Saturalabs / Voice & video
AI voice isolator. Bring the words forward.
Separate a speaking voice from a mixed recording using a simple text description. For interviews, podcasts, and voiceovers where the conversation matters more than the sound around it.
Start with your recordingThe track you’re looking for
The requested voice, separated from the rest of the recording.
Name one audible source. You can edit this prompt before running the separation.
Uploading opens the editor. Downloads require an account and credits. View pricing.
A focused approach
Voice isolation starts with choosing what to keep
A recording can contain speech, music, traffic, and room noise at the same time. Saturalabs uses your description to identify the sound you want to separate. Instead of naming every distraction, start by naming the voice: ‘The main person speaking’ is a useful first prompt for a single-speaker recording.
The result has two tracks. Isolated sound contains the requested source; Background audio contains the remaining mix. To keep speech, listen to Isolated sound. To keep the scene without that speech, audition Background audio. These are different uses of the same separation, and choosing the correct track matters as much as the prompt.
From mix to source
How to use the voice isolator
Start with the original recording
Upload an MP3, WAV, MP4, or MOV file up to 50 MiB (52.43 MB). Use your original export when possible, rather than audio that has already been heavily filtered or compressed.
Describe the audible voice
Name one source. Try ‘The person speaking in the foreground’ or ‘The narrator’s spoken voice.’ Describe what you can hear; a person's name alone may not distinguish them in a mix.
Compare speech and background
Listen to Isolated sound for the voice and Background audio for what was left behind. Check quiet words, breaths, and the moments where music or noise overlaps speech.
Use the voice in your edit
Download the separated voice as WAV when the result fits your project. Keep the original available so you can blend it back in if the isolated version sounds too dry or loses detail.
Be specific about the sound
A prompt for the result you want
Describe the source, then choose the output that matches your goal. These are starting points to try with your own recording.
| Your goal | Try this prompt | Keep this track |
|---|---|---|
| Keep a foreground speakerA clear starting point when one voice dominates the recording. | The main person speaking | Isolated sound |
| Keep narration over musicListen for music bleeding through during pauses and sustained words. | The narrator’s spoken voice | Isolated sound |
| Reduce a steady fanHere the target is the unwanted sound. Keep the remaining track and check that speech survived. | The steady fan noise | Background audio |
Made for a real next step
Put the separated audio to work
Podcast edits
Pull a host's voice forward when a recording includes music or a distracting environment. Review softer syllables before cutting the isolated track into your episode.
Interview excerpts
Prepare a spoken excerpt for an edit without carrying the entire surrounding soundscape. Overlapping speakers still need careful listening.
Voiceover recovery
Try separating narration from a rendered mix when the original voiceover file is unavailable. Compare against the mix so punctuation and sentence endings stay intact.
Listen before you commit
Three checks worth making
Compare a quiet passage and a busy passage with the original, at a similar listening level.
Word endings
Consonants can be quieter than vowels. Replay the ends of sentences and check that words remain intelligible.
Overlapping speech
Two people talking together can be difficult to separate. A generic ‘speech’ prompt may retain both voices.
Natural tone
Compare the voice at a similar listening level to the original. Notice metallic textures, missing breaths, or a hollow sound.
Know the limits
What a voice isolator can and cannot fix
Isolation separates audible sources; it does not recreate words that were never captured clearly. Severe clipping, a distant microphone, heavy echo, and simultaneous speakers can limit the result. The extracted voice may retain traces of the background or introduce audible artifacts.
Voice isolation also differs from transcription, voice cloning, and speaker identification. Saturalabs separates sound from your uploaded recording. It does not generate a new speaker or promise to select a named person from a crowded conversation.
Can I isolate voice from an audio file online?
Yes. Upload MP3 or WAV audio, describe the speaking voice, and review Isolated sound. Saturalabs also accepts MP4 and MOV video files, with a 50 MiB (52.43 MB) upload limit.
Is a voice isolator the same as a vocal remover?
They serve different goals. A voice isolator keeps spoken speech. A vocal remover usually separates singing so you can keep the instrumental backing. Choose a prompt that describes the actual source in your file.
Can I isolate one person from several speakers?
You can try describing an audible characteristic that distinguishes the desired voice, but a clean single-speaker result is not guaranteed. Overlapping voices and similar speaking tones are especially difficult.
Will the voice sound exactly like the original?
Not necessarily. Source separation can change tone or leave artifacts, particularly in dense recordings. Compare quiet passages and loud overlaps before using the output in a final edit.
Do I need credits to download the voice?
Downloads require an account and sufficient credits. The results screen shows the download cost before you download. Check the pricing page for the current credit options.
Start with a sound you know
Bring your own recording.
Describe one source. Listen to both tracks. Keep what works for your project.