Tool call Result is a mix of numbers and strings. It is received properly but while speaking the agent is speaking, it is randomly skipping numbers and just saying the text. When I see the message log, the numbers are being converted to Asterisk and dashes. Its a transient agent.

Hi there! 👋 I see you're having issues with your Text-to-Speech not properly handling numbers in tool call results. I don't have specific information about why numbers are being converted to asterisks and dashes in your message log, or if there are specific compatibility issues with transient agents. However, I can definitely help explain why your TTS might be skipping numbers and only speaking the text. Based on the information I've found, there are several potential causes for this behavior: 1. **Voice Formatting Plan**: The

voice.chunkPlan.formatPlan

configuration might be causing the TTS to skip numbers. This is a common issue that affects how numbers are processed before being sent to the TTS engine. [According to a Discord thread](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1334184334361628734), this can cause numbers to be skipped. 2. **`numberToDigitsCutoff` Setting**: If this setting is too low, numbers may be converted to individual digits instead of being spoken as words. You can adjust this in your configuration:

json
{
  voice: {
    chunkPlan: {
      formatPlan: {
        numberToDigitsCutoff: 100000 // Set this higher than your largest expected number
      }
    }
  }
}

[This solution was shared in a Discord thread](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1342349738032500736). 3. **Voice Input Formatting**: The "Voice Input Formatted" function, which is enabled by default, transforms numbers before they're passed to the TTS engine. [According to Vapi's documentation](https://docs.vapi.ai/assistants/voice-formatting-plan), this feature cleans up raw text to make it more readable for TTS, but it might be causing your issue. You can try turning this feature off.

022893ff-19a2-4c6e-95fa-a52f0f580fc8 this is the call id

This is happening because of Voice Formatting Plan which means right before sending your output to the Voice provider we apply some formatting to the LMM output and during this we remove by default markdown syntax contained content or other special characters so that the content which goes to the voice provider is of high quality and its easy to produce a better sounding sound. Under your voice settings you can toggle LLM output formatting utilities functions. This can only be done using the API, so let me know if you require any further help with this.

Thanks Shubham will try playing with it..also do you have suggestions for text to Voice Punjabi language model?

looking into it.

THe model introduces asterics and eats up my numbers along with it

can you try out new call with following voice config:

json
{
  "voice": {
    "provider": "VOICE_PROVIDERS.VAPI", // or whatever provider you're using
    "voiceId": "your-voice-id",
    "chunkPlan": {
      "enabled": true,
      "formatPlan": {
        "enabled": true,
        "formattersEnabled": [
          "markdown",
          "asterisk",
          "number"
        ]
      }
    }
  }
}

I think this works, will keep testing thanks

Hi, checking if this is resolved/solved for you?