Skip to main content
If the agent sounds robotic, unnatural, distorted, or inconsistent, the problem is usually in the voice setup rather than the platform infrastructure. In most cases, voice quality issues come from settings such as voice temperature, speed, interruption sensitivity, denoising, or the interaction between the voice provider and the selected language model.

​
Agent sounds robotic or flat

If the voice feels monotone or lifeless, the most common cause is that the voice settings are too conservative. Start by checking:
  • voice temperature,
  • voice speed,
  • and the overall voice choice.
If temperature is set too low, the voice can sound overly controlled and flat. Increasing it slightly often helps the delivery feel warmer and more natural. Voice speed matters too. If the speed is too slow, the agent can sound lethargic. If it is too fast, the result can feel rushed and unnatural.

​
Agent talks over the contact

If the agent keeps speaking when the contact starts talking, interruption sensitivity is usually too low. Increase it so the agent yields more quickly when it detects the other speaker. If the problem persists, check the denoising mode as well. In noisier environments, stronger denoising can help the speech recognition layer distinguish the contact’s voice from background sound more accurately.

​
Agent gets interrupted too easily

If the agent stops speaking every time it hears a small noise, interruption sensitivity is probably too high. Lower it slightly so the agent can continue through light background sounds, reactive noises, or non-verbal interruptions without constantly giving up the floor. This is especially common in calls where the contact is:
  • driving,
  • outside,
  • in a call centre,
  • or in another noisy environment.

​
Audio has static, echo, or distortion

If the audio contains static, echo, clipping, or distortion, the issue is often on the telephony side rather than in the voice engine itself. If it happens only on certain numbers or carriers, the likely cause is the call routing path. If it happens across many calls, check the audio settings and make sure denoising is enabled appropriately. If the issue continues, the best next step is to review an affected recording and use that specific call as the basis for troubleshooting.

​
The live voice sounds different from the preview

This is normal to some extent. The preview in the voice gallery is a clean sample. A live call uses the full voice pipeline:
  • speech recognition,
  • language model generation,
  • text-to-speech,
  • and telephony encoding.
That full chain can change how the voice feels in practice. In addition, the selected language model affects the wording, pacing, and sentence shape being sent into the voice engine. That means the same voice can sound noticeably different depending on which model is driving it. If a voice sounds wrong in production, do not assume the voice itself is the problem immediately. Test the same voice with a different model first.

​
What to adjust first

If voice quality feels off, work through the settings in this order:
  1. check voice temperature,
  2. check voice speed,
  3. adjust interruption sensitivity,
  4. review denoising,
  5. test the same voice with a different model,
  6. then compare with a different voice provider or voice option if needed.

​
Best practice

Always judge voice quality from real call recordings, not just the voice preview. The preview helps you narrow options, but the real test is how the voice sounds:
  • in a live conversation,
  • through the chosen model,
  • over the phone network,
  • with the actual script and pacing of the agent.
That is the only version that matters in production.