Separate voice, music and sound effects
One mixed track in, up to six clean stems out
A finished soundtrack is a single file: dialogue, footsteps and the score baked together. Seed Audio 1.5 can pull it apart again, returning each voice, each effect and the music as its own track, ready to drop into an editor and mix again.
Based on ByteDance's documentation. We describe our own results on the translation pages; we have not yet run separation on our films.
What comes out
ByteDance's own example splits a short fantasy scene into six tracks. It shows the level of detail to expect: not just “voice” and “music”, but each speaker and each sound.
Each speaker
In the example, the narrator, a girl and a boy each arrive as a separate voice track.
Each effect
Footsteps on one track, a door closing on another, so you can move or replace one without touching the rest.
The music
The background score on its own, clean enough to keep under a new voice.
What stems are good for
Dubbing and translation
Translate the voice and keep the original music and effects underneath. Having the voice alone is exactly how we prepared our 12-language film.
Fixing one element
Turn the music down under a line, remove a cough, replace a weak effect, without re-recording the scene.
Remixes and edits
Recut a trailer, a podcast clip or a short with the score running on its own stem.
Archive material
Old videos where the original project files are gone: rebuild the stems from the final mix.
How to split a soundtrack
Export the audio
Take the mixed audio out of the video as WAV or MP3. Keep each file between 2 seconds and 6 minutes.
Split long files
Longer soundtracks go in as parts. Cut at silence so no sound is torn in half.
Separate
The model returns up to six tracks: voices, effects and music.
Rebuild the mix
Bring the stems into your editor at the same start time and they line up with the picture.
Input and output
| Parameter | Limit |
|---|---|
| Input | One audio file |
| Formats | WAV or MP3 |
| Length | 2 to 360 seconds |
| File size | Up to 90 MB |
| Stems returned | Up to 6 |
| Output formats | WAV (default), MP3, PCM, OGG Opus |
Source: ByteDance Seed Audio 1.5 early access integration guide, October 2026.
Good to know
- Input is audio, not video. Extract the soundtrack first.
- Six stems is the ceiling per request.
- One file per request, up to 6 minutes and 90 MB. Split longer soundtracks at silence.
- Separate only material you have the rights to edit.
Stem separation FAQ
What is stem separation?
Splitting a mixed soundtrack back into its parts: voices, sound effects and music, each on its own track, so they can be edited or mixed again independently.
How many stems does Seed Audio 1.5 return?
Up to six. ByteDance's example returns a narrator, two other voices, the music, footsteps and a door as six separate tracks.
Can I separate audio from a video?
Separation takes an audio file. Export the soundtrack from your video as WAV or MP3 first, up to 6 minutes and 90 MB per file.
Can I remove the voice and keep the music?
Yes. Once the voice is on its own stem you can mute it, replace it or translate it, and keep the music and effects as they were.