Resemble AI

Read this first: Resemble AI stopped selling voice generation to new customers

If you came here for voice cloning and text-to-speech, the honest answer is that you are at the wrong product page. On 2026-08-13 Resemble AI published a post explaining that it has put the whole company behind deepfake detection, that it is supporting existing voice customers but not taking new ones, and that its identity going forward is a detection company (getcartsignal.com/tools/resemble-ai; bestaivoicegenerator.com carries a matching status correction dated 2026-09-02). The text-to-speech product page still exists — Chatterbox, real-time speech synthesis across 100 languages with custom pronunciation and PerTh watermarking — but it serves the installed base, not the market. For voice generation as a new buyer in October 2026, look at ElevenLabs or the speech APIs from the hyperscalers instead.

What Resemble sells now is one of the more complete deepfake-detection stacks available, and that is what I have reviewed below, because it is what you can actually buy. The pivot is also, in fairness, a coherent bet: as synthetic media spread and the EU AI Act's Article 50 watermarking requirements took effect on 2026-08-01, a company with both detection models and provenance tooling is positioned where the demand is heading.

What it actually does

The product line reads like a security suite. Resemble Detect analyses audio, images and video for synthetic tampering, returning a verdict with an explanation rather than a bare score — a single API call, with the vendor claiming sub-300ms responses. The underlying models are its own work: DETECT-3B Omni, a three-billion-parameter multimodal model benchmarked across 50-plus languages and 160-plus generative models, and DETECT-World, a newer "world model" that also checks physical plausibility — lighting, motion, continuity — so it can flag output from generators it was never trained on. The benchmark page claims top ranks across audio, image and video detection and 97.4% average audio accuracy; the product pages claim up to 99.5% detection accuracy tested against 250-plus generators. Those are vendor numbers and I treat them as such — I could not independently reproduce them.

Resemble Intelligence layers human-readable forensic reports on top: which artifacts drove the score, with a Reverse Search aimed at compliance teams and legal review. Resemble Meetings brings real-time detection into Zoom, Teams, Meet and Webex with alerts and a forensic audit trail, and it is the reason Microsoft Teams appears as a featured partner on the homepage. Resemble Identity does biometric speaker verification with enrollment from four seconds of audio. Resemble Watermarker embeds an imperceptible PerTh watermark across audio, video, image and text, positioned against EU AI Act Article 50, and an Inspector SDK checks for it. There is also Resemble Agent Detection at $1 per 1,000 sessions — a product aimed at spotting AI agents in your traffic.

My testing notes

First observation: the detection workflow is built for evidence, not vibes. A single call on an audio file returns both a verdict and the reasoning behind it, and the Intelligence layer turns that into a report a compliance officer can read. In my testing of the flow with audio I knew the provenance of, the confident cases were flagged clearly and the explanations referenced recognizable artifacts rather than generic warnings. That matters more here than raw accuracy: in fraud and trust-and-safety workflows, a number nobody can defend in a meeting is close to useless.

Second, the pricing mechanics reward buying the subscription only at real volume. On the Flex plan at time of writing, audio detection is $0.035 per second and video is $0.07 per second — an hour of audio runs about $126, which is fine for spot checks and painful for archives. The Team plan drops detection to $0.015 per second across the board, so the $350/month monthly price (or $280/month billed annually) pays for itself somewhere around ten hours of audio detection a month. My practical take: Flex for investigations and pilots, Team the moment detection becomes a workflow.

Third, the deployment story is deeper than most competitors in this category. On-premises via Docker and Kubernetes, air-gapped installations that make no external calls, Zero Retention Mode that deletes media after analysis, and GDPR and HIPAA compatibility claims on the products page. For a security team at a bank or a broadcaster, those deployment options are the difference between buying and not.

Two smaller products are worth understanding before you compare plans. Resemble Meetings is the one that lands in daily operations: real-time detection during live calls across Zoom, Teams, Meet and Webex, with alerts and a forensic audit trail afterwards — the use case being the voice-clone-CEO-call style of fraud that security teams now plan for explicitly. Resemble Agent Detection goes the other direction, scanning sessions for AI agents at $1 per 1,000 sessions on every tier; it is cheap enough to leave on. Neither replaces the API, but both show where the roadmap points: detection embedded in the tools people already use rather than a destination product you remember to visit.

The documentation deserves credit for treating detection as an engineering problem. The Detect API reference describes submitting media and retrieving structured metrics and visualizations, including an option to check for Resemble's own watermark in a single-file request. Identity enrollment from four seconds of audio and cross-app authentication reads like a product built for voice-based verification flows rather than a demo. Across the API surface I found the same shape everywhere: one call in, structured verdict out, with enough metadata to build a queue around.

Where it falls short

Start with the fact in the first section: no voice generation for new customers, which removes the half of the product this catalogue listing historically described. Then the detection economics — per-second pricing on video at $0.07 Flex makes continuous monitoring expensive without a subscription. The vendor's accuracy claims are self-published; the benchmarks page shows its own model ranked first, and independent comparative studies are scarce, so treat the percentages as claims to test against your own data before committing. Setup is API-first and technical: this is not a product a marketing team switches on before lunch, and the Team/Business seat structure means a solo investigator on Flex has no middle option between pay-as-you-go and a $350/month team plan. Finally, the churn itself is a signal — a company that just replaced its commercial identity will keep moving, and contract terms are worth reading closely.

A note on what "supporting existing customers but not taking new ones" means in practice: current voice customers can keep building on Chatterbox and the voice APIs, and the docs still describe real-time synthesis with custom pronunciation. If you are one of them, the migration question to ask your account contact is how long that support runs and what happens to voice workloads at renewal.

Verdict

I rate Resemble AI 3.3. The detection suite is deep and unusually deployment-friendly, and the Intelligence reports solve a real problem for compliance workflows; those earn strong core-functionality marks. Value is mixed — excellent at volume on Team, expensive at retail rates on Flex — and usability and ecosystem are held back by the technical, API-first surface and the missing mid-tier. The score is also discounted because the product this listing originally described is effectively closed: voice cloning and TTS remain for existing customers only. As a deepfake-detection and provenance platform, Resemble is a serious contender worth a pilot. As a voice-generation vendor, it is a redirect.

Sources

• https://www.resemble.ai/pricing
• https://www.resemble.ai/products
• https://www.resemble.ai/products/detect
• https://docs.resemble.ai/detect
• https://resemble.ai/deepfake-detection-software
• https://resemble.ai/benchmarks
• https://resemble.ai/products/text-to-speech
• https://getcartsignal.com/tools/resemble-ai
• https://bestaivoicegenerator.com/reviews/resemble-ai
• https://realaitest.com/reviews/elevenlabs-vs-resemble.html
• https://www.usagepricing.com/blueprint/activity/resemble-ai-2026-07-29-team-business-tiers
• https://www.g2.com/products/resemble-ai/pricing