How do I improve AI agent speech recognition with keywords and pronunciations?

Last updated: November 5, 2026

Boosted keywords and pronunciation entries help a voice AI agent recognize and say important terms more accurately during calls. Use them for names, product terms, locations, acronyms, and other words callers mention often.

Before you start

  • Identify the words or phrases the agent commonly mishears or mispronounces.

  • Review recent call transcripts to confirm the issue.

  • Prepare phonetic guidance for any words that need pronunciation overrides.

Steps

  1. Open the AI agent.

  2. Go to the voice or interaction settings.

  3. Find the Boosted keywords section.

  4. Add important words or phrases the agent should recognize reliably.

  5. Find the Pronunciation guide section.

  6. Choose the pronunciation notation format, such as IPA or CMU, if prompted.

  7. Enter the word and its phonetic spelling.

  8. Click Add entry or save the pronunciation.

  9. Use phone preview to test the updated recognition and pronunciation.

  10. Review new calls to confirm the changes improved accuracy.

Tips

  • Add only important terms so the keyword list stays focused.

  • Use boosted keywords for names, companies, cities, products, and technical terms.

  • Use pronunciation entries for uncommon names, brand-specific words, abbreviations, and acronyms.

  • Pair keywords with denoising and accuracy-focused speech-to-text settings for noisy calls.

  • Test pronunciation entries across multiple preview calls.

Troubleshooting

  • The agent repeatedly mishears a name: Add it as a boosted keyword and create a pronunciation entry.

  • Keywords still misrecognize: Remove low-priority keywords and focus on the most important terms.

  • A pronunciation entry does not work: Validate the IPA or CMU notation.

  • The agent struggles in noisy calls: Enable denoising or switch speech-to-text to accuracy mode.