Vapi voice not reading received text from model
# supporta
Alex QS
05/19/2025, 6:13 PMHi Vapi team, this morning we've seen a few calls using the Elliot Vapi Voice not read out text that is recieved by vapi. Ie, I can see the text recieved in the logs but it is never verbalized. This head led to failed calls due to silence timeout (user waiting for assistant to respond but never hears a response).
Call ids:
75ed7849-dabb-41be-a00e-6c9baa099117
613d44dc-9dda-4f87-b220-e8657a3d14c7
9a6607fd-9b2c-4a95-aab5-8976bd60d624
v
Vapi Support Bot
05/19/2025, 6:14 PMThanks for sharing all those details! I couldn't find enough info in our docs or community channels to solve this one, so I'm escalating your issue to a human teammate who can dig in further. They'll be able to review the call IDs and logs you provided. If you haven't already, you can check that the voice configuration for Elliot is set up correctly and that no fallback or provider issues are present, but otherwise, someone from the team will follow up with you soon.
For reference, you can find more about Vapi Voices [here](https://docs.vapi.ai/providers/voice/vapi-voices), but since your setup seems correct and the issue is specific, a human will take it from here!
a
Alex QS
05/19/2025, 8:51 PM@User
we're also seeing this with a cartesia voice now. We see the model response in the logs at around 20:25:47:114 - however there's just silence on the call until the user says "hello?" nearly 7 seconds later.
https://dashboard.vapi.ai/calls/c413c79c-de3d-477f-964d-e3dd3fbb501c -> right at the beginning of the call
_____
20:25:47:114 - Voice input: Looking to book a room or got questions about the event?
20:25:47:115 - Voice input:
20:25:47:115 - [CHECKPOINT] Model sent end token
20:25:47:115 - [LOG] Model request cost (attempt #1, $0, 5370 prompt, 36 completion)
20:25:53:877 - [LOG] Transcriber output: Hello?
Alex QS
05/19/2025, 9:01 PMThis one too 5486fac7-6410-4e73-9cce-9ab07df89b00 - from 20:57:33:65 to 20:57:48:023 - Vapi received model output but it wasn't read aloud.
Alex QS
05/20/2025, 12:08 AMAnother one - this Assistant is not using customLLM so I think this bug is unreleated: 6a07120b-d232-46ce-844c-4b1e32551c1f
The below model output is received but voice never reads it and silence timeout several seconds later:
20:50:01:150 - [CHECKPOINT] - Model sent end token
20:50:01:151 - [LOG] - Voice input: Thank you, Ms. Billones! Have you stayed with us before?
20:50:01:151 - [CHECKPOINT] - Voice audio requested
Alex QS
05/20/2025, 12:08 AMcc @jonathan lei
s
Shubham Bajaj
05/20/2025, 12:54 AMlooking into it.
Shubham Bajaj
05/20/2025, 1:03 AM🔵 15:00:55:857 Message Received (12-0150deb). Message:
{
"type": "error",
"context_id": "12-0150deb",
"status_code": 500,
"done": true,
"error": "error writing job replies: CUDA error: unknown error\nCUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.\nFor debugging consider passing CUDA_LAUNCH_BLOCKING=1\nCompile with
TORCH_USE_CUDA_DSAa
Alex QS
05/20/2025, 1:04 AMok. This was happening for both VapiVoice's and Cartesia voice. is VapiVoice built ontop of caretsia?
s
Shubham Bajaj
05/20/2025, 1:05 AMStrange, can you share other call ids as well?
a
Alex QS
05/20/2025, 1:07 AMyeah shared at the top.
75ed7849-dabb-41be-a00e-6c9baa099117 -> vapi voice
Alex QS
05/20/2025, 1:08 AMapprox timestamp for error: 18:05:43:631 - [LOG] Voice input: like a condo, hotel, or accessibility needs? (this line didn't get read by STT)
Alex QS
05/20/2025, 1:09 AMalso 6a07120b-d232-46ce-844c-4b1e32551c1f -> vapi voice
approx timestamp: 20:50:01:151 [LOG] Voice input: Thank you, Ms. Billones! Have you stayed with us before?
s
Sahil
05/20/2025, 7:36 PMsent you calendar link.
Sahil
05/20/2025, 7:36 PMLet debug it tomorrow.
a
Alex QS
05/21/2025, 9:41 PMall good, I think @Shubham Bajaj has debugged this and is following up on the resolution
a
Aditya
05/22/2025, 2:23 AMPlease reach out to us if you have any other questions regarding this issue