Audio Isolation with UVR5

Ultimate Vocal Remover (UVR) is a powerful, free, and open-source tool for separating audio sources. It’s an essential tool for creating the clean, isolated vocal tracks needed for high-quality voice model training.

Installation

You can run UVR on your local computer or on Google Colab.

Local Installation

1. Download the installer from the UVR website.
2. Run the installer to set up UVR on your computer.
3. Once installed, launch UVR and click the wrench icon (🔧) to open the settings.
4. Go to the Download Center and download the separation models you want to use.

Note

When you run the installer, Windows may flag it as potential malware. This is a false positive due to the installer not being signed with a commercial license. You can review the source code on their GitHub repository if you have any concerns.

Tip

To use RoFormer models, you’ll need the Roformer patch, listed as “UVR Roformer Version Patch Only” (or “Full Install” if you haven’t installed UVR yet).

How to Use UVR

1. Select Your Audio File: Choose the audio file you want to process.
2. Choose an Output Path: Select a folder where the separated audio files will be saved.
3. Select a Model: Choose the separation model you want to use. Different models are optimized for different tasks.
4. Click Convert: Click the Convert button to start the separation process.

Google Colab

1. Open the UVR5 NO UI Colab Notebook.
2. In your Google Drive, create a folder named Music for your input audio files and a folder named Vocales for the output files.
3. Run the Installation cell to set up UVR5.
4. Once the installation is complete, run the cell for whichever separation method you want to use.

Available Models

UVR has a wide variety of models available for different tasks. Here are some of the most popular ones organized by category.

Click to expand the list of recommended models

RoFormer Models

These models are currently the best-performing models for most tasks.

Model Name	Best For	Description	Download Link
Gabox Fv4 Mel-Roformer	Vocals	Clean, non-muddy vocals with a good noise/fullness balance, though it can miss some backing vocals	Model / Config
Gabox Fv7 Mel-Roformer	Vocals	Captures harmonies well with minimal bleed, but fullness varies by track and can sound a bit noisy	Model / Config
Gabox and Aufr33 Mel Roformer Small Karaoke	Main vs. Backing Vocals	The best model for separating main vocals from backing vocals	Model / Config
Anvuew BS-Roformer Dereverb	Reverb Removal	The best model for removing stereo reverb from audio	Model / Config
Aufr33 Mel Roformer Denoise 1-2	Noise Removal	The best model for removing background noise from audio	[V1] Model Config / [V2] Model Config

MDX-Net Models

A prize-winning model framework that is great for a variety of tasks.

Model Name	Best For	Description	Download Link
MDX23C-InstVoc HQ	Multiple Tasks	A set of models that work together to deliver high-quality results	Built-in
UVR-MDX-NET Voc FT	Vocals	A model focused specifically on separating vocals	Built-in
UVR-MDX-NET Crowd HQ 1	Background Vocals	Separates background vocals from main vocals	Built-in

VR Arch Models

A location-guided model that delivers great results in most tasks.

Model Name	Best For	Description	Download Link
UVR-DeEcho-DeReverb	Reverb & Echo	Separates reverb and echo from audio	Built-in
5_Hp-Karaoke-UVR	Background Vocals	Another excellent model for separating background vocals	Built-in
UVR-DeNoise	Noise Removal	Removes noise and some artifacts from audio	Built-in

Demucs Models

An all-in-one model capable of separating audio into 4 to 6 stems.

Model Name	Best For	Description	Download Link
htdemucs_ft	General Purpose	Slower processing but better results than the standard htdemucs model	Built-in
htdemucs_6s	Instrument Separation	Capable of separating guitar and piano stems in addition to vocals and drums	Built-in

Note

New models come out frequently, so this list may not always be up to date. For the latest models and more detailed, community-tested information, check out this community guide.