Resemble AI

Resemble has spent a couple of years turning from a voice-cloning vendor into something closer to a voice security company, and that reframing is the most important thing on its site today. The generation side is still substantial: instant and custom cloned voices, text to speech with emotion and prosody control, and a streaming API for conversational agents and games, where latency is the product. Around it sits verification and provenance — identity registration so a voice is enrolled against a consented owner, watermarking of generated audio, and real-time deepfake detection APIs covering audio as well as video and images. That stack is sold to platforms, banks and enterprises that must answer whether a recording really came from the person it claims to be from.

The blunt view: this is a purchase for a compliance department as much as an audio team, and that is the point. Where ElevenLabs wins on voice breadth and self-serve pricing, Resemble wins when you must prove consent and detect abuse. Before contracting, check how consent is captured and whether it is auditable per enrolled voice, whether detection and watermarking are included or priced separately, whether your languages work in the real-time path rather than only in batch, and how metering behaves for agent traffic — per-character or per-minute pricing scales badly the longer conversations run. On-premises deployment is offered, which usually means the price is a conversation rather than a checkout button.