HeyGen launches Voice cloning model, claims blind-test wins over ElevenLabs
On October 15, HeyGen released its voice cloning model HeyGen Voice and published multiple rounds of blind test data, claiming it outperforms mainstream competitors including ElevenLabs and Cartesia on several metrics, drawing widespread attention.
Confirmed
- HeyGen officially launched Voice, a video-scene-aware voice cloning model, saying it ranks first on the Artificial Analysis leaderboard and is the first voice model purpose-built for video: feed in everyday recordings, a script, and target footage, and it adjusts tone, pacing, and acoustics to match the scene.
- Environment-matching blind test: the same voice was placed into 50 settings including a cathedral, a factory floor, and the front seat of a car; judges, unaware of which was which, assessed whether the sound matched the visuals—results showed HeyGen Voice's environment match rate exceeded competitors like Cartesia.
- Real-user blind choice: thousands of users recorded their own voices, then listened unlabeled to versions from HeyGen, ElevenLabs v4, and Cartesia Sonic 3.6 and picked a favorite; HeyGen won 62.2% against ElevenLabs and 54.1% against Cartesia.
- Largest-scale preference test: 42,875 blind evaluations across five languages, asking "which voice would you want narrating an entire class." HeyGen's win rate was 77.9% over Cartesia, 68.1% over Inworld, and over 61% against ElevenLabs v4, with the margin widening to 87% in cross-language scenarios.
Why it matters
- The voice cloning space has been dominated by players like ElevenLabs; HeyGen's differentiation is "video-native, scene-aware," backed by a dense release of third-party user blind test data, which could reshape competition in video voiceover and localization markets.
- All data was released by HeyGen itself; test setups and statistics come from a single party, and independent third-party verification remains to be seen.
2026-10-15 ~ 2026-10-15 · 6 related posts
Primary sources
- HeyGen voice blind test with 42,875 ratings beats Cartesia 77.9% of the time — HeyGen · 2026-10-15
- [source] HeyGen Launches Scene-Aware Voice Cloning, Beats ElevenLabs in 42,875 Blind Ratings — HeyGen · 2026-10-15
- HeyGen Voice beats ElevenLabs and Cartesia in blind ambient-fit tests — HeyGen · 2026-10-15
- [source] HeyGen users blind-picked its voice over ElevenLabs 62.2% of the time — HeyGen · 2026-10-15
- [source] 42,875 blind ratings: HeyGen voice wins up to 87% in cross-language tests — HeyGen · 2026-10-15
- HeyGen launches video-first voice model ranked #1 on Artificial Analysis — Tegadesigns · 2026-10-15