Update app.py
Browse files
app.py
CHANGED
|
@@ -1,8 +1,57 @@
|
|
| 1 |
-
|
| 2 |
-
|
| 3 |
-
import gradio as gr
|
| 4 |
-
import requests
|
| 5 |
import tempfile
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 6 |
|
| 7 |
# Preset audio URLs
|
| 8 |
INPUT_ROOT = 'https://adasp.telecom-paris.fr/rc-ext/demos_companion-pages/vqvae_examples/'
|
|
@@ -137,7 +186,7 @@ with gr.Blocks(title="VQ-VAE Timbre Transfer", theme=gr.themes.Soft()) as demo:
|
|
| 137 |
|
| 138 |
Transfer the timbre (tone/texture) from one audio source to another while preserving the musical content.
|
| 139 |
|
| 140 |
-
**Content**: Musical notes/melody that will be preserved
|
| 141 |
**Style**: Instrument timbre/texture that will be applied
|
| 142 |
|
| 143 |
### How It Works:
|
|
@@ -230,5 +279,16 @@ with gr.Blocks(title="VQ-VAE Timbre Transfer", theme=gr.themes.Soft()) as demo:
|
|
| 230 |
outputs=[output_audio, status_msg]
|
| 231 |
)
|
| 232 |
|
| 233 |
-
|
| 234 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os
|
|
|
|
|
|
|
|
|
|
| 2 |
import tempfile
|
| 3 |
+
import requests
|
| 4 |
+
import gradio as gr
|
| 5 |
+
import torch
|
| 6 |
+
import librosa
|
| 7 |
+
import numpy as np
|
| 8 |
+
import subprocess
|
| 9 |
+
import sys
|
| 10 |
+
|
| 11 |
+
def install_dependencies():
|
| 12 |
+
"""Install required packages for deployment"""
|
| 13 |
+
try:
|
| 14 |
+
if not os.path.exists('ss-vq-vae'):
|
| 15 |
+
print("Cloning ss-vq-vae repository...")
|
| 16 |
+
subprocess.run(['git', 'clone', 'https://github.com/cifkao/ss-vq-vae.git'], check=True)
|
| 17 |
+
|
| 18 |
+
subprocess.run([sys.executable, '-m', 'pip', 'install', './ss-vq-vae/src'], check=True)
|
| 19 |
+
print("Dependencies installed successfully!")
|
| 20 |
+
|
| 21 |
+
except Exception as e:
|
| 22 |
+
print(f"Error installing dependencies: {e}")
|
| 23 |
+
raise
|
| 24 |
+
|
| 25 |
+
# Install dependencies for deployment
|
| 26 |
+
try:
|
| 27 |
+
install_dependencies()
|
| 28 |
+
import confugue
|
| 29 |
+
from ss_vq_vae.models.vqvae_oneshot import Experiment
|
| 30 |
+
except ImportError:
|
| 31 |
+
print("ss-vq-vae not found. Please install manually or run in Colab.")
|
| 32 |
+
sys.exit(1)
|
| 33 |
+
|
| 34 |
+
def download_model():
|
| 35 |
+
"""Download model files if they don't exist"""
|
| 36 |
+
model_dir = 'ss-vq-vae/experiments/model'
|
| 37 |
+
os.makedirs(model_dir, exist_ok=True)
|
| 38 |
+
|
| 39 |
+
model_path = os.path.join(model_dir, 'model_state.pt')
|
| 40 |
+
if not os.path.exists(model_path):
|
| 41 |
+
print("Downloading model...")
|
| 42 |
+
url = 'https://adasp.telecom-paris.fr/rc-ext/demos_companion-pages/vqvae_examples/ssvqvae_model_state.pt'
|
| 43 |
+
response = requests.get(url)
|
| 44 |
+
with open(model_path, 'wb') as f:
|
| 45 |
+
f.write(response.content)
|
| 46 |
+
print("Model downloaded successfully!")
|
| 47 |
+
|
| 48 |
+
# Initialize model
|
| 49 |
+
download_model()
|
| 50 |
+
logdir = 'ss-vq-vae/experiments/model'
|
| 51 |
+
cfg = confugue.Configuration.from_yaml_file(os.path.join(logdir, 'config.yaml'))
|
| 52 |
+
exp = cfg.configure(Experiment, logdir=logdir, device='cpu')
|
| 53 |
+
exp.model.load_state_dict(torch.load(os.path.join(logdir, 'model_state.pt'), map_location=exp.device))
|
| 54 |
+
exp.model.train(False)
|
| 55 |
|
| 56 |
# Preset audio URLs
|
| 57 |
INPUT_ROOT = 'https://adasp.telecom-paris.fr/rc-ext/demos_companion-pages/vqvae_examples/'
|
|
|
|
| 186 |
|
| 187 |
Transfer the timbre (tone/texture) from one audio source to another while preserving the musical content.
|
| 188 |
|
| 189 |
+
**Content**: Musical notes/melody that will be preserved
|
| 190 |
**Style**: Instrument timbre/texture that will be applied
|
| 191 |
|
| 192 |
### How It Works:
|
|
|
|
| 279 |
outputs=[output_audio, status_msg]
|
| 280 |
)
|
| 281 |
|
| 282 |
+
gr.Markdown("""
|
| 283 |
+
### 🔧 Troubleshooting
|
| 284 |
+
- **Poor transfer quality?** Try different instrument combinations or adjust max duration
|
| 285 |
+
- **Audio doesn't load?** Check internet connection or try different presets
|
| 286 |
+
- **Processing slow?** Reduce max duration or try shorter audio clips
|
| 287 |
+
|
| 288 |
+
### 📖 Citation
|
| 289 |
+
Original work by Ondřej Cífka (InterDigital R&D and Télécom Paris, 2020).
|
| 290 |
+
Demo by Ali Dulaimi.
|
| 291 |
+
""")
|
| 292 |
+
|
| 293 |
+
if __name__ == "__main__":
|
| 294 |
+
demo.launch(share=True, debug=True, height=1400)
|