AI BriefingBlind test of 32 ElevenLabs voices finds professional badges and popularity do not predict audiobook quality
·2026.11.12 06:51
Key point
Loudness and age tags significantly influence listener preference, while usage metrics and professional status show no correlation with quality.
Details
- 125 paid US listeners made 1,122 forced-choice judgments in a blind test of 32 ElevenLabs Voice Library voices using identical eleven_v4 settings.
- "Marie - Audiobook Narrator" and "Measured" ranked first with 66% win rates (95% CI 55–76), while "Alder - Warm, Raspy and Comforting" ranked last at 24%.
- Voices with the "professional" badge performed no better than instant clones, winning only 46% of 649 head-to-head pairs against them.
- Loudness was a significant factor, with louder clips winning 54% of pairs; "old"-tagged female voices underperformed at 43% win rate compared to 58% for young female voices.
- Usage volume, ranging from under 1M to nearly 1B characters/year, showed zero correlation with listener preference.
- Full data, including per-judgment CSV and charts, is published on woodstitch.com with a disclosed relationship to the testing platform Woodstitch.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
ElevenLabs Introduces ElevenAI Desktop App for Dictation and Meeting Notes