ElevenLabs

What it actually does

ElevenLabs turns typed text into speech, clones a voice from a short sample, dubs existing audio into other languages, and hosts a library of community voices. There is also sound effects generation and an API. On top of the voice work the company now markets speech to text, music, and agent products.

Voice quality is the reason to be here, and it is not a close contest. Across G2, Capterra and Trustpilot commentary the consistent praise is the same pair of things: natural delivery and low latency. Language coverage is broad, though the numbers we have disagree. The product listing says 32 languages while reviewer commentary cites 70 or more, so check the specific language you need rather than trusting a headline count. The APIs carry SOC 2 Type 2 certification, and the company ships watermarking plus a speech classifier for detecting its own output, which helps if you are the person filling in a compliance questionnaire.

What it costs

Free gives you 10,000 characters a month across three voices. That is enough to audition the voices honestly and nowhere near enough to publish on a schedule. Paid starts at $5 a month for Starter, $22 for Creator, and $99 for Pro.

The gap between $22 and $99 is where buyers get caught. The pro-tier features are expensive relative to the rest of the market, and if you need them you are paying five times the Creator price rather than a small step up. The other cost is less obvious. You are buying a voice tool from a company that has spread into image, video and music generation, so some of your subscription is funding a roadmap that has nothing to do with the thing you came for. If voice stops being the priority, you find out after you have built a workflow on it.

Where it falls down

The review scores split hard by audience. G2 sits at 4.5 out of 5 from 1,211 reviews, with 71% five-star and 0% one-star. Trustpilot sits at 3.0 from 1,147 reviews. We could only see that Trustpilot figure in a search listing because the site blocks extraction, so treat it as indicative rather than page-verified. Still, a 1.5-point spread between business reviewers and consumer reviewers means the consumer-side experience is worse than the B2B scores suggest, and G2 will not show you that.

The safety record is the real problem. ElevenLabs tools were used in the AI robocall impersonating President Biden. The company added a "no-go voices" policy afterward, and 404 Media then bypassed it. Emanuel Maiberg reported on 7 March 2024 that 404 Media "discovered a bypass to ElevenLabs' system that allows a user to generate cloned voices that under ElevenLabs' own policy should be not possible." Press scrutiny over cloned celebrity and political voices has continued in The Atlantic, the Washington Post, PCMag and ZDNet.

On output, users report occasional artifacts in longer samples, so long-form work needs a listening pass rather than a straight export. And the ethics question around cloning is not settled by a consent checkbox, whatever the marketing says.

Who should buy it

Podcasters, audiobook producers and video creators who need voice that does not sound synthetic. If your output is narration and quality is the binding constraint, this is the tool, and the free tier lets you prove that before paying anything.

Do not buy it if you want a vendor whose cloning reputation will survive a procurement review, because the robocall episode and the bypassed policy are on the public record and someone will find them. Do not buy it to listen to your own documents either. That is a different product category, and Speechify is built for it. And if you only need a handful of clips a month, stay on free or Starter rather than reaching for Creator out of optimism.

Our 4.0 reflects a superb product carrying ugly externalities. The score would be higher if the safety controls held up when journalists pushed on them.