WoodstitchHuman evals for voice AI

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.

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
Research report · 18 November 2026

Which Fish Audio voice is best for audiobooks?

998 blinded judgments · 72 paid listeners · 32 Fish Audio library voices · one 30-second nonfiction passage · loudness-matched

76%“Old”, first
95% CI 63–85
60%popular voices win
vs little-used ones
41%quality-check passed
vs unchecked: no better

Popularity is a usable signal on Fish, the quality badge is not: usage predicted preference where it did not on ElevenLabs, the "passed" badge told listeners nothing once popularity was accounted for, female voices were slightly ahead, and with level matched nothing won by being loud. Listen to all 32. Read the report →

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