Cartesia Voice -> Language Setting Not Working
# support
y
Hey team, running into an issue with Cartesia voice language and hoping someone can help verify this on your end. I'm creating a call via the API and passing assistant.voice.cartesia.language = "hi" (Hindi). But it seems like that language config isn't being applied — the voice is still pronouncing names with an English accent and mangling them. For example, if I pass the text "Hi Husen Bhai, How are you?" — it comes out something like "Hi, Huzenberg AI. How are you?" — clearly treating it as English. I tested this in the Cartesia playground directly: When the transcript language is set to English or Auto-detect → same mispronunciation. When I manually set it to Hindi → it pronounces perfectly. So the language setting in the Cartesia playground does fix the issue — but it doesn't seem to be getting passed correctly through the Vapi API. Could someone verify whether the language field is actually being sent to Cartesia perfectly on your side and working as expected ? API reference I'm using: https://docs.vapi.ai/api-reference/calls/create → request.body.assistant.voice.CartesiaVoice.language Org ID:
5e6b2415-b4b3-4b00-8a8e-5ceb82d10f6d
Call ID:
019cfaed-28ba-7001-8778-b61672f9f7db
.
v
According to the [October 30, 2024 changelog](https://docs.vapi.ai/changelog/2024/10/30), you can now set
CartesiaVoice.language
to
"hi"
for Hindi. To ensure the API is receiving the correct value, try sending your request with the following structure:
Copy code
json
"assistant": {
  "voice": {
    "cartesia": {
      "language": "hi"
    }
  }
}
If the issue persists, double-check that the payload matches this format and consider reaching out to support with your call details for further investigation. Source: - [October 30, 2024 Changelog](https://docs.vapi.ai/changelog/2024/10/30)
c
Hi Yash, we’ll look into the call IDs and get back to you soon.
Hi, I checked your call
019cfaed-28ba-7001-8778-b61672f9f7db
and the
language: "hi"
parameter is being passed to Cartesia correctly on our end. The config looks fine. The issue is likely the voice ID you're using (
898606ed...
) - it's not a Hindi-native voice, so even with the language set to Hindi, Cartesia defaults to English pronunciation. Fix: Switch to one of these Hindi-native voice IDs: •
ac7ee4fa-25db-420d-bfff-f590d740aeb2
7f423809-0011-4658-ba48-a411f5e516ba
c1abd502-9231-4558-a054-10ac950c356d
bdab08ad-4137-4548-b9db-6142854c7525
95d51f79-c397-46f9-b49a-23763d3eaa2d
One more thing - your first message is in English (
"Hi {{customerFirstName}}, How are you?"
), which may cause Cartesia to treat that utterance as English regardless. Using a Hindi-native voice should help with Hinglish content like this. Let me know if swapping the voice ID fixes it! Best, Vapi Support
y
Hi @Shaunak, Thanks for the quick reply — really appreciated! We tested the Hindi-native voice IDs you shared and they're working perfectly. However our cloned cartesia voice seems to have an issue — it's not pronouncing correctly even with the language set to Hindi, whereas the other Hindi voices you provided are working just fine. Thanks again for your support!
c
Hi Yash, Glad the Hindi-native voices are working! For the cloned voice issue, the most likely cause is the model. Can you check which Cartesia model you have set for the cloned voice in your assistant config under
voice.model
? Make sure it's
sonic-3
or
sonic-multilingual
- not
sonic-english
. The Hindi-native voices that worked for you are all using
sonic-3
, and
sonic-english
may ignore the language parameter even when set to Hindi. If you're already on
sonic-3
and it's still not working, share a call ID with the cloned voice and I'll check the exact request we're sending to Cartesia. Best, Vapi Support