Vocal SeparatorNEW
Split a song into an instrumental track and an extracted vocal, from a file or a Douyin / TikTok link
Checking your access…
AllFreeTool Vocal Separator splits a song into two files on our server: an instrumental track with the vocals removed, and a dry vocal track with the accompaniment removed. Upload an audio or video file, or paste a Douyin, WeChat Channels or TikTok share link — the backend downloads, decodes and separates it with the open-source MDX-Net vocal model, then hands you both 44.1 kHz stereo WAV files.
Cover recordings, karaoke backing tracks, practising an instrument and editing stems all start with pulling the voice out of a mix. This tool runs the very MDX-Net models UVR5 uses on our own server: your media is never forwarded to a third party, you do not have to download a clip and upload it again, and in a minute or two you get a clean vocal and a clean instrumental.
Why use Vocal Separator on AllFreeTool?
Two Outputs at Once
One separation run returns both the instrumental (vocals removed) and the vocals (extracted) as WAV files.
Works from Links
Paste a Douyin, WeChat Channels or TikTok share link and the server fetches the audio for you.
MDX-Net Models
Switch between Kim_Vocal_2, UVR-MDX-NET-Voc_FT and Inst_HQ_4 — all open-source UVR models.
Live Progress
Progress is reported chunk by chunk, with a clear stage message while the model runs.
How to use Vocal Separator
Pick a Source
Upload a local audio or video file, or switch to link mode and paste a Douyin, WeChat Channels or TikTok share link.
Choose a Model
Kim_Vocal_2 for general use, Voc_FT for cleaner vocals, or Inst_HQ_4 when the instrumental matters most.
Wait and Download
When the separation finishes, two download buttons appear — one for the instrumental, one for the vocals.
Common questions
How are Douyin, WeChat Channels and TikTok links handled?＋
The server calls the bundled yt-dlp binary to pull the media stream out of the link, and falls back to reading the media URL straight out of the share page if that fails. Only the official Douyin, TikTok and WeChat domains are accepted, which keeps the feature from being abused to reach internal network resources.
Which file is the “vocals removed” version?＋
“Instrumental (vocals removed)” keeps the instruments and background sound but drops the singing — that is the one to use as karaoke backing. “Vocals (extracted)” keeps the singer and drops everything else, which is what you want for covers, duets and practising.
Is the result as good as UVR5?＋
The tool uses the same open-source MDX-Net models and the same chunking parameters as UVR5 (overlap 0.25, hop 1024, 44.1 kHz stereo), so the processing matches. What differs is the material itself: the more a vocal overlaps the accompaniment in frequency, and the more complex the backing is, the more bleed you will hear in the result.
How long can one file be?＋
Up to 10 minutes per run and 60 MB per upload. Separation takes roughly one to two times the length of the audio, and the page keeps showing progress, so please leave it open until it finishes.
Is my audio sent to a third party?＋
No. The file only goes to our own server, the model runs there too, and nothing is forwarded to any external service. Temporary files are cleaned up automatically once the job is over.