Original Discord Post by sebssoriano | 2024-07-15 14:32:41
Hello,
Recently, Catalan language has been added to our account. We’ve been testing with it and here are our findings:
- When we only use catalan in the Language and Speech functionality, it recognizes our speech quite well.
- However, the more languages we add, the more it struggles to understand us, especially when English is added.
For these tests, we used ‘JennyMultilingualV2 Female American English Voice’ from Azure. This voice has the same accent inconsistencies we experienced with ElevenLabs during the last months (i.e: it ends up speaking Spanish with British accent). Looks like this happens especially when the answer is very long. OpenAI voices are available for Catalan, but they also exhibit the same inconsistencies.
Given this context, and to minimize issues, is it possible to…
- Implement a workaround to restrict the STT and TTS components to recognize and understand only Catalan?
- Since accent inconsistencies are more common in long replies, is there a way beyond prompting on the character description to ensure replies are more concise?
Thanks in advance for your help!