Like the title says, to reproduce, go here:
For style instructions, type in something like “In English with Indian accent” or “In English with Delhi Accent” or “In Hinglish” - it always fails and generates no audio. But if I just say “In English” it works fine.
Even when I specify exact origin, it fails 50% of the time, which is unusable in any production environment.
Yet, I can see in Google API docs for audio generation that you CAN specify tone/pacing/accent for this model/endpoint. So is this a bug?
Hi
I am working on reproducing this , i wanted to confirm which TTS model are you using ?
Hi
Using the gemini-2.5-flash-preview-tts model.
I was able to run into this error for an entire day by simply asking it to speak English in an Indian accent. You can test it with different origins too.
1 Like
I am having some difficulties reproducing this error .
I was able to get an audio back in AI Studio as well using API end point for English in an Indian accent
Does this always fail for you or is there specific prompts which are failing for you ?
