Skip to main content

Troubleshooting AI Agent Call Quality

Learn how to identify and resolve common issues that affect the clarity, accuracy, and consistency of your AI agent’s calls.

Maria Cairns avatar
Written by Maria Cairns
Updated over a week ago

Why it matters

High-quality audio and accurate transcription are essential for reliable call performance. If callers cannot hear the agent clearly, or if the agent mishears callers, instructions and workflows break down. Troubleshooting call quality ensures the agent sounds natural, understands callers correctly, and delivers consistent results. Although AI agents can also message, this guide focuses specifically on call-related quality issues.

Key Concepts

Speech-to-text mode: Controls how caller audio is transcribed. Accuracy mode improves precision, while speed mode reduces latency.

Denoising: Removes background noise and, optionally, background speech to improve transcription quality.

Voice stability: Controlled by voice temperature and speed settings.

Ambient sound: Optional background noise that can enhance or interfere with clarity.

Interruption behavior: Defines how easily the caller can interrupt the agent, shaping the flow of the conversation.

Step-by-Step: Diagnose Call Quality Issues

  1. Review a recent call transcript and recording in the Calls tab.

  2. Identify whether issues relate to audio clarity, transcription accuracy, or spoken output from the agent.

  3. Check the agent’s speech-to-text mode in the call-related settings.

  4. Review denoising settings to ensure appropriate noise reduction is enabled.

  5. Examine voice temperature and voice speed to confirm delivery sounds natural.

  6. Verify that boosted keywords and pronunciation entries are set for domain-specific terms.

  7. Use a phone preview to reproduce the behavior and validate your assumptions.

  8. Adjust settings and retest until call quality improves.

Tips and Best Practices

Improving Transcription Quality

  • Use Accuracy mode when callers use technical or domain-specific vocabulary.

  • Enable noise + background speech cancellation in noisy call environments.

  • Add boosted keywords for names, product terms, or jargon the agent commonly misinterprets.

Improving Agent Voice Quality

  • Keep voice temperature low (0 to 0.4) for stable and predictable delivery.

  • Adjust voice speed to natural ranges between 0.9 and 1.1.

  • Disable ambient sound if clarity is more important than realism.

Reducing Interruptions or Overlapping Speech

  • Lower interruption sensitivity if the agent stops speaking too easily.

  • Decrease responsiveness slightly if the agent reacts too quickly to caller speech, causing overlap.

Testing and Iteration

  • Always use phone preview to validate behavior under real audio conditions.

  • Compare call recordings over time to identify recurring patterns.

  • After changing settings, review multiple new calls to ensure improvements are consistent.

Troubleshooting

Issue

Possible Cause

Fix

Agent mishears caller

STT mode set to speed; denoising disabled

Switch to accuracy mode and enable denoising

Agent sounds robotic

Voice temperature too low or voice speed too high

Increase temperature slightly or reduce speed

Background noise in recordings

Insufficient denoising

Enable noise cancellation or noise + background speech cancellation

Agent cuts off mid-sentence

Interruption sensitivity too high

Lower interruption sensitivity

Caller and agent overlap frequently

Responsiveness too high

Reduce responsiveness

Incorrect transcription of key phrases

Missing boosted keywords

Add relevant boosted keywords or pronunciation entries

Did this answer your question?