Audio Separator Pro

Audio Separator Pro

Splits a song or recording into stems such as vocals, instruments and drums with over 50 models, with batch runs and a side by side model comparison. It suits musicians, DJs, karaoke makers and audio editors.

Made by nomadkaraoke on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: Audio Separator Pro is an enhanced edition of the Audio Separator Space that splits audio into separate parts, called stems, such as vocals and instruments. It offers more than 50 models across five families: Roformer, MDX23C, MDX-NET, VR and Demucs. It adds batch runs, a model comparison mode and three colour themes.

Who it suits: People who want to pull vocals, backing tracks, drums or other parts out of a recording without installing separation software.

Industry | Who uses it | How they use it | Example
Music | Producers and DJs | Split a track into stems for remixing | A 4 stem or 6 stem split
Entertainment | Karaoke makers | Remove the lead vocal to make a sing along track | A pop song with vocals out
Media | Podcasters and editors | Clean reverb or noise from a recording | A live recording
Archiving | Restoration hobbyists | Reduce noise on old vinyl or tape transfers | A digitised record
Education | Music students | Isolate drums or wind instruments to study them | A drum separation

If you only need to trim silence or convert a file format, you can skip it.

Where to use it:

• A web browser, on the Space's page at Hugging Face
• Windows, macOS or Linux with Python, from a downloaded copy

Use case | Who | What they get back
Karaoke and instrumentals | Karaoke makers | A track with the lead vocal taken out
Vocal isolation | Producers | A clean vocal stem
Batch separation | Editors with many files | A combined ZIP of all outputs
Model comparison | Audio engineers | Results from two to four models side by side

How to set it up:

• Python on your computer
• Python packages from requirements.txt, with Gradio 6.19.0
• No account or key is named

1. Install the Hugging Face hub package.
2. Download the Space with the hf command.
3. Move into the downloaded folder.
4. Install Gradio 6.19.0, then the requirements.
5. Start app.py.

pip install -U huggingface_hub
hf download nomadkaraoke/audio-separator --repo-type space --local-dir audio-separator
cd audio-separator
pip install gradio==6.19.0
pip install -r requirements.txt
python app.py

How to use it: Upload or drag in an audio file, pick a use case preset, which chooses a suitable model for you, and optionally change the model from the dropdown. Then choose WAV, MP3, FLAC, OGG or M4A and run it. Advanced parameters can be switched on for finer control.

python app.py

Tool | What it does
Use case presets | Pick a goal such as karaoke, de-reverb or a 4 stem split and the model is chosen for you
Batch mode | Runs many files with the same settings and offers one ZIP download
Comparison mode | Runs two to four models on one file for a side by side check
separate_batch, compare_models, model_metadata, stats | Server endpoints for batch runs, comparison, model details and status

Example: You upload a pop song and choose the karaoke preset. The app returns the song split into two stems, so you can use the one without the lead vocal as a sing along track.

For newcomers and experts: Start with one short song and the karaoke preset, since the preset picks a model for you. Compare the result with a different model only after you hear what the first one gives.

Presets map to model families. Roformer handles most two stem jobs, Demucs handles four and six stem splits, MDX-NET offers a fast instrumental option for CPU only use, MDX23C covers drum separation and VR covers wind instruments. Each model carries quality, speed and size notes, and comparison mode picks a best model from the quality scores.

You upload one file or a batch
A preset or model is chosen
The separator splits the audio into stems
Stems come back, as a ZIP for batches

Good to know:

• More than 50 models across five architectures, with quality, speed and size notes
• Three themes: dark, light and a retro one
• License: mit
• Cost to run: none named
• Data it reaches: the audio files you upload, plus a processing history kept by the app
• Not for: the source names no limits

On Hugging Face, 13 November 2026

• Likes: 9
• Built with: gradio
• License: mit

Links

• Space: https://huggingface.co/spaces/nomadkaraoke/audio-separator
• Files: https://huggingface.co/spaces/nomadkaraoke/audio-separator/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.