Turing TTS Benchmark: Best Voice Model Fooled Listeners Only ~10% of the Time
rohanpaul_ai · x · 2026-11-01
Voice Arena launched Turing TTS, a benchmark separating preference from human-likeness: listeners hear the same sentence from a person and a model and pick the human; the model's fool rate is the score.
- Best model, Gemini 3.8 Flash TTS, fooled listeners 10.5% of the time; all ranked models remain "detectable"
- Top four models (Gemini 3.8 Flash TTS/Flash-Lite, Cartesia Sonic-3.6, ElevenLabs v4) are statistically tied, with fool rates from 10.5% down to 6%
- Against real humans, the whole field still has a long way to go, exposing a gap between current evals and the indistinguishability goal
Related event: Turing TTS Benchmark: Best Models Fool Humans Only 10% of the Time(3 posts)→
More from Models
- Newest LLM isn't automatically best for business: reliability and cost beat benchmarks — DavidLinthicum · 2026-11-01
- Using AI to crack a spatial folding puzzle on a Punjabi TV quiz show — satnam6502 · 2026-11-01
- Friends resubscribed to ChatGPT for 'unlimited' use — worst timing ever — Angaisb_ · 2026-11-01
- GPT-6.1 Terra ranks #14 on Toloka Arena as pass^5 falls to 59% from ~72% — MParakhin · 2026-11-01
- OpenAI employee: GPT-6.1 Sol feels better than GPT-6 Astra, nearly as smart and much faster — willdepue · 2026-11-01
- CMU paper shows pretraining, SFT and RL can run as one unified likelihood-based algorithm — omarsar0 · 2026-11-01