Audio Stem Separator (AI)
Split any song into vocals, drums, bass and other instruments with AI, 100% in your browser.
Drag your song here or click to select
MP3, WAV, OGG, FLAC, M4A, AAC · no artificial size limit
Built by
Miguel Ángel Colorado Marin (MACM)
Built by
Miguel Ángel Colorado Marin (MACM)
Full-Stack Developer · Guadalajara, España
I develop web apps, digital tools and full projects — from design to deployment.
The stem separator uses HTDemucs, Meta AI Research's open-source audio source-separation model, exported to ONNX and running entirely inside your browser — your song is never uploaded to any server. Upload any song (MP3, WAV, OGG, FLAC, M4A or AAC) and within a few minutes you get 4 independent stems: vocals, drums, bass and the rest of the instruments. Listen to them separately with the built-in mixer (per-stem solo/mute, all in sync), or download any of them as lossless WAV — perfect for making a karaoke version by removing the vocals, isolating the bass or drums to study a song, or extracting an acapella. The first time you use it the AI model (~300 MB) downloads and stays cached in your browser, so following separations are much faster. It's the same technology paid karaoke and remix tools use, running on your own device, at no cost and with your audio never leaving your computer or phone.
Features
- ✓Meta's HTDemucs model (open source)
- ✓4 stems: vocals, drums, bass and other
- ✓Mixer with per-stem solo/mute
- ✓Download each stem as lossless WAV
- ✓100% in browser, your audio is never uploaded
How to split a song into stems?
- 1
Upload your song
Drag the file or click to select it. No artificial size limit.
- 2
Tap Separate
The first time it downloads the AI model (~300 MB); after that it stays cached.
- 3
Listen with the mixer
Use per-stem solo/mute to hear vocals, drums, bass or other on their own.
- 4
Download the stems
Each stem downloads as lossless WAV, ready to use.
Frequently asked questions
Is my song uploaded to a server?
No. The HTDemucs model runs inside your browser. The audio never leaves your device.
How long does splitting a song take?
It depends on your device and whether WebGPU acceleration is available — anywhere from close to real time to several times the song's length on more modest devices using WebAssembly. A progress bar shows real-time progress.
Why is the model ~300 MB?
HTDemucs is a large neural network (a hybrid spectrogram/waveform architecture with a transformer). It downloads once and stays cached in your browser — subsequent uses don't need to download it again.
How good is the separation quality?
HTDemucs is one of the best open-source audio source-separation models available, comparable to paid tools. Even so, no separation AI is perfect: slight bleed from other stems can remain, especially in very dense mixes.
Can I use it to make karaoke tracks?
Yes — separate the song, download every stem except vocals, and load them into your favorite audio or video editor to mix them back together without the voice.
Related tools
Embed Audio Stem Separator (AI) on your site
Add Audio Stem Separator (AI) to any web page with a simple iframe. Free, with attribution to miguelacm.es.
<iframe
src="https://miguelacm.es/embed/audio-stem-separator"
width="100%"
height="700"
frameborder="0"
title="Audio Stem Separator (AI) — miguelacm.es"
></iframe>