weierstrass-walker
commited on
Commit
•
955a729
1
Parent(s):
d36098a
Update README.md
Browse files
README.md
CHANGED
@@ -58,7 +58,7 @@ These are the codes for the bird call generation score sde model.
|
|
58 |
|
59 |
<code>core-sde-sampler.py</code> will execute the sampler. The sampler uses pretrained weight to generate bird calls. The weight can be found [here](https://github.com/mandelbrot-walker/Birdcall-score-sde/blob/main/ckpt.flax)
|
60 |
|
61 |
-
For using different sample generation parameters,
|
62 |
```bash
|
63 |
python main.py --sigma=25 --num_steps=500 --signal_to_noise_ratio=0.10 --etol=1e-5 --sample_batch_size = 128 --sample_no = 47
|
64 |
```
|
@@ -68,7 +68,7 @@ In order to generate the audios, these dependencies are required,
|
|
68 |
pip install librosa
|
69 |
pip install soundfile
|
70 |
```
|
71 |
-
In order to train the model from scratch, please generate the dataset using this [link](www.kaggle.com/ibraheemmoosa/birdsong-spectogram-generation). The dataset is generated in kaggle. Therefore, during training your username and api key is required in the specified section.
|
72 |
```bash
|
73 |
python main.py --sigma=35 --n_epochs=1000 --batch_size=512 --lr=1e-3 --num_steps=500 --signal_to_noise_ratio=0.15 --etol=1e-5 --sample_batch_size = 64 --sample_no = 23
|
74 |
```
|
|
|
58 |
|
59 |
<code>core-sde-sampler.py</code> will execute the sampler. The sampler uses pretrained weight to generate bird calls. The weight can be found [here](https://github.com/mandelbrot-walker/Birdcall-score-sde/blob/main/ckpt.flax)
|
60 |
|
61 |
+
For using different sample generation parameters change the argument values. For example,
|
62 |
```bash
|
63 |
python main.py --sigma=25 --num_steps=500 --signal_to_noise_ratio=0.10 --etol=1e-5 --sample_batch_size = 128 --sample_no = 47
|
64 |
```
|
|
|
68 |
pip install librosa
|
69 |
pip install soundfile
|
70 |
```
|
71 |
+
In order to train the model from scratch, please generate the dataset using this [link](www.kaggle.com/ibraheemmoosa/birdsong-spectogram-generation). The dataset is generated in kaggle. Therefore, during training your username and api key is required in the specified section inside the script.
|
72 |
```bash
|
73 |
python main.py --sigma=35 --n_epochs=1000 --batch_size=512 --lr=1e-3 --num_steps=500 --signal_to_noise_ratio=0.15 --etol=1e-5 --sample_batch_size = 64 --sample_no = 23
|
74 |
```
|