Separate vocals and instrumentals for free
Separate vocals and instrumentals for free with local AI. Preview both tracks and download WAV files in your browser, without uploading your audio.
Choose your audio
Auto will use CPU because a suitable discrete GPU could not be confirmed. You can still select GPU manually.
CPU processing can take several times the audio length. Start with a short preview; you can cancel at any time.
Checking your browser and looking for recent results…
Model and local storage
The model downloads only when you start (about 63.7 MB).
Estimated temporary space: 297.0 MB. Browser space available: 0 KB.
Your input and separated tracks are processed locally and are not uploaded. The browser downloads static model and runtime files from Hugging Face and jsDelivr by default. Clearing the model cache does not delete your audio results.
Desktop preview: one file at a time, mono or stereo, up to 10 minutes. Separation may leave instruments or alter vocals; this is not lossless restoration. Storage is temporary and may be cleared by your browser.
Model: Kim Vocal 2 — Kimberley Jensen · MIT. Runtime: ONNX Runtime Web · MIT. Model permission · Third-party notices
How it works
Extract vocals from a song in three steps
Choose a file already on your device. The model downloads when you start, while the audio stays in your browser for separation and export.
Choose your audio
Select or drop a mono or stereo MP3, M4A, WAV, FLAC or Ogg file. Check the duration and listen to the original before processing.
Preview, then separate
Try the first 30 seconds or process up to 10 minutes. Auto chooses CPU unless the browser can confirm a suitable GPU; you can also choose either mode manually.
Compare and save both tracks
Switch between the original, vocals and instrumental at the same playback position. Download each track as WAV or pass it to another local audio tool.
Use the vocal and instrumental tracks
One run produces both stems. Listen before saving: source mixing and model limitations affect how clean each track sounds.
Isolate singing for edits
Keep the vocal stem for a rough acapella, timing edits or a new arrangement. Backing voices may remain with the lead.
Make a practice backing track
Use the instrumental stem for rehearsal or karaoke practice. Faint vocal remnants can remain, especially with layered or reverberant music.
Prepare the next audio task
Send either stem to the converter or trimmer. You can also try transcribing the vocal track, although separation does not guarantee better lyrics recognition.
Input formats, output and limits
The browser must be able to decode the audio inside the file; a supported filename extension alone does not guarantee success.
| Input | What the tool does | Keep in mind |
|---|---|---|
| MP3, M4A/AAC, Ogg | Decode the audio, then separate both stems locally. | Codec support varies by browser and file. |
| WAV, FLAC | Decode mono or stereo audio and prepare it for the model. | Large uncompressed files need more browser space. |
| Vocal + instrumental WAV | Export two 44.1 kHz, 16-bit stereo files. | Use Audio Converter if you need MP3 or another format. |
Current desktop preview: one file at a time, up to 500 MB input and 10 minutes of audio. A preview processes only the first 30 seconds. Multi-channel audio, video files and live microphone input are outside this tool's current scope.
What AI separation can and cannot recover
The model estimates the vocal part of a mixed recording. It may leave instruments in the vocal track, leave quiet voices in the backing track, or change the sound of the voice. It does not identify one singer, remove arbitrary noise or recreate the original studio stems.
For a cleaner result, start with a well-mixed file and compare both tracks before using them. A longer file does not make separation more accurate; try the 30-second preview first.
Frequently asked questions
Is my audio uploaded?
No. Your selected file and the separated tracks are handled in your browser. The browser downloads the static AI model and runtime files when needed; those requests do not include your audio.
Does this remove vocals completely?
Not always. Reverb, backing singers and instruments with similar frequencies can appear in either stem. Listen to the result before you use it for karaoke or editing.
Can I save MP3 instead of WAV?
This tool exports two WAV files. Use the linked Audio Converter to create MP3 or another supported format. Converting does not recover detail lost in the source recording.
Why is GPU processing unavailable?
Auto uses CPU if the browser cannot offer a suitable GPU. You can try a current desktop Chrome or Edge browser; CPU processing may take much longer.
What happens when I refresh or leave?
Recent completed tracks can be restored for about 10 minutes in the same browser. Leaving this tool releases its files unless a download or another tool still needs them. Save important results to your device.
Can I separate a video or a long song?
First extract the audio from a video with Audio Extractor. For audio longer than 10 minutes, trim it into shorter files. Each file must still meet the 500 MB input limit.
Continue with another audio tool
These tools can help prepare an input file or work with a separated track in the same browser.
Audio Extractor
Take an audio track from a video before separating vocals.
Audio Converter
Convert a downloaded WAV stem to MP3 or another supported format.
Audio Trimmer
Cut a section from the original or a separated track.
Audio Volume Booster
Make a quiet recording louder or reduce an overly loud track directly in your browser. Adjust gain, protect peaks, preview the result, and export a WAV copy.
Audio to Text
Try transcribing the vocal stem; recognition quality depends on the recording.
Audio Noise Remover
Reduce background noise in voice recordings for free. Preview, compare and download a WAV file with local AI in your browser, without uploading audio.