Hi team,
We are currently using the ElevenLabs V2 voice engine and the custom pronunciation dictionary. We would like to use the V2.5 engine as we feel the voice sounds better and is more consistent than the V2 version, however according to the docs and confirmed with our testing, the custom pronunciations are not obeyed when using V2.5.
This feature only works with English agents using 11Labs voices that selected the
eleven turbo v2(English only) voice model.
We currently have a bit of a clunky workaround in our system prompt as follows, however we would prefer to use the custom dictionary as that is the whole purpose of the feature. We do this because some of our MCP results return words the TTS struggles with, otherwise we would just put the weird phonetic spellings directly in the prompt.
# Word substitution
If you encounter the following words, instead say their replacement word and not the original.
- “example” should be replaced with “an example replacement”
- “test” should be replaced with “this is a test”
Is this a limitation on the Retell side or the ElevenLabs side, and is it on the roadmap to bring the custom pronunciation to the V2.5 engine?
