2024-10-12 01:16:20 +08:00
|
|
|
# MusicGen
|
|
|
|
|
|
|
|
|
|
An example of Meta's MusicGen model in MLX.[^1] MusicGen is used to generate
|
|
|
|
|
music from text descriptions.
|
|
|
|
|
|
|
|
|
|
### Setup
|
|
|
|
|
|
|
|
|
|
Install the requirements:
|
|
|
|
|
|
|
|
|
|
```
|
|
|
|
|
pip install -r requirements.txt
|
|
|
|
|
```
|
|
|
|
|
|
|
|
|
|
### Example
|
|
|
|
|
|
|
|
|
|
An example using the model:
|
|
|
|
|
|
|
|
|
|
```python
|
2024-10-14 21:21:25 +08:00
|
|
|
from musicgen import MusicGen
|
2024-10-12 01:16:20 +08:00
|
|
|
from utils import save_audio
|
|
|
|
|
|
|
|
|
|
model = MusicGen.from_pretrained("facebook/musicgen-medium")
|
|
|
|
|
|
|
|
|
|
audio = model.generate("happy rock")
|
|
|
|
|
|
|
|
|
|
save_audio("out.wav", audio, model.sampling_rate)
|
|
|
|
|
```
|
|
|
|
|
|
|
|
|
|
[^1]: Refer to the [arXiv paper](https://arxiv.org/abs/2306.05284) and
|
|
|
|
|
[code](https://github.com/facebookresearch/audiocraft/blob/main/docs/MUSICGEN.md) for more details.
|
