Freya's Adam TTS tops Design Arena blind listening bench, 43 Elo points from real human speech
socialwithaayan · x · 2026-09-22
Voice startup Freya released Adam and Eve, TTS voices it calls the most human-like ever built. Adam ranks #1 on Design Arena's Audio Realism Bench with an Elo of 1419 vs 1462 for real human speech in blind listening tests, well ahead of Cartesia Sonic 3.6 (1306), Eleven v3 Conversational (1202), Grok TTS (1150), Google Gemini 2.5 Pro TTS (1101) and OpenAI GPT Realtime 2 (1071). The team claims the uncanny valley has been crossed. API access opens in batches; they also open-sourced FreyaTTS-small, a from-scratch Turkish TTS model (48kHz, Apache 2.0) with paper, code and eval dataset.
Related event: Freya Releases Adam/Eve TTS Models, Claims to Cross the Uncanny Valley(2 posts)→
More from Multimodal
- Blue noise sampling yields stable, coherent generations without lowering temperature — amplifiedamp · 2026-09-23
- Fully AI-generated short film 'The Race for AGI' made with Seedance 2.5 — HeyToha · 2026-09-23
- Author shares the full prompt for AI-assisted, beat-synced music video editing — piotrbinkowski · 2026-09-23
- fal releases State of Generative Media Report Vol. 2 as gen media hits production — adamho · 2026-09-23
- GPT-6 Astra plans, Higgsfield renders, Jev picks: an AI motion graphics pipeline — PrajwalTomar_ · 2026-09-23
- Redditor Speculates Qwen Image 2.1 Was Distilled from GPT Image 2 — Friendly-Fig-6015 · 2026-09-23