S2R-Adaptation-TTS · cubic

Project page · Paper PDF · GitHub

OmniVoice Burmese native routed LoRA weights and inference helpers.

What this checkpoint contains

Property	Value
Release	cubic
Base	k2-fsa/OmniVoice
Language	Burmese
Format	Native routed LoRA; not a standard Hugging Face PEFT adapter
LoRA	Rank 8, alpha 16; 28 transformer layers
Routing	text_prediction, including an audio-head delta

Paper results are reported on the project page and are not benchmark claims for this download.

Download and run

Use Python 3.12 with a compatible PyTorch installation. Install the upstream OmniVoice 0.1.5 inference package and Hugging Face client:

pip install omnivoice==0.1.5 huggingface_hub soundfile accelerate
hf download InsiderX-Pro/S2R-Adaptation-TTS --local-dir s2r-omnivoice
cd s2r-omnivoice
python inference.py \
  --text "သင် မင်္ဂလာပါ။" \
  --ref-audio /path/to/reference.wav \
  --ref-text "The exact transcript of your reference recording." \
  --output output.wav

Replace the text and reference inputs with your own. The reference transcript must match the supplied recording. Use --device cpu if CUDA is unavailable. Use --base-dir /path/to/OmniVoice to reuse a local base-model directory.

The loader downloads the base at revision c5fdb5ccb189668d56333f77ba2629f4cd7535f4 and verifies its weight SHA-256: 730839316de585f4c8298ec0e1712efc10fb19c6fa4e36eb741cb8d51ebcf6aa. It restores the native adapter and installs per-forward text/prediction routing. Loading only adapter_model.safetensors with a generic PEFT loader is insufficient. The base and its audio codec are separate downloads; they are not duplicated here.

Inference files

• adapter_model.safetensors: original adapter tensors.
• peft_config.json: native adapter specification, with a portable base identifier.
• tokenizer.json, tokenizer_config.json, chat_template.jinja: checkpoint tokenizer assets.
• inference.py, native_adapter/: loading and routing helpers from the project.

Base-model licenses apply separately. No additional blanket license is declared for the project-specific adapter or helpers in this upload.

Citation

@unpublished{lu2026reliable,
  title = {From Reliable Text to Real Voices: Trust-Aware Progressive Adaptation for Low-Resource TTS},
  author = {Lu, Jiayi and Geng, Yizhong and Yang, Jinghan and Jiang, Tianhan and An, Boxun and Gao, Yingming and Li, Ya},
  year = {2026},
  note = {Manuscript}
}

The arXiv entry is pending. Use the project page for the current manuscript and the paper's experimental results.