Issue with assistant not speaking input parameters...
# support
s
Hi team it's little frustrating when despite of passing the tools correctly the assistant does not ask the input parameters properly. Please refer below call Ids where you can see the assistant does not speak out the input parameters properly it just omits the words.
157e3f15-5a58-4db7-95c2-2ce92cf8b768
a0471f25-c35b-4f20-8e6b-6c5cbf1dcaa7
Being honest, from a user's perspective this kind of issues makes us loose our interest to work with vapi. Please look into this on priority and help me with some solution.
v
Hi there! I understand your frustration with the assistant not speaking input parameters properly. This can definitely be a challenging issue to deal with. I'm unable to access the specific details of your call IDs, but I can help explain some potential causes and solutions for this problem. Based on the information I've found, there are several possible reasons why your Vapi assistant might not be speaking input parameters properly: ## Potential Configuration Issues 1. **Voice Formatting Plan**: This issue could be related to your assistant's voice formatting plan. According to the [documentation](https://docs.vapi.ai/documentation/core-concepts/assistants/voice-formatting-plan), you may need to change the default
voice.chunkPlan.formatPlan
config to match your requirements. 2. **Speaking Plan Settings**: The
startSpeakingPlan
configuration might need adjustment. Setting Deepgram's endpointing value to 300 has helped resolve similar issues for other users. 3. **Transcriber Issues**: You might want to try using
modelOutputInMessages
instead of transcribed output in messages. This can be enabled by making an UPDATE request to your assistant. ## Known Bugs There are some known bugs that could cause this behavior: 1. **Values Not Being Read**: There's a known issue where "the voice agent fails to read out loud the values retrieved from a search," resulting in silence or placeholders not being replaced with retrieved values. 2. **Tool Call Interpretation**: The model may interpret a tool call as spoken output instead of executing the command, which can affect how input parameters are spoken. ## Possible Solutions Here are some steps you can take: 1. **Update Assistant Configuration**: Make an UPDATE assistant request using the API client or Postman to adjust the speaking plan and transcriber settings. 2. **Check SSML Settings**: Enabling SSML can give you more control over speech synthesis, including pauses and pronunciation.
3. **Review Language Settings**: Ensure that the language setting for the voice model used for outbound communication is correctly specified. 4. **Contact Vapi Support**: Since you've already identified the call IDs, you can reach out to Vapi support with these IDs. They can use them to check call status updates, logs, and debugging artifacts to identify the specific issue in your case. Would you like me to provide more details on any of these potential solutions? Sources: - [Reasons why Vapi assistant is not speaking input parameters properly](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1334184334361628734) - [Known bugs in Vapi that cause the assistant to not speak input parameters properly](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1334190896090058894) - [Vapi assistant configuration options that affect how input parameters are spoken](https://docs.vapi.ai/customization/speech-configuration) - [How to use Vapi call Ids to investigate issues with the assistant](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1334159219649417289)
@Vapi please help me out here
k
Hey! To help track down this issue, could you share: - What response you expected to get - What response you actually got instead This would really help us figure out what went wrong!
Here is my interpretation of your issue, do correct me if I am wrong here. When the assistant asks the user for ServiceNow ticket details, assistant omitting key words in her requests for parameters, making prompts confusing and incomplete. For example: 1\. When asking for ticket information, assistant says: "Could you please provide me with the following details for the ticket? 1, of the issue. 2, of the ticket if applicable. 3. High, medium, low, 4, high, medium, low, 5, of the issue." 2\. Later assistant says: "Please provide the following details. 1, tickets state new in progress, on hold, resolved, closed, canceled. 2," The assistant is clearly missing critical words when requesting parameter inputs, such as: \- Not saying "short description" when asking for point 1 \- Not saying "priority" or "urgency" when listing high/medium/low options \- Incomplete parameter requests that end abruptly This is causing confusion and frustration .
s
you are correct this is exactly what i wanted to convey
s
@Sahil 1. **Enable SSML Parsing**: The code shows that when
enableSsmlParsing
is false, the system removes angle bracket content from the text via the
removeAngleBracketContent
function. This could be stripping out parameter names if they're enclosed in formatting tags. 2. **Adjust Format Plan**: The assistant's voice format plan needs to be configured to preserve parameter names:
Copy code
json
{
  "voice": {
    "enableSsmlParsing": true,
    "chunkPlan": {
      "enabled": true,
      "formatPlan": {
        "enabled": true,
        "replacements": []
      }
    }
  }
}
3. **Add Custom Replacements**: If needed, you can add custom replacements to ensure parameter names are spoken correctly:
Copy code
json
"replacements": [
  {
    "type": "exact",
    "key": "short description",
    "value": "short description"
  },
  {
    "type": "exact",
    "key": "urgency",
    "value": "urgency"
  },
  {
    "type": "exact",
    "key": "priority",
    "value": "priority"
  }
]
4. **Implement Tool Response Guidelines**: Based on the prompting guide found, include specific instructions in your system prompt to ensure proper parameter handling:
Copy code
[Warning]
Do not modify or attempt to correct user input parameters or user input, Pass them directly into the function or tool as given.

[Response Guidelines]
When asking for input parameters, always state the complete parameter name before requesting its value.
For example, say "Please provide the short description of the issue" rather than just "of the issue."
5. **Update Assistant Configuration**: Make an API call to update your assistant with these changes:
Copy code
javascript
// Example API call to update assistant configuration
const updateAssistant = {
  voice: {
    enableSsmlParsing: true,
    chunkPlan: {
      enabled: true,
      formatPlan: {
        enabled: true
      }
    }
  },
  model: {
    systemPrompt: "... [include tool parameter handling instructions from above] ..."
  }
};

// PUT request to /assistant/{assistantId}
2 Views