Voice & Response Issues

If you’re experiencing issues with voice quality or response timing, you can adjust several settings to optimize performance.

Response Speed vs. Accuracy

Your AI assistant has adjustable engine settings:

Dualplex: lowest latency, voice cloning available
Pipeline: transcription, model, speech — the widest choice of voices, with a pause between turns
Speech-to-speech: native voice generation, no transcription step

Pick one in General → Engine type. How accurate the answers are comes from the model, in Brain & prompt → Model.

To minimize silence during processing:

Click Enable fillers in Voice & speech → Filler audios
The AI will use natural sounds (like “hmm” or “one moment”) while processing
The section only appears on the Pipeline engine, which is the one with a pause to cover

Slow Responses
- Switch to the Dualplex engine
- Enable Filler Audio
- Simplify your system prompt
Inaccurate Responses
- Pick a stronger model in Brain & prompt → Model
- Provide more context in your prompt
- Add specific examples
Voice Quality
- Try different voice models
- Adjust speech speed settings
- Try cloning a voice
Assistant Not Responding or Missing Speech in Transcript
- Disable noise cancellation in Call behavior → Audio & recording
- Noise cancellation can sometimes filter out speech or cause the assistant to not respond
- Test with noise cancellation disabled to see if speech is properly detected

For sales calls: Dualplex, or Pipeline with filler audio if you need a specific voice
For support: keep the engine and pick a stronger model in Brain & prompt → Model

Need help? Contact support if these adjustments don’t resolve your issues.