conscious-sapphire•3mo ago

Weird problem...my agent refuses to say a phone number correctly.

I have a simple agent that takes pizza orders. If the caller needs further assistance my agent is supposed to say "please call 310-555-5555" But strangely, it will not say the entire phone number correctly no matter what I do. It screws up the "0" in the number and says some mumbo jumbo. Then if you ask it to repeat the number it rushes through the number like there's a weird fast glitch. I've tried changing LLMs and every one does the same thing. Any ideas?

9 Replies

Vapi•3mo ago

Try disabling the default voice formatting for phone numbers, which may be causing the mispronunciation.
Add this to your assistant configuration:

"voice": {
  "chunkPlan": {
    "formatPlan": {
      "enabled": false
    }
  }
}

"voice": {
  "chunkPlan": {
    "formatPlan": {
      "enabled": false
    }
  }
}

This will make the TTS read the phone number exactly as written in your prompt, which should resolve the issue with "310-555-5555" being spoken incorrectly.
See documentation for details. Source: - Voice Formatting Plan documentation

⤵️ Mention @Vapi Support Bot for follow-ups 🐥

Vapi•3mo ago

Try disabling the default voice formatting for phone numbers, which may be causing the mispronunciation.
Add this to your assistant configuration:

"voice": {
  "chunkPlan": {
    "formatPlan": {
      "enabled": false
    }
  }
}

"voice": {
  "chunkPlan": {
    "formatPlan": {
      "enabled": false
    }
  }
}

⤵️ Mention @Vapi Support Bot for follow-ups 🐥

conscious-sapphireOP•3mo ago

ok do i put that code right into the system prompt area?

ratty-blush•3mo ago

You can place the SSML or spaced-digit format directly in the system prompt to improve phone number pronunciation..

conscious-sapphireOP•3mo ago

@Duckie ok do i put that code right into the system prompt area? What do you mean by SSML?

Vapi•3mo ago

No, do not put that code in the system prompt area.
Instead, add the following configuration to your assistant's JSON settings (outside the prompt text):

"voice": {
  "chunkPlan": {
    "formatPlan": {
      "enabled": false
    }
  }
}

"voice": {
  "chunkPlan": {
    "formatPlan": {
      "enabled": false
    }
  }
}

This change should be made in your assistant's configuration file or via the API, not in the prompt or message content.
See documentation for details. Source: - Voice Formatting Plan documentation No, do not put that code in the system prompt area.
Instead, add the following configuration to your assistant's JSON settings (outside the prompt text):

"voice": {
  "chunkPlan": {
    "formatPlan": {
      "enabled": false
    }
  }
}

"voice": {
  "chunkPlan": {
    "formatPlan": {
      "enabled": false
    }
  }
}

This change should be made in your assistant's configuration file or via the API, not in the prompt or message content.
See documentation for details. Source: - Voice Formatting Plan documentation

ratty-blush•3mo ago

Speech Synthesis Markup Language (SSML) is a markup language used to control how text is spoken by AI voices, like ensuring phone numbers are read as individual digits or in standard formats, which improves pronunciation..

conscious-sapphireOP•3mo ago

Cool, how do I use that within Vapi?

ratty-blush•3mo ago

Choose a TTS provider that supports SSML like Azure, wrap your assistant’s responses in <speak>...</speak> with tags like <say-as> or <break>, and send it via the Vapi API or SDK using type: "ssml" to control pronunciation, pacing, and emphasis..

Weird problem...my agent refuses to say a phone number correctly.

Did you find this page helpful?