I’ve implemented [Pronunciation Dictionaries](https://docs.vapi.ai/assistants/pronunciation-dictionaries) in my Voice AI integration using ElevenLabs voices, and they are working properly during conversation. However, I’ve noticed that if a word with a custom pronunciation rule appears in the assistant’s

firstMessage

property, the pronunciation dictionary is not applied. This behavior seems to indicate that Vapi may not be applying pronunciation dictionary transformations at the time the

firstMessage

is rendered. Steps to Reproduce: 1. Create a pronunciation dictionary in ElevenLabs (via Vapi API) with a rule for a specific word. 2. Associate the dictionary with the assistant’s voice configuration in Vapi. 3. Set the assistant’s firstMessage property to include the target word. 4. Start a call or conversation. Expected Result: The pronunciation dictionary rule should be applied to words inside the

firstMessage

. Actual Result: The rule is ignored, and the word is pronounced incorrectly. Can you confirm whether pronunciation dictionaries are currently supported for the

firstMessage

property? Some example calls: Organization ID: e4f4156e-e34d-4d6d-92dc-65ad983fde37 - 23e3e6c3-a4e0-4389-be23-be4f47a9558c - 1c68c5f0-0f98-46b5-9ad7-06c8591b7fca - d13916df-90a6-47ec-8e05-852d2a82aa1b - dad668a0-27be-4b7a-a278-22c92def486c - 6edb9437-65d4-41e7-9796-14887bd3d5fd - 2814245b-09e7-4fad-88ae-2c40962ad611

Based on the documentation I have found, there is no explicit confirmation that pronunciation dictionaries are applied to the

firstMessage

property. The documentation only states that custom pronunciations are used "when your assistant encounters the specified words during conversation," without mentioning

firstMessage

specifically. Suggestion: To clarify this behavior and get an authoritative answer, I recommend reaching out to [Vapi Support](https://docs.vapi.ai/support) or submitting a feature request, as this may be an undocumented limitation. Sources: - [Pronunciation dictionaries](https://docs.vapi.ai/assistants/pronunciation-dictionaries) - [Vapi Support](https://docs.vapi.ai/support)

It seems the pronunciation dictionary doesn’t apply to firstMessage since that part is pre-rendered before the rules kick in. A simple workaround is to trigger the same text as the first system message right after the call starts. I can help you set that up if needed. Have you noticed if the issue also happens when sending assistant messages mid-call through the API? @mk

I didn't try sending messages through the API, will try the workaround you suggested, and let you know here. Thanks for clarifying!

Perfect! Try the workaround and let me know how it goes. If it still doesn’t work, I can suggest another way to apply the pronunciation rules consistently. @mk