At a glance
- Accepts
- MP3, WAV, M4A, AAC, OGG, Opus, FLAC, and the audio track of a video
- Outputs
- Two 44.1 kHz stereo WAV files, one voice and one music
- Where it runs
- Entirely in your browser, on your device
- What leaves your device
- Nothing - the audio never leaves your device; the accurate mode downloads model weights to you
- Limits
- AI mode downloads a model once: Fast quality (Spleeter, 38 MB) is quick on any device but leaves more bleed between voice and music; Best quality (HT-Demucs, 158 MB) is cleaner but slow without WebGPU and can run out of memory on phones; both models are trained on sung vocals, so spoken word is out of distribution; the instant mode only works on a steady musical bed
- Account
- Not required - there is no sign-up
- Price
- Free, with no watermark and no usage cap
- Made by
- Visrut
- Last updated
- 7 November 2026
Splitting a voice out of background music is not a filtering problem, even though it sounds like one. Speech and music occupy the same part of the spectrum: a speaking voice has its fundamental around 85 to 255 Hz, most of what makes it intelligible sits between 300 Hz and 4 kHz, and its consonants reach past 10 kHz. Guitars, piano, strings and snare live in exactly that band. Any filter wide enough to keep the voice keeps most of the instruments with it.
The two approaches
- Background subtraction, instant. A musical bed holds roughly the same spectral shape for minutes at a time while a voice constantly moves. Take the median level of each frequency across the whole file and you have a decent estimate of the bed; subtract it and what is left leans heavily towards voice. Good on a steady backing track, poor on music that changes as much as the speech does.
- A separation model, in two sizes. Fast quality is Spleeter 2-stems (38 MB), Best quality is HT-Demucs fine-tuned for vocals (158 MB), both run in your browser through ONNX Runtime. They learned what a voice sounds like rather than relying on any rule about frequency, which is why they handle cases the first method cannot. Fast takes seconds; Best is cleaner but much slower without WebGPU. Downloaded once, then cached.
How to use it
- Drop in an audio file, or a video to pull its audio track.
- Pick the AI separation (Fast or Best quality) or the instant background subtraction, which costs nothing and tells you quickly whether your file is an easy case.
- Play both results in the page, then download either as WAV.
What it will not do
The model was trained on sung vocals, so spoken word over music is slightly outside its training. It usually still works on dialogue and podcast beds, but that is the case worth testing rather than assuming. Neither method can invent detail that was masked in the original recording, so a voice buried under a loud mix comes back with audible artefacts. And the music track is derived by subtracting the voice from your original, which means it inherits whatever the voice track got wrong.
Sources
Spleeter's vocal SDR on MUSDB18 (6.55 dB with soft masking, for its 4-stem model): Hennequin et al., JOSS 2020. HT-Demucs FT median vocal SDR on MUSDB18-HQ (9.19 dB): the model card of the ONNX export used here, based on Rouard, Massa and Défossez, 2022. Our own test numbers use the simple SDR of the MDX challenge, so they are not directly comparable. Checked October 2026.
Frequently asked questions
Is my audio uploaded to a server?
No. The audio is decoded and separated in your browser tab. The only network request this tool makes is downloading the separation model itself, which is a download, not an upload - your file never leaves your device.
Can you separate voice from music just by filtering frequencies?
Not really. Speech sits at roughly 85-255 Hz for the fundamental with intelligibility up to about 4 kHz and consonants to 10 kHz, and music covers all of that same range. A filter that keeps the voice keeps most of the instruments too. Removing the very low bass and the very high cymbal air helps a little, but it is cleanup rather than separation.
What does the fast mode actually do?
It estimates the steady musical background from the median level of each frequency over time, then subtracts it. A musical bed holds the same spectral shape for minutes while a voice moves around, so the median is mostly the bed. It works on a steady backing track and falls apart when the music changes as much as the voice does.
What is the difference between Fast and Best quality in the AI mode?
They are two neural networks. Fast is Spleeter 2-stems from Deezer, a 38 MB download that splits a 3-minute file in seconds on a recent computer and is light enough for phones. Best is HT-Demucs fine-tuned for vocals, a 158 MB download that is cleaner but needs much more memory and is slow without WebGPU. Fast leaves more music in the voice track and more voice in the music track. Published vocal scores on the MUSDB18 benchmark are about 6.6 dB SDR for Spleeter (its 4-stem model) and 9.19 dB for HT-Demucs FT, higher being cleaner; in our own test on a song with known stems, Fast scored 5.8 dB and Best 8.7 dB. Both are cached after the first download.
Does it work on speech, or only singing?
The model is trained on sung vocals, so spoken word over music is slightly outside what it learned. In practice it usually still separates dialogue and podcast beds well, but it is the one case where it is worth trying the file before relying on it.
Why do the two output files sum back to my original?
The music track is calculated as your original minus the isolated voice, rather than separately. That preserves the balance between the two. If separation creates peaks above full scale, both tracks are lowered equally to prevent distortion, so layering them back gives a quieter version of the original.
Can I get one file without the vocals and another without the piano?
The first, yes: the music file is your audio with the voice taken out, and the voice file is the voice with the music taken out. A file with only the piano removed is not something this tool can make. Both of its methods split audio into exactly two parts, voice and everything else, so the piano stays in the music file with every other instrument. We tried the only open model small enough for a browser that has a separate piano output (Demucs htdemucs_6s). On a test of solo piano mixed with a speaking voice, its 'no piano' version still had much of the piano in it, which matches its authors' own warning about piano quality, so we left it out rather than offer a result that does not work.
What formats can I put in?
Anything your browser can decode: MP3, WAV, M4A, AAC, OGG, Opus, FLAC, and the audio track of a video file. Output is 44.1 kHz stereo WAV.