Inference with Audiocraft - arXiv:2402.04825v2

[TtesseractT]

Hello,

So Ive read the paper, and it seems very impressive.

I wanted to know if there was any way of getting inference working with Audiocraft?

I'm a little unsure on the steps in doing this as the repo is still being developed with docs 👍

The goal of recreating the same process times as the paper did is what really interests me.

Any ideas or recommendations.

[danishbelim]

Yes, you can perform inference using Audiocraft, Meta's toolkit for generating and manipulating audio with AI. Audiocraft includes tools such as MusicGen for generating music and AudioGen for general audio generation. Here's how you can get inference working with Audiocraft

[danishbelim]

1. Set Up Your Environment
   Install Dependencies
   First, ensure you have the necessary dependencies. You’ll need Python and some Python packages.

bash
Copy code
pip install torch
pip install audiocraft
You may need to install additional dependencies based on your specific environment and hardware (like CUDA for GPU support).

[danishbelim]

If you want to work directly with the source, you can clone the Audiocraft GitHub repository:

bash
Copy code
git clone https://github.com/facebookresearch/audiocraft.git
cd audiocraft
pip install -e .

[danishbelim]

Using Audiocraft for Inference
Inference with Pre-Trained Models
Audiocraft provides pre-trained models that you can use for inference. Here’s how you can use MusicGen for generating music:

python
Copy code
from audiocraft.models import MusicGen

Load the pre-trained model

model = MusicGen.get_pretrained('melody')

Generate music

output = model.generate([{
'prompt': 'A serene and calm piano piece in the style of classical music',
'duration': 10, # seconds
}])