Mel-Band RoFormer Deux

Mel-Band RoFormer Deux

A dual Mel-Band RoFormer model from becruily that splits a song into vocals and instrumental, with an SDR score listed for each. For non-commercial audio work.

Made by becruily on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: This repository holds mel-band-roformer-deux, an audio model that splits a song into a vocal track and an instrumental track. It is built on the Mel-Band RoFormer design and made by becruily. The card calls it a dual model, so one release covers both outputs.

Who it suits: It suits people who want to pull the singing or the backing music out of a track they have the right to use. The card is short, so check the author's page for more.

Industry | Who uses it | How they use it | Example
Music | Producers | Pull a vocal or a backing track from their own mix | Making a karaoke-style backing track
Content | Creators | Reuse the music of a track they own without the voice | Adding a backing track to a video
Education | Music students | Listen to the vocal and the backing part on their own | Studying how a song is built
Research | Audio researchers | Compare separation scores across models | Checking the two scores in the card

If you need a model for commercial work, you can skip it, since the license is non-commercial.

Where to use it: The card names no program. You load the files in an audio separation tool that supports the Mel-Band RoFormer design, so check that your tool lists it.

• An audio separation tool that supports this design
• Windows, macOS or Linux with Python, to download the files

Use case | Who | What they get back
Vocal track | Producers | A track with the singing pulled out
Instrumental track | Creators | A track with the backing music left
Score check | Researchers | Two scores to compare with other models

How to set it up: You need Python and enough disk space for the model files. The card shows only the download step, so how you run it depends on your tool.

1. Install the Hugging Face download tool.
2. Download the files from the repository.
3. Point your separation tool at them.

pip install -U huggingface_hub
hf download becruily/mel-band-roformer-deux

How to use it: The card gives no run command. After the download, you load the files in your separation tool and open a song you have the right to use. The tool then writes a vocal track and an instrumental track.

Example: You run the two commands above, load the files in your tool and open a song you own. The tool gives you one file with the singing and one file with the backing music.

For newcomers and experts: Start with a short song and listen to both outputs. If the vocal track still carries some backing sound, compare it with another model before you decide.

For experts, the card gives only the two SDR scores, with no training details or test set. Compare it with other models on your own audio before you rely on it.

You download the model files
You load them in a separation tool
You open a song you may use
The model splits it into two parts
You get a vocal track and an instrumental track

Good to know: The card lists one score for each part. Here SDR is a common way to rate how clean a separated track is, and a higher number is better.

Title: SDR scores in the card
Instrumental: 17.5466 SDR
Vocals: 11.3695 SDR

• License: CC BY-NC 4.0, which is non-commercial
• Cost to run: none named
• Data it reaches: the audio you give it, run locally, and the files you download
• Not for: commercial use, or audio you have no right to use

On Hugging Face, 13 November 2026

• Downloads in the last 30 days: 391
• Likes: 47
• License: cc-by-nc-4.0

Links

• Model: https://huggingface.co/becruily/mel-band-roformer-deux
• Files: https://huggingface.co/becruily/mel-band-roformer-deux/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download becruily/mel-band-roformer-deux

Then follow the usage in its model card. Nothing in this kit runs by itself.