# 🎵 MMAG: A Multi‑Control Mixed Audio Generation Benchmark

**MMAG** is a comprehensive benchmark for evaluating mixed audio generation under multiple control conditions. It assesses a model's ability to generate coherent acoustic scenes containing speech, music, and sound effects simultaneously, while supporting fine-grained control over speaker identity and temporal alignment.

[![arXiv](https://img.shields.io/badge/arXiv-2608.06900-b31b1b.svg)](https://arxiv.org/abs/2608.06900)
[![Project Page](https://img.shields.io/badge/Project-Page-1a6b8a.svg)](https://hirookie9.github.io/MMAG-Page)
[![Hugging Face](https://img.shields.io/badge/HuggingFace-Dataset-fcd022.svg)](https://huggingface.co/datasets/rookie9/MMAG)

---

## 🚀 Updates

- [ ] Evaluation pipeline
- [ ] Raw multi-expert model pipeline  
- [ ] New version release (more precise sound event and timestamp annotations)

---

## 🧪 Evaluated Models

We evaluate the following models on MMAG:

| Model | Description |
|-------|-------------|
| [AudioDirector](https://github.com/Riddae/AuDirector) | Agentic orchestrator |
| [LTX-2.3](https://github.com/Lightricks/LTX-2) | Unified audio-visual generation |
| [Ovi 1.1](https://github.com/character-ai/Ovi) | Unified audio-visual generation |
| [MOVA-720p](https://github.com/OpenMOSS/MOVA) | Unified audio-visual generation |
| [JavisDiT++](https://github.com/JavisVerse/JavisDiT) | Unified audio-visual generation |
| [UniAVGen](https://github.com/MCG-NJU/Sora2-mini) | Unified audio-visual generation |
| [Dasheng-AudioGen](https://github.com/xiaomi-research/dasheng-audiogen) | Native mixed-audio generation |
| [Ming-Omni-TTS](https://github.com/example/Ming-Omni-TTS) | Native mixed-audio generation |

---

## 🙏 Acknowledgements

We thank the authors of [MECAT](https://github.com/xiaomi-research/mecat) for their valuable work.

---

## 📜 License

This project is licensed under the **CC BY 4.0** license.

---