can I pre-generate audio or play a recording
# ❓・ai-evy-supportm
meadow
05/26/2026, 1:45 AM@AI Evy can I pre-generate audio or play a recording?
a
AI Evy
05/26/2026, 1:45 AMThanks for tagging me @meadow. I'll look through relevant sources and get back to you shortly.
AI Evy
05/26/2026, 1:45 AM### Answer
Based on the available sources, Retell AI currently does not support playing pre-recorded audio files or injecting pre-generated audio directly into conversations. [(1) ]() [(2) ]()
However, there are a couple of workarounds you could consider: [(2) ]()
1. Use Twilio IVR to play the audio before transferring to the Retell AI agent
2. Add the content as text to be spoken by the agent, and optionally use voice cloning to match a specific person's voice (like a CEO message)
3. Use a multi-agent transfer approach where agent A speaks the message, then transfers to agent B for the main conversation
If you need to inject audio from one conversation into another (like repurposing information from one individual to another) , this feature is not currently available in Retell AI. [(1) ]()
AI Evy
05/26/2026, 1:45 AM
To learn more, see the sources I considered:
[[1] inject pre-recorded audio into a conversation](https://discord.com/channels/1186424376300679209/1301663894946975864)
[[2] Playing audio file](https://discord.com/channels/1186424376300679209/1376969263780200498)
AI Evy
05/26/2026, 1:45 AMFeel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.
AI Evy
05/26/2026, 1:50 AMThanks for the follow-up, @meadow. I'll get back to you soon.
m
meadow
05/26/2026, 1:50 AM@AI Evy lowest latency model
a
AI Evy
05/26/2026, 1:50 AM### Answer
Retell LLM is designed for low latency voice interactions with an average end-to-end latency of approximately 800ms [(1) ]() . This proprietary model has been fine-tuned specifically for conciseness, conversational tone, and reliable function calls, making it ideal for voice-based interactions .
The Retell LLM achieves sub-600ms to 800ms latency performance, which is significantly faster than traditional IVR systems (200-400ms but with limited understanding) and competitor AI systems (3-4 seconds) [(2) ]() [(3) ]() . This low latency ensures natural conversation flow and maintains user engagement [(3) ]() .
You can use Retell LLM when creating agents through the dashboard or programmatically via the API [(1) ]() . The model supports both single-prompt agents for straightforward tasks and stateful multi-prompt agents for more complex scenarios .
AI Evy
05/26/2026, 1:50 AM
To learn more, see the sources I considered:
[[1] Introducing Retell LLM Beta](https://www.retellai.com/changelog/introducing-retell-llm-beta)
[[2] ](https://www.retellai.com/resources/retell-ai-receptionist-understands-calls-better)
[[3] Why Low Latency Matters for AI Agents | Retell AI](https://www.retellai.com/blog/why-low-latency-matters-how-retell-ai-outpaces-traditional-players)
AI Evy
05/26/2026, 1:50 AMFeel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.
m
meadow
05/26/2026, 1:53 AMthat's not quite right, fastest model please