During voice responses, the agent’s speech is intermittently unintelligible. Symptoms include slurred or mumbled delivery, individual words truncated mid-syllable, and occasional audio that isn’t recognizable as words at all.
agent: agent_8aede261f638a4acb05ecd9cb6
call id: call_d9f787d32de692ba37a2eacaa7f
voice choppiness occurred between 5:31 - 6:00
Hello @vincentstanfordlead Thanks for sharing the Call ID. I’ll check it with the team and get back to you as soon as I have an update.
Thank You
Hello @vincentstanfordlead We reviewed call_d9f787d32de692ba37a2eacaa7f and checked the 5:31 to 6:00 window across the audio path, voice engine health, and latency. All of those were clean for the entire call, with no provider fallback, TTS p90 around 180 ms, and turn-taking p90 around 21 ms.
What you heard is a text formatting artifact rather than an audio quality issue. After the caller asked the agent to spell the email address more slowly, the agent produced the address with an ellipsis character between every single letter (i… n… t… a… k… e… @… a… b…). The voice engine treats the ellipsis as a pause marker, so it renders one letter, then a pause, then the next letter. On a phone call that sounds exactly like the mumbled, slurred, and truncated delivery you described. Earlier in the same call the agent spelled the same address using hyphens between the letters and it played back normally, on the same voice, the same agent, and the same call leg.
To control pacing, use the Voice Speed slider in the agent’s voice settings, which ranges from 0.5x to 2.0x, or enable “Dynamically adjust based on user input” so the agent matches the caller’s pace during the call. Pacing can also be instructed directly in the prompt, for example “slow down when spelling out addresses”. Speed handled through the voice settings will not produce this artifact, while pauses expressed as punctuation in the text will.
Voice settings reference: Configure a Retell agent's model, voice, and welcome message
Thank You