HeyGen, the AI avatar company best known for cloning faces, has turned to voices. On Friday it launched HeyGen Voice, an in-house voice model that debuted at No. 1 on Artificial Analysis' Controlled Voice leaderboard with an Elo score of 1,201 in independent evaluations of AI voice models.
The model is built for what the company calls authentic delivery: emphasis, emotion, pacing and character, so a cloned voice does not come across as a flat synthetic copy. CTO Rong Yan framed quality as the entire product. "Voice is one of the most powerful and personal ways we communicate, and quality makes all the difference," he said, adding that listeners lose interest the moment a voice lacks those cues.
Pricing is the aggressive part. HeyGen Voice is free inside the HeyGen platform and API. A Professional Voice Clone add-on costs $99 per month and is trained, with the owner's explicit consent, on 30 minutes to three hours of their speech. API pricing is temporarily discounted 50% to $15 per million characters through October 31.
Strategically, the launch closes the loop on HeyGen's stack. Avatar V remains the core model, and voice becomes a new format inside the same stack, so a person's face and voice come from one place rather than a patchwork of vendors. The company has made consent and control central to its pitch as realistic digital representations of people get easier to generate.
The launch follows HeyGen Video and lands on strong commercial numbers. Founded in 2020, HeyGen says it now reaches more than 40 million users across 196 countries, has powered more than 118 million videos, and is used by 85% of the Fortune 100. The company recently passed $200 million in annual recurring revenue after doubling revenue in eight months.
A company survey of more than 1,000 small business owners in its user base found that 71.6% had recorded a video for their business and then decided not to post it, usually over production quality or a lack of confidence in front of the camera — exactly the gap a convincing voice model is meant to close.
The independent leaderboard top spot is a marketing asset, but holding it is the harder part. The voice layer of the AI stack has become one of the most contested corners of generative AI, with ElevenLabs, Cartesia and the frontier labs' own TTS models trading positions on the same boards all year. HeyGen's bet is that owning the full identity stack — face, voice and consent controls in one place — matters more to business customers than any single leaderboard rank.
Comments (0)
Log in to join the discussion
Log InNo comments yet