Add model card for GibbsTTS

README.md

@@ -1,3 +1,56 @@
----
-license: mit
----
+---
+license: mit
+pipeline_tag: text-to-speech
+---
+
+# GibbsTTS
+
+GibbsTTS is a zero-shot text-to-speech model based on metric-induced discrete flow matching (MI-DFM). It incorporates a kinetic-optimal scheduler and finite-step CTMC moment correction to address heuristic scheduling and path-tracking errors in discrete generation.
+
+- **Paper:** [Kinetic-Optimal Scheduling with Moment Correction for Metric-Induced Discrete Flow Matching in Zero-Shot Text-to-Speech](https://arxiv.org/abs/2605.09386)
+- **Project Page:** [https://ydqmkkx.github.io/GibbsTTSProject](https://ydqmkkx.github.io/GibbsTTSProject)
+- **Code:** [https://github.com/ydqmkkx/GibbsTTS](https://github.com/ydqmkkx/GibbsTTS)
+
+## Model Description
+GibbsTTS addresses two primary limitations of MI-DFM:
+1. **Kinetic-Optimal Scheduler:** A training-free numerical schedule that traverses the probability path at constant Fisher-Rao speed, replacing heuristic schedulers.
+2. **Moment Correction:** A finite-step correction that adjusts jump probabilities to reduce errors while preserving the jump destination distribution.
+
+The model was trained on the [Emilia](https://huggingface.co/datasets/amphion/Emilia-Dataset) dataset and supports English and Chinese Mandarin.
+
+## Sample Usage
+
+To use the model, first clone the [official repository](https://github.com/ydqmkkx/GibbsTTS) and install the requirements.
+
+```python
+from models import GibbsTTS
+from config import ModelConfig
+from IPython.display import Audio
+import soundfile as sf
+
+# Initialize model
+configs = ModelConfig()
+model = GibbsTTS(configs)
+
+# English Synthesis
+prompt_audio = "./prompt_examples/common_voice_en_188092-common_voice_en_188093.wav"
+prompt_text = "This man looked exactly the same, except that now the roles were reversed."
+target_text = "He also tried to remember some good stories to relate as he sheared the sheep."
+language = "en"
+
+audio = model.synthesize(prompt_audio=prompt_audio, prompt_text=prompt_text, target_text=target_text, language=language)
+
+# Save and play
+sf.write('target.wav', audio, 24000, 'PCM_24')
+Audio(data=audio, rate=24000) 
+```
+
+## Citation
+```bibtex
+@article{GibbsTTS,
+ author    = {Dong Yang and Yiyi Cai and Haoyu Zhang and Yuki Saito and Hiroshi Saruwatari},
+ title     = {Kinetic-Optimal Scheduling with Moment Correction for Metric-Induced Discrete Flow Matching in Zero-Shot Text-to-Speech},
+ year      = {2026},
+ journal   = {arXiv preprint arXiv:2605.09386},
+}
+```