|
--- |
|
license: mit |
|
tags: |
|
- audio-generation |
|
--- |
|
|
|
[Dance Diffusion](https://github.com/Harmonai-org/sample-generator) is now available in 🧨 Diffusers. |
|
|
|
## FP32 |
|
|
|
```python |
|
# !pip install diffusers[torch] accelerate scipy |
|
from diffusers import DiffusionPipeline |
|
from scipy.io.wavfile import write |
|
|
|
model_id = "harmonai/jmann-large-580k" |
|
pipe = DiffusionPipeline.from_pretrained(model_id) |
|
pipe = pipe.to("cuda") |
|
|
|
audios = pipe(audio_length_in_s=4.0).audios |
|
|
|
# To save locally |
|
for i, audio in enumerate(audios): |
|
write(f"test_{i}.wav", pipe.unet.sample_rate, audio.transpose()) |
|
|
|
# To dislay in google colab |
|
import IPython.display as ipd |
|
for audio in audios: |
|
display(ipd.Audio(audio, rate=pipe.unet.sample_rate)) |
|
``` |
|
|
|
## FP16 |
|
|
|
Faster at a small loss of quality |
|
|
|
```python |
|
# !pip install diffusers[torch] accelerate scipy |
|
from diffusers import DiffusionPipeline |
|
from scipy.io.wavfile import write |
|
import torch |
|
|
|
model_id = "harmonai/jmann-large-580k" |
|
pipe = DiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16) |
|
pipe = pipe.to("cuda") |
|
|
|
audios = pipeline(audio_length_in_s=4.0).audios |
|
|
|
# To save locally |
|
for i, audio in enumerate(audios): |
|
write(f"{i}.wav", pipe.unet.sample_rate, audio.transpose()) |
|
|
|
# To dislay in google colab |
|
import IPython.display as ipd |
|
for audio in audios: |
|
display(ipd.Audio(audio, rate=pipe.unet.sample_rate)) |
|
``` |
