fabiogra/moseca

• Setup
  • Local environment
  • Docker
  • (Optional) Preprocess samples
• About
  • High-Quality Stem Separation
  • Advanced AI Algorithms
  • Karaoke Fun
  • Easy Deployment
  • Open-Source and Free
  • Support
• FAQs
  • What is Moseca?
  • Are there any limitations?
  • How does Moseca work?
  • How do I use Moseca?
  • Where can I find the code for Moseca?
  • How can I get in touch with you?
• Disclaimer

Create a new environment with Python 3.10 and install the requirements:

pip install -r requirements.txt

set the PYTHONPATH to the root folder:

export PYTHONPATH=path/to/moseca

download the vocal remover model:

wget --progress=bar:force:noscroll https://huggingface.co/fabiogra/baseline_vocal_remover/resolve/main/baseline.pth

then run the app with:

streamlit run app/header.py

You can also run the app with Docker:

docker build -t moseca .
docker run -it --rm -p 7860:7860 $(DOCKER_IMAGE_NAME)

or pull the image from Hugging Face Spaces:

docker run -it -p 7860:7860 --platform=linux/amd64 \
	registry.hf.space/fabiogra-moseca:latest

You can set the following environment variables to limit the resources used by the app:

• ENV_LIMITATION=true
• LIMIT_CPU=true

If you want to preprocess the samples used in the demo, you need to set the env variable PREPARE_SAMPLES=true as a secret (create the file run/secrets/PREPARE_SAMPLES with true value inside).

If you run locally you also need to move the files inside scripts to the root folder and run prepare_samples.sh.

cp scripts/ .
chmod +x prepare_samples.sh
python prepare_samples.sh

Welcome to Moseca, your personal web application designed to redefine your music experience. Whether you're a musician looking to remix your favorite songs, a karaoke enthusiast, or a music lover wanting to dive deeper into your favorite tracks, Moseca is for you.

Separate up to 6 stems including 🗣voice, 🥁drums, 🔉bass, 🎸guitar, 🎹piano (beta), and 🎶 others.

Moseca utilizes state-of-the-art AI technology to extract voice or music from your original songs accurately.

Engage with your favorite tunes in a whole new way!

Moseca offers an immersive online karaoke experience, allowing you to search for any song on YouTube and remove the vocals online.

Enjoy singing along with high-quality instrumentals at the comfort of your home.

With Moseca, you can deploy your personal Moseca app in the or locally with in just one click.

You can also speed up the music separation process by with GPU support.

Moseca is the free and open-source alternative to lalal.ai, splitter.ai or media.io vocal remover.

You can modify, distribute, and use it free of charge. I believe in the power of community collaboration and encourage users to contribute to our source code, making Moseca better with each update.

• Show your support by giving a star to the GitHub repository .
• If you have found an issue or have a suggestion to improve Moseca, you can open an
• Enjoy Moseca?

Moseca is an open-source web app that utilizes advanced AI technology to separate vocals and instrumentals from music tracks. It also provides an online karaoke experience by allowing you to search for any song on YouTube and remove the vocals.

Yes, in this environment there are some limitations regarding lenght processing and CPU usage to allow a smooth experience for all users.

If you want to remove these limitations you can deploy a Moseca app in your personal environment like in the or locally with

Moseca utilizes the Hybrid Spectrogram and Waveform Source Separation (DEMUCS) model from Facebook. For fast karaoke vocal removal, Moseca uses the AI vocal remover developed by tsurumeso.

1. Upload your file: choose your song and upload it to Moseca. It supports a wide range of music formats for your convenience.

2. Choose separation mode: opt for voice only, 4-stem or 6-stem separation depending on your requirement.

3. Let AI do its magic: Moseca’s advanced AI will work to separate vocals from music in a matter of minutes, giving you high-quality, separated audio tracks.

4. Download and enjoy: preview and download your separated audio tracks. Now you can enjoy them anytime, anywhere!

The code for Moseca is readily available on GitHub and Hugging Face.

For any questions or feedback, feel free to contact me on or LinkedIn.

Moseca is designed to separate vocals and instruments from copyrighted music for legally permissible purposes, such as learning, practicing, research, or other non-commercial activities that fall within the scope of fair use or exceptions to copyright. As a user, you are responsible for ensuring that your use of separated audio tracks complies with the legal requirements in your jurisdiction.