Ultimate Vocal Remover separates vocals and instruments from songs with local AI models.
Ultimate Vocal Remover is an open-source desktop application for separating vocals and instrumental parts from audio. It can produce karaoke-style instrumentals, isolated acapellas, or individual stems, with processing handled locally rather than through an online service. The program is developed by Anjok07 and aufr33 and is released under the MIT license.
For daily work such as preparing a practice track, isolating a vocal, or exploring a song’s parts, Ultimate Vocal Remover brings several separation approaches into one interface. Its choice of models makes it useful when one result needs a different separation method from another, though the interface is dense and may take time to navigate.
Choosing a separation method
The available model families are MDX-Net, Demucs, and VR Architecture. They can be used to extract vocals, create instrumentals, or split a song into stems. Many bundled models were trained by the UVR team; the Demucs v3 and v4 four-stem models are exceptions. Having multiple model families gives users options for different source-separation tasks without changing applications.
UVR also includes tools for changing pitch and time-stretching audio with Rubber Band. These adjustments can help adapt a processed track for a different key or tempo after separation. The central purpose remains source separation, so the practical workflow is to choose a model for the desired output and then apply pitch or timing changes if needed.
Local processing and hardware
All processing runs offline on the user’s computer. CPU conversion is available, so an NVIDIA graphics card is not required, but CPU processing is slower. For GPU conversion, the listed minimum is an NVIDIA GTX 1060 6GB, and 8 GB of VRAM is recommended; NVIDIA acceleration is the faster option described for the stable setup.
AMD Radeon and Intel Arc users can try a separate DirectML build, but it is marked beta. That distinction matters for anyone who depends on accelerated processing: the stable GPU path is centered on NVIDIA, while support for these other GPU families is not presented as a stable build.
Installation and maintenance considerations
The Windows installer bundles Python, PyTorch, and the required dependencies, avoiding a separate setup of those components. The installation guidance specifies the main C: drive, because installing to a secondary drive can cause instability. Users with limited disk space should account for the installation before starting, rather than treating UVR as a small utility.
The last stable installer dates from 2023, while the next major version remains in development. This means the program’s broad selection of models and offline workflow come with a maintenance caveat: the stable installer is not recent, and users should weigh that when planning a long-term setup.
Who may find it useful
Musicians, DJs, and karaoke fans can use UVR to prepare instrumentals, acapellas, or separate stems on a personal computer. Its combination of model choices and local processing suits users who want control over the separation workflow and do not want to send audio to an online service. The strongest fit is for people comfortable working through a crowded interface and, for faster conversions, using compatible NVIDIA hardware.
Ultimate Vocal Remover is a practical free, open-source option for local vocal and stem separation, with CPU processing available and faster conversion possible on supported NVIDIA GPUs. Its main trade-offs are the dense interface, the age of the stable installer, and the beta status of AMD and Intel GPU acceleration.
