https://vapi.ai/ logo
Join Discord
Powered by
Support Request: Code Switching + Transcriber Not ...
# support
a
Hi Vapi Team, I'm encountering an issue with my assistant configuration while using the Vapi Dashboard: Transcriber Provider: #Deepgram Model: #nova-2 Language: #multi codeSwitchingEnabled: true (intended) Voice Model: #ElevenLabs Turbo 2.5 I’m trying to support both English and Spanish using the multi language setting with codeSwitchingEnabled: true. However, when I configure this via the dashboard and publish the assistant: The codeSwitchingEnabled setting is not saved (it either reverts to false or disappears). The transcriber section appears disabled in the Vapi Dashboard, and I can’t manually re-enable or edit it reliably. After saving and publishing, the assistant does not behave correctly during multilingual calls. I'm using the Vapi Dashboard to configure everything — no API calls involved at this stage. Has there been a recent update affecting transcriber config or code-switching in the dashboard UI? Is there a workaround (such as using Advanced JSON or the Vapi API) to ensure codeSwitchingEnabled: true works correctly with language: "multi"? Here is the assistant I’m working with: Assistant ID: c40dd973-403e-4210-910b-bc80b6a1f491 Appreciate any help or clarification. Thanks so much!
By the way it was perfectly working a few hours ago...
v
Hi there! 👋 I understand you're having trouble with the codeSwitchingEnabled setting not saving properly in the Vapi Dashboard. I'm unable to access information about your specific assistant ID, but I can definitely help with the issues you're experiencing with saving the codeSwitchingEnabled setting and provide some workarounds. ## Known Issues with Vapi Dashboard There are indeed known issues with saving settings in the Vapi Dashboard. According to [multiple reports in the Discord community](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1334552736133546076), users have experienced problems with various settings not being saved correctly, even when they receive a "Success" message. ## About codeSwitchingEnabled and Language Support There are some important limitations to be aware of: 1. According to [Discord discussions](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1308120552140964034), codeSwitchingEnabled is only supported for English and Spanish. 2. The codeSwitchingEnabled setting is only compatible with 
nova-2
and 
nova-2-general
models, not with 
nova-3
which only supports 'en' and 'en-US' [as reported in Discord](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1341463939246198784). ## Workarounds Here are some workarounds you can try: ### 1. Using the Vapi API Several users have reported success using the API instead of the dashboard to configure settings. You can use JSON configuration like this:
Copy code
json
{
    "transcriber": {
        "provider": "deepgram",
        "model": "nova-2",
        "language": "es",
        "codeSwitchingEnabled": true
    }
}
### 2. Try Automatic Language Detection
According to [Discord discussions](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1328595920899674152), you can try using automatic language detection instead of manually setting the language. Don't pass any language and instead use automatic multiple languages detection. ### 3. Squad of Assistants Approach Some users have implemented a workaround by [creating a squad of assistants](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1308120552140964034), where the first assistant asks which language the user prefers and then transfers them to the appropriate language-specific assistant. ### 4. Advanced JSON Configuration You might be able to adjust the 
chunkPlan
configuration using Advanced JSON to improve multilingual support:
Copy code
json
{
  "voice": {
    "chunkPlan": {
      "enabled": true,
      "minCharacters": 30,
      "punctuationBoundaries": [".", "!", "?", ";"]
    }
  }
}
## Dashboard Improvements Coming The good news is that Vapi is working on a new dashboard version (dashboard v2) that is expected to fix many of these issues, [according to Discord discussions](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1334868141280329850). Is there anything specific about these workarounds you'd like me to elaborate on? Sources: - [Discord thread about code switching issues](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1308120552140964034) - [Discord thread about model compatibility](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1341463939246198784) - [Discord thread about automatic language detection](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1328595920899674152) - [Discord thread about dashboard issues](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1334552736133546076)
p
Known issue, vapi team on it
a
message has been deleted
m
what does codeSwitchingEnabled do exactlu
it seems to be mostly for formatting?
t
where can I add this section in vapi: { "transcriber": { "provider": "deepgram", "model": "nova-2", "language": "es", "codeSwitchingEnabled": true } }
k
checking if this is resolved/solved for you?
PreviousNext
Powered by