### Answer Based on the available sources, Retell AI's real-time ASR models are trained on healthcare vocabularies and deliver word-error rates under 7% in clinical settings. [(1) ]() However, the sources do not explicitly state which specific transcription model Retell uses under the hood. There is a mention in a Discord thread of Deepgram being referenced in the context of transcription issues, but this is not confirmed as the official ASR provider in the documentation. [(2) ]() What the documentation does cover: - Retell offers two transcription modes: "optimize for speed" (uses latest interim results with low endpointing) and "optimize for accuracy" (waits longer with more context, adding ~200ms latency) . [(3) ]() - Both modes have similar Word Error Rate (WER) , with the main difference being accuracy in capturing entities like numbers and dates—if your use case relies heavily on these, "optimize for accuracy" is recommended. - You can also use Boosted Keywords to add up to 100 custom keywords to improve recognition of domain-specific terms. [(4) ]() For the specific transcription model name/provider, I'd recommend reaching out via: - [Contact Sales]() - [Retell AI GitHub]() Would you like a general answer about common ASR providers used in voice AI platforms?

​ To learn more, see the sources I considered: [[1] AI Voice Agents for Prescription Refills in Healthcare | Retell AI | Retell AI](https://www.retellai.com/blog/automating-prescription-refills) [[2] Transcription issues leading to inactivity](https://discord.com/channels/1186424376300679209/1315673935333036122) [[3] Balance between transcription accuracy and latency](https://docs.retellai.com/build/transcription-mode) [[4] Increase transcription accuracy](https://docs.retellai.com/reliability/wrong-transcript) ​

Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.