Skip to main content
Voice choice strongly affects trust, clarity, and perceived competence. A good voice setup makes the agent feel calm, intentional, and predictable — especially in call scenarios where users cannot re-read responses. This step should be completed after agent behaviour is stable, but before serious call testing.

​
Where this lives

  • Configure → Agent voice
You will see two configurable sections:
  • Agent (the primary speaking voice)
  • Disclaimer (used only for legal or informational disclosures, if enabled)

​
What you are configuring

Each voice entry has three layers:
  1. Voice selection (who the agent sounds like)
  2. Weighting (if multiple voices are used)
  3. Voice tuning (stability, clarity, speed)
You can use a single voice or blend multiple voices together.

​
Step-by-step

​
1. Select the agent voice

In the Agent section:
  • Click Change
  • Browse voices by:
    • Language / locale
    • Gender
    • Voice type (for example, PolyVoice vs standard)
  • Preview each voice using the play button
Choose a voice that:
  • Matches your audience’s expectations
  • Sounds neutral and professional
  • Remains pleasant over repeated listening
Example (hotel concierge):
  • Language: English (United States)
  • Voice: Female, calm, mid-range pitch
  • Avoid novelty or character voices
Example (retail support):
  • Language: English (United Kingdom)
  • Voice: Neutral, friendly, slightly upbeat
Once selected, click Select.

​
2. (Optional) Add multiple agent voices

You can add more than one agent voice and split usage between them.
  • Click + Voice
  • Select an additional voice
  • Assign a percentage weight (for example, 70% / 30%)
This is useful when:
  • You want subtle variation across calls
  • You are testing two voices in parallel
Example:
  • Voice A (primary): 70%
  • Voice B (secondary): 30%
Avoid adding more than 2–3 voices unless you are deliberately testing variation.

​
3. Configure voice settings (gear icon)

Next to each voice, click the Settings (gear) icon to open advanced tuning. You will see: Voice model
  • Leave as default unless advised otherwise.
Stability (%)
  • Controls how consistent the voice sounds.
  • Higher = more predictable, less expressive.
Recommended ranges:
  • Call agents: 70–90
  • Informational agents: 60–80
Example: Stability: 80 Result: calm, consistent delivery across long calls
Clarity and similarity (%)
  • Controls how close the generated voice is to the original recording.
  • Higher = more realistic, but slightly less forgiving.
Recommended ranges:
  • Dynamic conversations: 80–95
  • Scripted or repetitive content: 95–100
Example: Clarity: 100 Result: very natural, human-like voice
Voice speed
  • Controls pacing.
  • Default is usually correct.
Recommended ranges:
  • Calls: 0.95–1.05
  • Long explanations: slightly slower (0.95)
Example: Speed: 1.05 Result: efficient without sounding rushed
Click Done to save settings.

​
4. Configure disclaimer voice (if enabled)

If your agent uses a disclaimer:
  • Expand the Disclaimer section
  • Select a voice (can be the same or different from the agent)
Best practices:
  • Use a clear, neutral voice
  • Avoid expressive or casual tones
  • Keep speed slightly slower than the main agent
Example disclaimer text: “This call may be recorded for quality and training purposes.”
Example settings:
  • Stability: 85
  • Clarity: 100
  • Speed: 0.95

​
5. Preview real utterances

Before publishing, preview the voice using real agent lines, not generic samples. Test:
  • Greeting
  • Clarifying question
  • SMS offer
  • Handoff offer
  • Closing line
Example preview phrases: “How can I help you today?” “I can send you a text with the link — would you like me to do that now?” “I can connect you to a colleague if you’d prefer.”
Listen for:
  • Natural pauses
  • Clear pronunciation
  • No rushing or slurring

​
6. Publish changes

Once satisfied:
  • Click Publish
  • Changes will apply to the selected environment
Remember: voice settings are environment-aware. Always test in Sandbox before Live.

​
Common mistakes to avoid

  • Using overly expressive or novelty voices
  • Setting stability too low (causes inconsistency)
  • Speaking too fast for phone audio
  • Forgetting to preview handoff language
  • Changing voices without retesting call flows

​
Verify

After publishing:
  • Make at least one test call
  • Listen to:
    • Greeting
    • Mid-call responses
    • Handoff or SMS offers
  • Confirm:
    • Voice sounds consistent
    • No awkward pacing
    • Audio matches expectations across turns
If something feels off, adjust stability first, then speed. Voice tuning is iterative — small adjustments often make a big difference.