What does Voice Extractor do?

It removes background noise, music and other non-speech sounds from a recording and returns only the voice. It has no settings, so you only wire in audio and run it.

Can I clean only part of a recording?

Voice Extractor always processes the whole input. To clean one part, cut it first with Trim Audio and wire the trimmed audio into Voice Extractor.

What is the difference between Voice Extractor and Audio Separation?

Voice Extractor keeps only a clean voice and discards everything else. Audio Separation splits any audio into a vocal track and an instrumental track, or into stems such as drums and bass, so you keep every part.

Does Voice Extractor work on music without vocals?

It needs speech or singing to extract. Purely instrumental audio gives little or nothing back.

Voice Extractor node — remove background noise and isolate the voice in a recording

Remove background noise, music and other sounds from a recording and keep only the voice. Clean audio before transcription, lip sync, dubbing or a voice change.

The Voice Extractor node removes background noise, music and other sounds from a recording and returns only the voice. It uses ElevenLabs voice isolation and has no settings to tune. Put it in front of a voice, lip sync or transcription node, which all work better with clean speech.

When to use it

Clean an interview or a field recording before you edit it.
Clean noisy audio before Transcribe, for a more accurate transcript.
Take the vocals out of music or mixed audio before a voice change.
Remove background noise from a podcast recording.
Isolate the dialogue of a video before Dubbing.

Quick start

Add the node

Press Tab on the canvas and choose Audio › Clean & Separate › Voice Extractor.

Wire the recording

Wire an audio node, such as Upload Audio, into the Audio input. For the sound of a video, wire the video through Extract Audio first.

Run it

Click Run on the node. The clean voice appears on the node with a player.

A noisy interview is cleaned before it is transcribed.

Input and output

Input	Accepts	What it does
Audio	Audio nodes, such as Upload Audio, Reference Audio and Text to Speech	The recording to clean.

The output, Audio, is the cleaned recording with the voice alone. It is audio, not data: for word timings, use Transcribe or Forced Alignment.

Settings

Voice Extractor has no settings besides its Label, the name shown on the canvas.

Models

Model	Maker	Modes	Credits	Details
ElevenLabs Voice Isolation	ElevenLabs	Voice isolation	74	Strip background noise / music from a vocal track.

Which cleaning node to use

You want	Use
Only a clean voice, with noise and music removed	Voice Extractor
The vocals and the instrumental of any song, as two tracks	Audio Separation, Vocal / Instrumental mode
Separate stems, such as drums, bass and piano	Audio Separation, Full stems mode
The vocals and stems of a song made with Suno	Suno Separate

Tips

Trim first. The node always processes the whole input. Cut the part you need with Trim Audio before you clean it.
Clean before voice work. Put Voice Extractor in front of Voice Changer or Lip Sync so that background noise does not carry into the result.
Expect limits on buried speech. The result depends on the input. Speech that is deeply buried under noise can come back with artifacts.

From the API

POST /v1/audio-isolation takes audioUrl. The MCP tool is audio_isolation. See Voice and media.

Voice Extractor