Audio separation

AI voice isolator, live and on demand

Isolate the voice from music, crowd and noise, and keep the background as its own stem. From a file or a live feed, for music, sports and broadcast.

CH 01 Race
Original mix
Channel
Commentary
0 dB
Crowd + engines
0 dB

  • Batch

    Files and folders, in Studio or by API

  • Live

    Live feeds in real time. Enterprise, on request

Live broadcasts powered by CAMB.AI

Ligue 1
FanCode
Australian Open
NASCAR
MLS
YES
Ligue 1
FanCode
Australian Open
NASCAR
MLS
YES
Ligue 1
FanCode
Australian Open
NASCAR
MLS
YES
Ligue 1
FanCode
Australian Open
NASCAR
MLS
YES

CAMB.AI already runs live multilingual broadcasts for sports, including Major League Soccer and the Australian Open. See our sports work

Two stems

One mix in. Two stems out.

Real-world audio is never clean. Commentary sits under 60,000 fans, a singer under the band, a reporter under the traffic.

CAMB.AI splits any audio or video into the foreground, with speech or vocals, and the background, with music, crowd, noise and ambience. Keep the voice, keep the atmosphere, or use both.

Live · Enterprise

Real-time voice isolation for live broadcasts

Connect your live feed and get the voice and the background back as two separate streams, in real time. We set it up with your team, on your own feed.

  • Voice and background as two live streams
  • Any language, several speakers at once
  • Straight into CAMB.AI live translation and dubbing
Race feedMove the level to bring the crowd back0:00
Commentary
Crowd + engines

Side by side

Cleaner than AudioShake and ElevenLabs on all 8 measures

CAMB.AI

Tap a stem to hear it

Less noise between words

 vs AudioShake

 vs ElevenLabs

Benchmark results

Cleaner voice, less leakage

CAMB.AI, AudioShake, ElevenLabs and LALAL.AI on the same seven real-world clips. CAMB.AI comes first on every separation measure.

The cleanest voice of the four

2.8× AudioShake and 1.7× ElevenLabs on how cleanly the voice is pulled from the mix.

SI-SDR, estimated by SQUIM: how much of the separated voice is the voice and not what was around it.

CAMB.AI2.84 dB
AudioShake1.01 dB
ElevenLabs1.64 dB
LALAL.AI2.81 dB

CAMB.AI2.84 dB
LALAL.AI2.81 dB
ElevenLabs1.64 dB
AudioShake1.01 dB

The cleanest voice of the four

2.8× AudioShake and 1.7× ElevenLabs on how cleanly the voice is pulled from the mix.

SI-SDR, estimated by SQUIM: how much of the separated voice is the voice and not what was around it.

CAMB.AI0.376%
AudioShake0.437%
ElevenLabs0.447%
LALAL.AI0.529%

The least voice left behind

14% less than AudioShake, 16% less than ElevenLabs, 29% less than LALAL.AI: a background stem you can really reuse.

Average probability that speech is still heard in the background stem.

CAMB.AI1.895
LALAL.AI1.852
ElevenLabs1.544
AudioShake1.527

The voice that sounds best

24% above AudioShake and 23% above ElevenLabs on how good the separated voice sounds.

PESQ, estimated by SQUIM: perceived speech quality, without a reference recording.

CAMB.AI0.744
LALAL.AI0.740
ElevenLabs0.726
AudioShake0.697

The voice that is easiest to understand

Every word stays clear, ahead of all three.

STOI, estimated by SQUIM: how intelligible the separated speech is.

CAMB.AI0.258
LALAL.AI0.255
ElevenLabs0.229
AudioShake0.209

The most voice in the voice stem

23% above AudioShake and 13% above ElevenLabs.

Average CLAP speech-focus gap: how clearly the foreground stem is about speech.

MeasureCAMB.AIAudioShakeElevenLabsLALAL.AI
Separation quality (SI-SDR)2.84 dB1.01 dB1.64 dB2.81 dB
Voice left in the background (Leakage)0.376%0.437%0.447%0.529%
Perceptual quality (PESQ)1.8951.5271.5441.852
Intelligibility (STOI)0.7440.6970.7260.740
Voice kept in the foreground (CLAP gap)0.2580.2090.2290.255
DNSMOS speech quality (SIG)2.6622.3872.5232.684
DNSMOS background quality (BAK)2.8842.3072.5382.973
DNSMOS overall quality (OVRL)2.3031.9232.0942.327

Internal benchmark: CAMB.AI, AudioShake, ElevenLabs and LALAL.AI on the same seven clips from Wikimedia Commons (gaming commentary, a film trailer, live music, an outdoor interview, a live newsroom, archival TV and a red carpet interview), averaged. SQUIM and DNSMOS are reference-free estimates. On the three DNSMOS scores LALAL.AI is slightly ahead; CAMB.AI leads every other measure.

Use cases

Made for the teams that work with sound

Music labels and production

Music labels and production

Pull a clean acapella and instrumental from finished masters, for sync, karaoke, remixes and catalog work.

Explore
Sports

Sports

Lift commentary and interviews out of stadium crowds, live or after the match.

Explore
Motorsport

Motorsport

Engine, wind and pit-lane noise out of onboard audio, team radio and interviews.

Explore
Media and broadcast

Media and broadcast

Clean dialogue for dubbing, subtitles, news and archives, and a background stem ready for new languages.

Explore
Remove music from video

Remove music from video

Keep every word and drop the soundtrack from clips, highlights and re-edits.

Explore

Live and batch

Every way you work with audio

Live separationEnterprise

Your live feed, back as two separate streams in real time: the voice and the background. For broadcasts, live sports, events and streams. An enterprise feature, set up with your team on request.

Talk to us

Batch separation

Upload .mp4, .mov, .mp3, .wav, .flac or .aac, one file or a whole folder. Download both stems as FLAC, WAV, MP3 or AAC. 22.6 seconds to process a 30-second video.

Audio separation API

Create a separation task, check its status and get the results, one file at a time or in bulk. The same two stems, inside your own workflow.

POST /apis/audio-separationGET  /apis/audio-separation/{task_id}GET  /apis/audio-separation-result/{run_id}POST /apis/audio-separation-results

Your audio stays yours

CAMB.AI is SOC 2 compliant and doesn't train its models on your audio. Your past runs are in your History, and you can delete them whenever you want.

Separate once, use it everywhere

The clean voice stem goes straight into the rest of CAMB.AI: subtitles, translation and AI dubbing, on files or live. The background stem keeps the original music and atmosphere under the new voice.

Who it's for

Built for every team that works with voice

Music labels and publishers

Vocal and instrumental versions from catalogs that never had stems.

Sports leagues and broadcasters

Clean commentary and a separate atmosphere stem, live and after the match.

Newsrooms and post-production

Dialogue pulled out of streets, crowds and wind, ready to edit.

Localization teams

A clean voice to translate and a background stem to dub over.

Developers

Two stems from one API call, one file or thousands.

Creators and social teams

Clips without their soundtrack, ready for a new one.

How it works

From one mix to two stems in four steps

01

Upload or connect

Add a file or a folder. Enterprise teams can also connect a live feed.

02

Separate

CAMB.AI splits the voice from everything else, in any language.

03

Get two stems

Download the foreground and the background, or receive them as two live streams.

04

Use them anywhere

Send the voice to subtitles, translation and dubbing, and keep the background under the new voice.

Pricing

Pick your hours. Both stems in every plan.

Every plan gives you the voice and the background as two stems, in Studio and by API. Try any plan free for 3 days, with 30 minutes of audio.

10 h
Billing

Solo

$40/month

$480 billed yearly

3 hours of audio a month

For creators and hobbyists. Pull the voice out of songs, podcasts and videos in minutes.

Start free trial
  1. Today 30 minutes to try. $0 today.
  2. Day 3 Your plan starts: $480 a year.

Session

Recommended

$130/month

$1,560 billed yearly

10 hours of audio a month

For producers and content creators working every week. Enough for full albums, podcasts and soundtracks.

Start free trial
  1. Today 30 minutes to try. $0 today.
  2. Day 3 Your plan starts: $1,560 a year.

Catalog

$375/month

$4,500 billed yearly

30 hours of audio a month

For professionals and teams at volume. Clean voice and background stems for music, post-production and localization.

Start free trial
  1. Today 30 minutes to try. $0 today.
  2. Day 3 Your plan starts: $4,500 a year.

Enterprise

High volumes, live separation on your own feeds, or separation inside your own workflow.

  • Live separation, in real time
  • Volume pricing
  • SLA and dedicated support
  • Custom delivery and storage

FAQ

Questions about audio separation

An AI voice isolator separates the voice from everything else in a recording. Most return only the voice. CAMB.AI's audio separation gives you two stems: the foreground (speech or vocals) and the background (music, noise, crowd and ambience).

Stem separation splits a mix into stems, and many stem splitters aim for four: vocals, drums, bass and other. Voice isolation usually returns only the voice. CAMB.AI's audio separation returns two stems, the voice and the background, so you can use both.

Two: foreground and background. Multiple speakers stay together in the foreground stem. Drums, bass or other instruments are not split into separate stems.

Yes, as an enterprise feature. Talk to us and we set it up on your own feed: you get the voice and the background as two separate streams in real time.

Upload .mp4 and .mov video, or .mp3, .wav, .flac and .aac audio. Download each stem as FLAC, WAV, MP3 or AAC.

Yes. Separation works on speech in any language.

Yes. You can create tasks, check their status and get results one by one or in bulk.

Separating audio doesn't give you any rights to it. Make sure you have the rights you need before you publish.

Hear the difference on your own audio

  • SOC 2 compliant
  • Your audio isn't used to train our models
  • Batch and live