#SESA Audio separation

1 messages · Page 1 of 1 (latest)

lilac ore
#

🎵 Sharing My Passion Project: SESA Audio Separation! 🎙️

Hey everyone! I’m thrilled to share a project I’ve been working on solo: SESA Audio Separation. This tool separates audio into vocals, instrumentals, drums, bass, and more, with quality enhancement features. I don’t know coding, so I built this with AI help. I’d love your feedback! 😊

What’s SESA Audio Separation About?
I made this to make audio separation easy for music production or audio processing fans. Despite no coding skills, AI helped me add features to experiment with audio creatively.

  • Audio Separation: Split audio into vocals, instrumentals, drums, bass, etc., using models like Vocal Models.
  • At the same time, a phase fixer is also available, allowing you to obtain audio with less residual noise.
  • all newly made models are available (some of the older models are also available
  • Apollo Quality Enhancement: Boost audio with MP3 Enhancer, Lew Vocal Enhancer, or Apollo Universal Model (Normal or Mid/Side methods).
  • Auto & Manual Ensemble: Combine model outputs for better results using average, median, min, or max methods.
  • Google Drive Integration: Copy files to Google Drive for sharing.
  • User-Friendly Interface: Built with Gradio, featuring progress bars and logs.
  • Output Formats: Export in WAV, FLAC, MP3, and more.
    Tools & Tech (With AI Help)

I relied on AI since I don’t code. Here’s what was used:

  • Languages & Libraries: Python, Gradio (UI), Librosa, Soundfile (audio processing), PyTorch (model inference).
  • Other Tools: Apollo for enhancement, subprocess for scripts.
  • Features: Multi-language support with I18nAuto, progress tracking with tqdm.

Colab: https://colab.research.google.com/drive/1CH2JWd6YculmKSug9zpzuxvM6mSYysdB?usp=sharing

Note: Auto ensemble might have a few bugs, but I’m doing my best to fix them.

image.png image.png image.png image.png image.png
Google Colab
lilac ore
#

Update:

  • Ensemble error partially resolved
  • Date: April 12, 2025
lilac ore
#

Future update:

  • Adaptation to Kaggle
  • Developing Windows software
lilac ore
#

Next update:

  • Model training feature will be added.
#

If you have any desired feature or idea, you can mention it; if it's a very useful idea, I'll add it.

lilac ore
#

Update

  • Naming Improvements: We've refined the naming conventions for better clarity and consistency.
  • Favorites Feature: You can now star your favorite models! They’ll always appear at the top for quick access.
  • Persistent Settings: No need to reselect your changes. Your previous selections and settings will now be saved automatically.
  • Auto Ensemble Presets: A new preset option for Auto Ensemble allows you to easily and quickly select your preferred models.

Note: You’ll need to change the model category once for the update to take effect.

lilac ore
#
  • Matchering feature added. (Not great, but decent enough)
lilac ore
#
  • SESA Audio Separation now saves the separated audio to the drive.