What Voice Isolator is
Inside the sprawling ElevenLabs platform — voice synthesis, dubbing, agents — sits a small tool with one sentence of ambition: remove everything from a recording that is not the voice. Background noise, music, street rumble, interference; in goes the file, out comes isolated speech. It is available in the browser and through the API, and it is the tool we mean when a guide on this site says "one-shot web rescue."
The billing unit is the platform's credits: 1,000 credits per minute of audio processed. The free plan's 10,000 monthly credits therefore buy roughly ten minutes of isolation; the $6 Starter plan about thirty. Simple enough once you know the conversion — and the reason a calculator appears twice in this review.
Where the points came from
Scope — 2.5/5
One documented job: separate speech from everything else. It does not level, does not EQ, does not touch loudness, does not claim to repair clicks or plosives. The score is not a punishment for focus — it is the rubric honestly recording that five of six problem families are out of scope. If your only problem is noise, this number overstates the pain; if you want a finished file, it understates it.
Control — 2.0/5
The lowest control score we issued this term, tied with Magic Dust. There is no strength slider, no conservative mode, no way to tell it "less." When the model decides your narrator's breathiness is noise, that texture is gone, and your only recourse is not using the tool. The two points acknowledge that the one thing it does, it does with genuinely modern separation quality.
Workflow — 4.0/5
Frictionless where it counts: a drag-and-drop page with no learning curve at all, plus an API for anyone gluing it into a pipeline. Processing is quick. It loses the fifth point because file-level convenience is the whole story — there is no preset system, no batch queue in the web UI, no production-line temperament.
Value — 3.5/5
Verified 2026-09-20: Free plan 10k credits/month ≈ 10 minutes; Starter $6/month, 30k ≈ 30 minutes; Creator $22/month at 121k ≈ two hours (intro discounts advertised); Pro $99/month at 600k. Annual billing works out to roughly two months free. As a rescue kit, superb — ten free minutes a month solves most people's actual emergency, and $6 is impulse territory. As a routine tool, do the arithmetic: paid credits work out near $12 per processed hour, triple LALAL.AI's pack rate and a different sport from Auphonic's $2.40. One more caveat worth its half point: free-tier output excludes commercial use.
Transparency — 4.0/5
Plans and the credit-per-minute rate are public and consistent; the deducted point is for the credit abstraction itself. Nobody wakes up wanting 30,000 credits; they want half an hour of clean speech, and the conversion is one step of algebra a pricing page should not require.
Who it is for
- The one important bad recording: this month's voice memo, interview, lecture — free tier, ten minutes, done.
- Developers: the API makes it the easiest isolation step to bolt into a media pipeline.
- Not for finishing work: plan a leveling/loudness pass elsewhere.
- Not for texture-critical voices: ASMR-adjacent, breathy or performance narration can lose character with no dial to save it.
Credit where due
- Fastest zero-to-rescued path on our board
- Genuinely useful free allowance, monthly, not lifetime
- API access from cheap tiers
- Modern separation quality on plain noise and music
Points deducted
- Zero controls — the purest black box we graded
- Isolation only; no leveling, EQ or loudness
- ~$12/processed hour at paid rates
- Credit maths; free tier is non-commercial
To raise the grade
A single strength slider — even two positions — would be worth a full point of control and would lift the total to 66.5. Minute-denominated pricing would earn the transparency point back. Neither seems technically hard; both would signal that the tool trusts its users with decisions.
Ten free minutes is a fair trial
Upload your actual problem file — the noisy one, not a test — and judge the isolation on the voice you care about. If the free result convinces you, Starter is $6.
Marked link: subscribe through it and ElevenLabs may pay ClarityGrade a commission. The C+ above came out of the scoring script before any link existed. Disclosure.
Ranked against the field: best AI voice enhancer 2026 · If the noise is a live-meeting problem, an isolator is the wrong species — see the category boundary.