block #0009 in --:--:--Join the pool
PENDING…
ai ✓ confirmed 6/6 1h ago · 4 min read

HeyGen Voice TTS Tops a Cloned-Voice Arena, With Asterisks

HeyGen Voice TTS debuted at number one on Artificial Analysis' Controlled Voice arena, but it is unranked on the bigger Provider Voices board, and pricing is murky.

HeyGen Voice TTS Tops a Cloned-Voice Arena, With Asterisks
tl;dr
  • HeyGen Voice scored an Elo of about 1,201 to 1,202 on the Controlled Voice arena, ahead of Qwen-Audio-3.1-TTS-Plus and ElevenLabs' Eleven v4 Turbo.
  • It isn't ranked on the larger Provider Voices arena, where Eleven v4 Turbo leads at 1,328, per OrcaRouter.
  • Instant clones use one recording; professional clones need more audio and a paid slot or add-on.
in this block
  1. What actually happened
  2. What the leaderboard actually measures
  3. The asterisks
  4. Cloning: two different requirements
  5. Why it matters
  6. What to do as a reader

HeyGen Voice TTS debuted at number one on Artificial Analysis' Controlled Voice leaderboard, and HeyGen is shouting "world's best" from the rooftops. The AI avatar company launched HeyGen Voice, its in-house text-to-speech model, on October 9. The ranking is real, but it's one specific board, and the fine print on pricing and cloning is worth reading before you switch.

What actually happened

HeyGen's launch release, distributed via PR Newswire on October 9, calls HeyGen Voice the company's new in-house voice model and says it ranks first on Artificial Analysis' leaderboard. CTO Rong Yan said the goal was a voice "that always sounds like you at your best."

The release says HeyGen Voice is free inside the HeyGen platform and API, and it sits in the same stack as HeyGen's Avatar V model, so identity and voice come from one vendor. HeyGen also says it requires the voice owner's explicit consent, with safeguards in its avatar products.

The company also leaned on its scale. HeyGen says it reaches over 40 million users in 196 countries, has seen more than 118 million videos created and recently passed $200 million in annual recurring revenue. It cites its own survey of more than 1,000 small business owners, where 71.6% said they had recorded a business video and then decided not to post it. These are company figures, not independent audits.

What the leaderboard actually measures

Artificial Analysis' Controlled Voice leaderboard compares TTS models using the same eight cloned voices, four US and four UK. That isolates how well an engine handles a voice it didn't choose.

Artificial Analysis' post, mirrored by Vibeleaderboard, put HeyGen Voice at an Elo of 1,201 across 1,468 appearances, ahead of Qwen-Audio-3.1-TTS-Plus at 1,182, Eleven v4 Turbo at 1,166 and Eleven v4 at 1,162. OrcaRouter's later read shows 1,202, with Qwen at 1,186 and Eleven v4 Turbo at 1,167. Votes keep coming in, so the numbers drift; we're quoting both snapshots.

HeyGen Voice also ranked first in Assistants and Customer Service, second in Knowledge Sharing and first in UK English. On pronunciation robustness it scored 83.1%, tenth of 29 models.

The asterisks

OrcaRouter's point is that "one engine won one board." On the far larger Provider Voices arena, Eleven v4 Turbo leads at 1,328 and HeyGen Voice isn't ranked at all. Several ElevenLabs tiers sit close behind on the controlled board too.

Pricing is murky. Artificial Analysis lists HeyGen Voice at $30 per million characters, versus $40 for Eleven v4 Turbo and $80 for Eleven v4. OrcaRouter notes that $30 is Artificial Analysis' conversion, since HeyGen's own pricing page sells credits rather than a published voice rate. And HeyGen's release says the model is free on its platform and API. Those statements describe different things, so check your actual bill.

Speed is another trade-off. Artificial Analysis clocked HeyGen Voice at 40 characters per second of generation time.

Cloning: two different requirements

HeyGen's developer docs describe two modes. An instant voice uses one recording, of which the first 3 minutes are used, and is usually ready within seconds. Instant voices are free to create during the preview.

A professional voice needs one to ten recordings of the same speaker totaling 20 minutes or more, plus one purchased voice clone slot per voice. The press release describes the Professional Voice Clone as a $99-per-month add-on trained on 30 minutes to three hours of speech. The docs and the release give different minimums, so we're reporting both. Professional speech costs 0.6 API credits per generated minute, per the docs.

Why it matters

Voice cloning is getting commoditized fast. Our Open TTS Leaderboard and MAI voice models pieces tracked the same race. HeyGen's angle is bundling: if you already make avatar videos there, HeyGen Voice TTS keeps everything in one place.

What to do as a reader

If you're choosing a voice engine, test HeyGen Voice TTS on your own scripts and compare it with ElevenLabs on both cloned and stock voices. Don't treat one leaderboard as the final word.

Only clone voices you have clear consent to use, and keep the original recordings. Watch whether HeyGen joins the Provider Voices arena and publishes a direct per-character rate. Nothing here is investment advice.

Not financial advice. DYOR, ser.

More in the pool

all ai
gm ser

Get confirmed before the crowd

Daily block at 07:00 UTC. No spam, just the block, ser.