Taming Multimodal Joint Training for High-Quality Video-to-Audio Synthesis

Ho Kei Cheng, Masato Ishii, Akio Hayakawa, Takashi Shibuya, Alexander Schwing, Yuki Mitsufuji

University of Illinois Urbana-Champaign, Sony AI, and Sony Group Corporation

[Paper (being prepared)] [Project Page]

Note: This repository is still under construction. Single-example inference should work as expected. The training code will be added. Code is subject to non-backward-compatible changes.

Highlight

MMAudio generates synchronized audio given video and/or text inputs. Our key innovation is multimodal joint training which allows training on a wide range of audio-visual and audio-text datasets. Moreover, a synchronization module aligns the generated audio with the video frames.

Results

(All audio from our algorithm MMAudio)

Videos from Sora:

https://github.com/user-attachments/assets/82afd192-0cee-48a1-86ca-bd39b8c8f330

Videos from MovieGen/Hunyuan Video/VGGSound:

https://github.com/user-attachments/assets/29230d4e-21c1-4cf8-a221-c28f2af6d0ca

For more results, visit https://hkchengrex.com/MMAudio/video_main.html.

Installation

We have only tested this on Ubuntu.

Prerequisites

We recommend using a miniforge environment.

• Python 3.8+
• PyTorch 2.5.1+ and corresponding torchvision/torchaudio (pick your CUDA version https://pytorch.org/)
• ffmpeg<7 (this is required by torchaudio, you can install it in a miniforge environment with conda install -c conda-forge 'ffmpeg<7')

Clone our repository:

git clone https://github.com/hkchengrex/MMAudio.git

Install with pip:

cd MMAudio
pip install -e .

(If you encounter the File "setup.py" not found error, upgrade your pip with pip install --upgrade pip)

Pretrained models:

The models will be downloaded automatically when you run the demo script. MD5 checksums are provided in mmaudio/utils/download_utils.py

The 44.1kHz vocoder will be downloaded automatically.

The expected directory structure (full):

MMAudio
├── ext_weights
│   ├── best_netG.pt
│   ├── synchformer_state_dict.pth
│   ├── v1-16.pth
│   └── v1-44.pth
├── weights
│   ├── mmaudio_small_16k.pth
│   ├── mmaudio_small_44k.pth
│   ├── mmaudio_medium_44k.pth
│   └── mmaudio_large_44k.pth
└── ...

The expected directory structure (minimal, for the recommended model only):

MMAudio
├── ext_weights
│   ├── synchformer_state_dict.pth
│   └── v1-44.pth
├── weights
│   └── mmaudio_large_44k.pth
└── ...

Demo

By default, these scripts use the large_44k model. In our experiments, inference only takes around 6GB of GPU memory (in 16-bit mode) which should fit in most modern GPUs.

Command-line interface

With demo.py

python demo.py --duration=8 --video=<path to video> --prompt "your prompt" 

The output (audio in .flac format, and video in .mp4 format) will be saved in ./output. See the file for more options. Simply omit the --video option for text-to-audio synthesis. The default output (and training) duration is 8 seconds. Longer/shorter durations could also work, but a large deviation from the training duration may result in a lower quality.

Gradio interface

Supports video-to-audio and text-to-audio synthesis.

python gradio_demo.py

Known limitations

1. The model sometimes generates undesired unintelligible human speech-like sounds
2. The model sometimes generates undesired background music
3. The model struggles with unfamiliar concepts, e.g., it can generate "gunfires" but not "RPG firing".

We believe all of these three limitations can be addressed with more high-quality training data.

Training

Work in progress.

Evaluation

Work in progress.

Acknowledgement

Many thanks to:

• Make-An-Audio 2 for the 16kHz BigVGAN pretrained model
• BigVGAN
• Synchformer