Deepgram Aura beats OpenAI TTS in Autor test of 12,000 live calls
Tool · dev.to · stat: 180ms Autor Tech split-tests four text-to-speech engines across 12,247 live patient calls for its healthcare receptionist assistant, Loquent. Deepgram Aura secures an 81.2%…
Tool · dev.to · stat: 180ms
Autor Tech split-tests four text-to-speech engines across 12,247 live patient calls for its healthcare receptionist assistant, Loquent. Deepgram Aura secures an 81.2% completion rate, beating ElevenLabs and OpenAI. The team attributes the win to Deepgram's low 180ms latency, proving that speed matters more to patient trust than isolated voice quality.
Low latency beats high fidelity when building real-time voice agents Founders building voice agents should prioritize response speed over hyper-realistic vocal tones to prevent user hang-ups.
Autor Tech blog post
Per Autor Tech's production split-test of 12,247 live patient calls.
Lower latency engines drove higher completion rates, while the most natural-sounding voice had the highest abandonment.
Every claim ties to a primary source. See our methodology.