Can anyone provide any guidance or tips on how to get the agent to spell back names without grossly mispronouncing certain letters? This issue makes trying to confirm spellings impossible.
Text I want spelled back is being generated like this: “g - - m - - a - - i - - l - - dot - - c - - o - - m”
The hyphens slow the pace, but certain letters are completely mispronounced.
Based on the available sources, here are some strategies to improve letter pronunciation when spelling back names:
Retell AI offers a custom pronunciation feature that works with English agents using 11Labs voices with the eleven turbo v2 (English only) voice model(1). You can set a pronunciation dictionary for the agent that includes the phonetic alphabet (IPA or CMU) and the phonetic pronunciation of specific words or letters(1).
For example, you can specify:
actually)ipa or cmu)æktʃuəli)(1)You can find phonetic pronunciations using tools like the IPA pronouncing dictionary tool or the CMU pronouncing dictionary tool(1).
According to community feedback, switching to a better model like GPT-4 and using the pronunciation function in the speech settings can help train the model for correct pronunciation(2). The pronunciation function works well for words, though individual letters may require more specific configuration(2).
Retell AI also offers a “normalize for speech” option that normalizes text (numbers, currency, dates, etc.) to its spoken form for more consistent speech synthesis(3).
Note: The custom pronunciation feature is currently available for English agents with specific 11Labs voice models(1). If you need help implementing this for your specific use case, you may want to reach out to Retell AI support through their Contact Sales, Discord, or GitHub channels.