Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,116 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: mit
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
tags:
|
| 4 |
+
- audiocraft
|
| 5 |
+
- Musicgen
|
| 6 |
+
- Music Generation
|
| 7 |
+
- micro-musicgen
|
| 8 |
+
---
|
| 9 |
+
|
| 10 |
+
# micro-musicgen-acid
|
| 11 |
+
|
| 12 |
+
Curated and trained by Aaron Abebe.
|
| 13 |
+
|
| 14 |
+
|
| 15 |
+
|
| 16 |
+
> [!WARNING]
|
| 17 |
+
> WARNING: **These models WILL sound bad to a lot of people.** The goal is not create pleasant sounding music,
|
| 18 |
+
> but to spark creativity by using the weird sounds of Neural Codecs for music production and sampling!
|
| 19 |
+
|
| 20 |
+
Micro-Musicgen is a new family of super small music generation models focussing on experimental music and latent space exploration capabilities.
|
| 21 |
+
These models have unique abilities and drawbacks which should enhance creativity when working with them while creating music.
|
| 22 |
+
|
| 23 |
+
- **only unconditional generation**: Trained without text-conditioning to reduce model size.
|
| 24 |
+
- **very fast generation times**: ~8secs for 10x 10sec samples.
|
| 25 |
+
- **permissive licensing**: The models are trained from scratch using royalty-free samples and handmade chops,
|
| 26 |
+
which allows them to be released via the MIT License.
|
| 27 |
+
|
| 28 |
+
|
| 29 |
+
This is the second entry in the series and is called `micro-musicgen-acid`. It's trained on different 303 sample packs as well as audio I made with my Behringer TD-3.
|
| 30 |
+
|
| 31 |
+
If you find this model interesting, please consider:
|
| 32 |
+
|
| 33 |
+
- following me on [GitHub](https://github.com/aaronabebe)
|
| 34 |
+
- following me on [Twitter](https://twitter.com/mcaaroni)
|
| 35 |
+
|
| 36 |
+
## Samples
|
| 37 |
+
|
| 38 |
+
All samples are from a single run, without cherry picking.
|
| 39 |
+
|
| 40 |
+
<table style="width:100%; text-align:center;">
|
| 41 |
+
<tr>
|
| 42 |
+
<td>
|
| 43 |
+
<audio controls>
|
| 44 |
+
<source src="https://huggingface.co/pharoAIsanders420/micro-musicgen-jungle/resolve/main/assets/output_3.wav?download=true" type="audio/wav">
|
| 45 |
+
Your browser does not support the audio element.
|
| 46 |
+
</audio>
|
| 47 |
+
</td>
|
| 48 |
+
<td>
|
| 49 |
+
<audio controls>
|
| 50 |
+
<source src="https://huggingface.co/pharoAIsanders420/micro-musicgen-jungle/resolve/main/assets/output_2.wav?download=true" type="audio/wav">
|
| 51 |
+
Your browser does not support the audio element.
|
| 52 |
+
</audio>
|
| 53 |
+
</td>
|
| 54 |
+
<td>
|
| 55 |
+
<audio controls>
|
| 56 |
+
<source src="https://huggingface.co/pharoAIsanders420/micro-musicgen-jungle/resolve/main/assets/output_6.wav?download=true" type="audio/wav">
|
| 57 |
+
Your browser does not support the audio element.
|
| 58 |
+
</audio>
|
| 59 |
+
</td>
|
| 60 |
+
</tr>
|
| 61 |
+
<tr>
|
| 62 |
+
<td>
|
| 63 |
+
<audio controls>
|
| 64 |
+
<source src="https://huggingface.co/pharoAIsanders420/micro-musicgen-jungle/resolve/main/assets/output_9.wav?download=true" type="audio/wav">
|
| 65 |
+
Your browser does not support the audio element.
|
| 66 |
+
</audio>
|
| 67 |
+
</td>
|
| 68 |
+
<td>
|
| 69 |
+
<audio controls>
|
| 70 |
+
<source src="https://huggingface.co/pharoAIsanders420/micro-musicgen-jungle/resolve/main/assets/output_7.wav?download=true" type="audio/wav">
|
| 71 |
+
Your browser does not support the audio element.
|
| 72 |
+
</audio>
|
| 73 |
+
</td>
|
| 74 |
+
<td>
|
| 75 |
+
<audio controls>
|
| 76 |
+
<source src="https://huggingface.co/pharoAIsanders420/micro-musicgen-jungle/resolve/main/assets/output_5.wav?download=true" type="audio/wav">
|
| 77 |
+
Your browser does not support the audio element.
|
| 78 |
+
</audio>
|
| 79 |
+
</td>
|
| 80 |
+
</tr>
|
| 81 |
+
</table>
|
| 82 |
+
|
| 83 |
+
## Benchmarks
|
| 84 |
+
|
| 85 |
+
|
| 86 |
+
## Usage
|
| 87 |
+
|
| 88 |
+
Install my [audiocraft](https://github.com/facebookresearch/audiocraft) fork:
|
| 89 |
+
|
| 90 |
+
```
|
| 91 |
+
pip install -U git+https://github.com/aaronabebe/audiocraft#egg=audiocraft
|
| 92 |
+
```
|
| 93 |
+
|
| 94 |
+
Then, you should be able to load this model just like any other musicgen checkpoint here on the Hub:
|
| 95 |
+
|
| 96 |
+
```python
|
| 97 |
+
import torchaudio
|
| 98 |
+
from audiocraft.models import MusicGen
|
| 99 |
+
from audiocraft.data.audio import audio_write
|
| 100 |
+
model = MusicGen.get_pretrained('pharoAIsanders420/micro-musicgen-acid')
|
| 101 |
+
model.set_generation_params(duration=10)
|
| 102 |
+
wav = model.generate_unconditional(10)
|
| 103 |
+
|
| 104 |
+
for idx, one_wav in enumerate(wav):
|
| 105 |
+
# Will save under {idx}.wav, with loudness normalization at -14 db LUFS.
|
| 106 |
+
audio_write(f'{idx}', one_wav.cpu(), model.sample_rate, strategy="loudness", loudness_compressor=True)
|
| 107 |
+
```
|
| 108 |
+
|
| 109 |
+
## Dataset
|
| 110 |
+
|
| 111 |
+
I sourced the datasets from these royalty free sources:
|
| 112 |
+
|
| 113 |
+
- [https://www.samplescience.info/2022/05/abstract-303.html](https://www.samplescience.info/2022/05/abstract-303.html)
|
| 114 |
+
- [https://www.musicradar.com/news/free-303-samples](https://www.musicradar.com/news/free-303-samples)
|
| 115 |
+
- [https://www.musicblip.com/product/c-am-114bpm-418mb-61loops/](https://www.musicblip.com/product/c-am-114bpm-418mb-61loops/)
|
| 116 |
+
- about 1 hour of different sequences I recorded from my Behringer TD-3.
|