audiocraft-webui v2.0!

Local web UI for Facebook's Audiocraft model: https://github.com/facebookresearch/audiocraft

Features:

• Long Audio: Generate audio as long as you like.
• Processing Queue: Add multiple prompts to the queue and process them automatically.
• Generation History: Replay generated audio and view the settings used for each.

Install:

If you'd like GPU acceleration and do not have torch installed, visit PyTorch Installation Guide for instructions on installing GPU-enabled torch correctly.

Option 1: Manual Installation (Classic Method)

If you prefer manual installation, install dependencies using:

pip install -r requirements.txt

(If you encounter errors with audiocraft, please refer to their official documentation).

Option 2: Using the Install Script (Automated Setup)

To automate installation and environment setup, use the provided script:

• Linux/macOS:

  ./install.sh

• Windows:

  install.bat

This will check for Python 3.10, create a virtual environment (venv), and install all required dependencies automatically. If Conda is available, it can be used instead of venv.

---

Run:

Option 1: Manual Run (Classic Method)

Start the web UI manually using:

python webui.py

Option 2: Using the Run Script (Automated Execution)

Alternatively, use the run script for easier execution:

• Linux/macOS:

  ./run.sh

• Windows:

  run.bat

This will automatically activate the appropriate environment (virtualenv or Conda) and start webui.py. Once the script is stopped, it ensures the environment is properly deactivated.

---

There's no need to download any external models. Simply pick a model from the dropdown menu, and when you run it for the first time, it will be automatically downloaded via Audiocraft.

If you want to use Melody Mode, select the Melody model, and an option to upload your melody audio file will appear.

---

Notes:

• Generated files are saved in the static/audio/ directory.
• The currently active model remains loaded in memory by default. To unload it after each generation, launch with:

  python webui.py --unload-after-gen

• The UI could use an improved design—contributors are welcome!

---

Parameters:

• Top-K: Affects the variety of generated audio. Higher values lead to more diverse results but may increase randomness.
• Top-P: Recommended around 0.7. Controls sampling diversity, with lower values producing more stable results.
• Duration: Length of generated audio.
• CFG (Classifier Free Guidance): Determines how strictly the output follows the prompt (recommended 3-5).
• Temperature: Controls randomness. Higher values (1.05-1.5) create more chaotic results, while lower values yield structured outputs.

---

Troubleshooting:

• Torch Installation Issues: Ensure you have the correct CPU/GPU version installed (PyTorch Guide).
• Virtual Environment Issues: If activation fails, try:

  source venv/bin/activate  # Linux/macOS
  venv\Scripts\activate    # Windows

• Conda Not Found? Ensure Conda is installed and added to PATH. Verify by running:

  conda --version

• Installation Script Issues? Ensure scripts have execution permissions:

  chmod +x install.sh run.sh

---

Changelog:

Feb-25-2024:

• Rewrote everything.
• Added generation history for audio outputs.
• Removed outdated dependencies.
• Removed deprecated parameters (overlap and segments).