The human benchmark for speech AI.

Woodstitch runs blinded listening studies on text-to-speech models with paid human raters. Every study states its method, its numbers and what it does not show.

Interim research report · updated 5 November 2026

Can listeners distinguish AI narration from human narration?

346 blinded judgments · 63 paid listeners · ElevenLabs v4 vs 10 human narrators · clips of 15, 30 and 60 s

48.0%accuracy
95% CI 42.3–53.8

+5.3 pp60 s vs 15 s
no length effect detected

46.6%when “very confident”
confidence did not track accuracy

Listeners did not classify above chance, longer clips did not detectably help, and stated confidence did not track accuracy. Acoustically the AI differs most on pause rhythm; listeners’ answers did not follow it. Read the report →

For AI teams

Testing a release candidate? We run blinded human evaluations on your models, voices and scripts. Private by default — paid work never changes a public result.

Evaluate your model

For listeners

Get paid to compare short audio clips and pick the one that sounds better. Work whenever you want.

Become a rater

How every study is run — listeners, payment, blinding, statistics, publication rules — is in the platform methodology.