lj1995/GPT-SoVITS](https://huggingface.co/lj1995/GPT-SoVITS) (MIT) —
via [Genie-TTS](https://github.com/High-Logic/Genie-TTS) (MIT).
Runtime assets (chinese-hubert-base, speaker encoder, RoBERTa, G2P dictionaries)
mirrored from [High-Logic/Genie](https://huggingface.co/High-Logic/Genie) (MIT;
hubert originally [TencentGameMate/chinese-hubert-base](https://huggingface.co/TencentGameMate/chinese-hubert-base), MIT;
RoBERTa Apache-2.0).

Weight bins under `model/` and `genie_data/chinese-hubert-base/` are fp16;
the Sokuji sidecar expands them to fp32 in place at load time.

Default voice clips: en — JFK 1961 inaugural address excerpt (US government
work, public domain); zh/ja — fully synthetic speech generated with
Qwen3-TTS 1.7B (Apache-2.0), no human speaker involved.