Eleven Voice Isolator — clean speech from a noisy recording
Street noise, music under the dialogue, an echoing room: Voice Isolator strips it all away and leaves only the voice. Run it on a clip to get the clip back with clean sound, or on an audio file of up to 30 min.
ElevenLabs: expressive speech in 70+ languages, dialogue, music written to your cut, sound effects, dubbing and captions — the sound layer for every clip.
Open your galleryUse it through MCP
Voice isolation runs on a clip or audio file you already have: open it in your gallery and pick the clean audio action.
What it does
- Noise, music and reverb outTraffic, wind, hum, background music and room echo are removed. The voice stays.
- The picture staysA clip goes in and the same clip comes out with clean sound. Audio files come back as audio.
- Long recordingsPodcasts, interviews and lectures of up to 30 min in one run.
- Better captions and dubbingClean speech makes transcripts, captions and dubbing more accurate further down the line.
Specification
- Provider
- ElevenLabs
- Provider model
audio-isolation- Limits on Mira
- source up to 30 min
- Output
- mp3, mp4
- Where it works
- Gallery: an action on your clip or audio fileMCP tools:
isolate_voiceMira for Adobe panel (After Effects, Premiere Pro)
Cost in credits
4 credits per minute
Every started unit is billed in full.
| Example | Credits |
|---|---|
| 1 min of source audio | 4 |
| 10 min of source audio | 40 |
| 30 min of source audio | 120 |
The composer always shows the current price before you generate.
Questions about Eleven Voice Isolator
What does it remove?
Background noise, music, reverb and any other sound that is not speech.
How long can a file be?
Up to 30 min per run.
Can it pull the vocals out of a song?
For music, use stem separation in Eleven Music instead: it splits a track into vocals and instrumental, or into six stems.
How much does it cost?
4 credits per minute of audio. The table above has worked examples.
Can an AI agent clean audio for me?
Yes: over the Mira MCP server, isolate_voice takes a generation or an uploaded file and returns the cleaned version.