I’m currently facing an issue with our voice agent built on Vapi. The agent is misinterpreting user inputs for names and email addresses. Even when users say things like “john dot doe at gmail dot com,” the agent is either transcribing it incorrectly or not processing it into a valid email format. The agent often returns incorrect spellings of names. Email inputs like “name at gmail dot com” are not being converted into proper format (e.g., name@gmail.com). Confirmation logic is either missing or not reliably verifying the inputs with the user.

Thanks for sharing the details. To help you further, could you let me know if you are currently using the keyword boosting feature (e.g., adding names or email domains as keywords) in your Vapi assistant’s transcriber configuration? This will help determine if further tuning or configuration is needed for better transcription accuracy. If you need, I can provide an example of how to set up keyword boosting for names and emails. Sources: - [Custom Keywords (Keyword Boosting) in Vapi](https://docs.vapi.ai/customization/custom-keywords) - [Voice Formatting Plan](https://docs.vapi.ai/assistants/voice-formatting-plan)

Hi Jhon Smith, To address the issues with your voice agent misinterpreting inputs, especially names and email addresses, you can take the following steps: 1. **Transcription Accuracy**: - Use more accurate transcriber models. Check the [Multilingual documentation](https://docs.vapi.ai/customization/multilingual) for a list of available transcriber providers and models. Switching to a provider that supports better language models can improve transcription accuracy. 2. **Custom Keyterms for Spelling**: - Implement Deepgram’s Keyterm Prompting technology to boost the recall for specific terms, particularly for names and email patterns. More information on this can be found in the [changelog update](https://docs.vapi.ai/changelog/2025/2/20) related to Deepgram Keyterm Prompting. 3. **Test Components Individually**: - Use the [dashboard's testing tools](https://docs.vapi.ai/debugging) to test individual components like speech recognition and spelling correction separately: - **Call Logs**: Check error messages and confirm if inputs are being transcribed correctly. - **Tool Testing**: Validate the agent’s logic when parsing email addresses. 4. **Response Logic**: - Ensure that your response logic is configured to handle common patterns when users are spelling out email addresses. Consider enabling patterns that intelligently recognize "dot" and "at" as "." and "@". 5. **Debugging Tools**: - Make use of the Quick Diagnostics tools detailed in the [Debugging voice agents guide](https://docs.vapi.ai/debugging). This includes reviewing the call logs for incorrect transcription instances and testing the assistant in the “Talk to Assistant” feature. Implement these steps to enhance the agent’s performance in handling names and email addresses correctly.