https://vapi.ai/ logo
Join Discord
Powered by
General Transcriber Deafness/low volume/indistingu...
# support
o
Some Call IDs where there is a distinct problem....but it clearly is a general problem 3421672f-02a5-47fe-9491-cf7e46023004 a9582d0f-1e00-4aea-bef2-16c04c0be153 9c42c937-bbd0-4c6f-8dba-6ea57c560795 b766e68e-08cf-4660-9a69-adc6c4171266
v
# Troubleshooting Transcriber Deafness in Vapi If you're experiencing the "deafness bug" where the AI doesn't hear you, this is likely due to issues with the Deepgram transcriber. Try changing your transcriber with this simple fix:
Copy code
javascript
// Switch from Deepgram to another transcriber
// In your Vapi configuration:
{
  "transcriber": {
    "type": "whisper", // Try "whisper", "gladia", or "azure" instead of "deepgram"
    "useVapiKey": true // Use Vapi's key instead of your own
  }
}
This addresses the most common cause where Deepgram might be blocking your IP. Sources: - [Troubleshooting steps for General Transcriber Deafness in Vapi](https://discord.com/channels/1211482211119796234/1340235729204285510) - [Possible causes of General Transcriber Deafness in Vapi](https://discord.com/channels/1211482211119796234/1343612796717371393)
o
deepgram is the only transcriber that works well for our use case - with the optimum latency - for optimum quality of transcription
Other thing - this is not a pure deepgram issue... - this is a call with gladia transcriber: f8478746-1192-4f2c-a8ed-700bcd0acfea Notice how the transcriber can't hear very clear audio
k
To help track down this issue, could you share one or two: - When exactly this happened (the timestamp) - What response you expected to get - What response you actually got instead This would really help us figure out what went wrong!
o
https://dashboard.vapi.ai/calls/3421672f-02a5-47fe-9491-cf7e46023004 - at 6 second mark I say "haan bolo", then again at 11 second mark (both are affirmatives in hindi)...and yet the bot doesn't pick that up and asks "are you still there?" at about 15 second mark - at 27 second mark I again say "kitna chahiye" translated as "how many do you want?" and so on till the end of the call, but the transcriber fails to pick up all of those messages from the other end Expected was that the transcriber picks up the message and I get a reply back from the assistant
https://dashboard.vapi.ai/calls/f8478746-1192-4f2c-a8ed-700bcd0acfea [gladia transcriber] the whole call is just me saying "hello" and the transcriber not picking up literally any of the hellos
https://dashboard.vapi.ai/calls/a9582d0f-1e00-4aea-bef2-16c04c0be153 - around 21 second mark, I say "haan mil jayega" which is basically "yes you'll find it here"... but transcriber doesn't pick it up....and I say it again at the 28 second mark Both of these aren't picked up by the STT at all
k
Looking into it
o
@User Did you guys find anything?
c
current
Copy code
"startSpeakingPlan": {
        "waitSeconds": 0.1,
        "transcriptionEndpointingPlan": {
          "onPunctuationSeconds": 0.1,
          "onNoPunctuationSeconds": 0.8,
          "onNumberSeconds": 0.5
        },
        "customEndpointingRules": [],
        "smartEndpointingPlan": {
          "provider": "vapi"
        }
      }
updated
You need to change your start speaking plan.
Copy code
"startSpeakingPlan": {
  "waitSeconds": 0.6,
  "transcriptionEndpointingPlan": {
    "onPunctuationSeconds": 0.1,
    "onNoPunctuationSeconds": 1.2,
    "onNumberSeconds": 0.5
  }
}
- removed 
smartEndpointingPlan
  because it only works for English languages. - changed 
onNoPunctuationSeconds
is changed to 1.2, because  when set to 0.8 value it may be slightly aggressive for non-English language. This change gives your assistant more breathing room to correctly identify when a user has actually finished speaking versus just interrupting momentarily. If interruptions happen, increase onNoPunctuationSeconds by 0.2 until it feels natural. If responses seem too slow, decrease by 0.1.
3 Views
PreviousNext
Powered by