FROM THE KULTURE ARCHIVE07 / KULTURE LAB
Free AI stem splitter.
Find the voice. Isolate the bass. Hear what makes the track move.
AI STEM SPLITTER · BETA
Choose the sound you need.
Choose the part you want, then load your song. Preview and download only your selected result as MP3 or WAV.
One result means less export time and memory. The AI still analyses the full mix, so separation itself may take several minutes.
Up to 10 minutes / 60 MB. All-stems mode: up to 3 minutes. Desktop recommended. Your audio stays on your device. Starting downloads an AI model (about 166 MB); processing may take several minutes and uses significant memory. Keep this tab open.
One separation at a time. Your device does the processing; choose a single stem for longer tracks. Use Cancel to stop the job.
Choose an audio file to begin.
Powered by Demucs / StemSplit ONNX and ONNX Runtime. Open source credits & licences. AI separation is an estimate; some sounds may bleed between stems.
TAKE THE TRACK APART
Remove vocals online.
Keep the part you need.
Project Kulture’s free AI stem splitter separates a song into vocals, drums, bass and other instruments. Choose Remove vocals / instrumental for a backing track, or Vocals only / a cappella to extract the voice. Listen to the result, then download it as MP3 or WAV.
The audio is processed in your browser and stays on your device. There’s no account or audio upload. Start with a short, clear recording and choose one stem if that’s all you need.
Try the stem splitter ↑How to remove vocals from a song
Choose your output.
Select the instrumental option to remove the singing. For vocal isolation, select the vocals-only option. You can also choose bass, drums, other instruments or all stems.
Load an audio file.
Choose an MP3 or WAV from your device. The tool reads it locally. Keep files within 60 MB and 10 minutes; all-stems mode is limited to 3 minutes to reduce memory use.
Separate, listen, download.
Press Create selected stem and keep the tab open. The first run downloads an AI model of about 166 MB. Processing can take several minutes. Preview the stem before saving an MP3 or WAV.
An instrumental, an acapella or the rhythm section?
A vocal remover and a stem splitter use the same separation process here. The difference is the part of the song you choose to keep.
| Your goal | Choose | What you get |
|---|---|---|
| Remove the singer | Remove vocals / instrumental | Drums, bass and other instruments combined as one backing track. |
| Extract an acapella | Vocals only / a cappella | The estimated vocal part, including backing vocals the model captures. |
| Study the low end | Bass only | The estimated bass part for hearing notes, rhythm and phrasing. |
| Hear the groove | Drums only | The estimated drum part for studying breaks and patterns. |
| Hear the harmony | Other instruments only | A combined group that can include guitar, piano, synths and other sounds. |
| Explore the whole arrangement | All stems + instrumental | Four separated parts plus a combined instrumental, with a 3-minute limit. |
“Other instruments” is one combined output. This version doesn’t give you separate guitar, piano and synth tracks. For the difference between separated audio and original recording files, read stems vs multitracks.
Make room for your own voice.
Use the instrumental maker to prepare a karaoke backing track or rehearse a vocal part. Play the preview through headphones first: vocal reverb or harmonies may remain in the instrumental.
For remix ideas, the acapella extractor lets you hear phrasing without the full arrangement. Work with music you own or have permission to reuse; separating a recording doesn’t grant publishing or sampling rights.
Learn what makes the track move.
Isolate a drum break to hear its timing, or solo the bass to understand how it works against the kick. A stem can help you listen closely even when it isn’t clean enough for a finished release.
Build on what you hear with the beat maker, or read how to mix kick and bass without losing impact.
Get a cleaner vocal separation.
Start with the best source you have. A clear WAV or good-quality MP3 gives the model more useful detail than a distorted recording or a file that’s been compressed repeatedly. Converting a poor recording to WAV won’t restore missing detail.
Dense arrangements, live crowd noise, heavy reverb and layered voices can leave bleed or a watery sound. That’s why the tool includes a preview. Check quiet sections as well as the chorus, and compare the result at a similar volume to your original.
AI stems are estimates made from a finished mix. They don’t recreate the untouched tracks from the original session. For an important edit, listen to each transition before exporting your final mix.
Your song stays with you.
The separation model runs on your device rather than a shared audio-processing server. The page downloads the model and software it needs, but it doesn’t send your selected audio file for processing. Downloads are created locally in the browser.
Desktop Chrome or Edge is recommended. Use one separation at a time, close heavy apps if memory is tight, and keep this tab open. Choose a single output for longer tracks. Cancel stops the active job; the tool releases its AI worker after a period of inactivity.
Stem splitter questions
Is this stem splitter free?
Yes. Project Kulture’s stem splitter is free to use and doesn’t require an account. It runs in your browser. The first separation downloads an AI model of about 166 MB, so you need an internet connection and a device with enough memory.
How do I remove vocals online and keep the instrumental?
Choose Remove vocals / instrumental, load your audio file and press Create selected stem. The tool combines the estimated drums, bass and other instruments into one backing track. Preview it, then download MP3 or WAV. Some vocal reverb or backing vocals may remain.
Can I use it as an acapella extractor?
Yes. Choose Vocals only / a cappella to extract the estimated vocal part of your song. You can preview it and download it as MP3 or WAV. Lead vocals, harmonies and reverb may be grouped together; instruments can sometimes bleed into the result.
Which files and song lengths can I use?
MP3 and WAV are good starting formats. Other audio formats depend on your browser’s decoder. The file limit is 60 MB. A single-output job accepts up to 10 minutes of audio; All stems + instrumental accepts up to 3 minutes to reduce memory use. Video files and streaming links aren’t supported inputs.
Why is the stem splitter taking a long time?
Your device downloads and runs the AI model. Separation may take several minutes, depending on the track and your hardware. Keep the tab open and try desktop Chrome or Edge. A shorter clip uses less memory. Choosing one output reduces export work, although the AI still analyses the full mix.
Does my audio get uploaded or stored on Kulture’s server?
No. Your selected audio is decoded and separated on your device, and the resulting downloads are created in your browser. The page fetches its software and model files over the internet. It doesn’t upload your selected audio to Kulture for separation.
Can I get separate guitar and piano tracks?
This version separates vocals, drums, bass and other instruments. Guitar, piano, synths and similar sounds share the Other instruments output. It doesn’t offer individual guitar or piano extraction, and separated stems aren’t the original multitrack session files.