Voice, language & behavior

Open Settings → AI moderator to change anything on this page. Changes are held in your draft until you click Save at the bottom of Settings. If you close with unsaved edits, Strella asks whether to discard them.

Steps

1. Pick a language (or languages)

   Under Moderator language, choose one or more from the dropdown. Set Default language as the fallback when a session can't otherwise pick.

   • One language: every interview happens in that language. The voice picker filters to voices that speak it.
   • Multiple languages: the participant picks at the start of their call.
   • All languages (leave empty): same as picking every supported language.

   See Supported languages for the full list.

2. Pick a voice

   What you see depends on languages:

   • One English-only study: browse English voices by name and provider.
   • One non-English study: voices filter to ones that speak that language.
   • Multiple languages or all-languages: pick a gender (Feminine / Masculine / Neutral); Strella picks the right voice per language automatically.

   Click the play icon next to any voice to preview a sample. Set Voice speed with the slider. Participants can still adjust their own playback speed during the call.

   

3. Rename the moderator (optional)

   Use Change display name to override the default "Strella." The moderator introduces itself with that name, and the call UI refers to the moderator by it.

4. Pick a behavior mode

   Question behavior:

   • Adaptive (default): conversational; references prior answers; rephrases questions naturally.
   • Strict: reads each question verbatim; minimal improvisation.

   For one-off exact wording on a specific question, use the per-question Ask question verbatim toggle in your guide. It does the same thing for a single question.

5. Pick a flow

   Question flow:

   • Automatic (default): the moderator decides when each question has been answered and moves on.
   • Participant-controlled: the moderator stays silent and waits for the participant to click next. Won't interrupt.

   Automatic is more conversational. Participant-controlled is useful for studies where you need long pauses (e.g. think-aloud usability work).

6. Set reaction length

   Reaction length controls how much the moderator says between answers: Low (brief acknowledgements), Medium (default, natural conversational warmth), or High (longer reflections). Setting it to Low is one way to make the moderator less effusive between answers; to change which acknowledgements it uses, or stop it affirming answers altogether, use Affirmation tone (below).

7. Set task completion handling

   For Task steps, Task completion handling decides how the moderator knows the participant is done:

   • Default: the moderator judges based on what the participant says and confirms before moving on.
   • End task button: the participant clicks a button.
   • Explicit confirmation: the moderator waits for a verbal "done."

   A goal-based Figma task is the exception: reaching the goal screen ends the task automatically.

8. Turn on Adaptive pause detection (optional)

   Adaptive pause detection lets the moderator deliberately wait in silence, rather than filling every gap, when a participant is thinking or working through something. It resumes when the participant re-engages. Useful for tasks and think-aloud research.

9. Turn on Agent vision (beta, optional)

   Agent vision lets the moderator periodically look at the participant's camera and shared screen while media is up, so it can react to what they're doing visually instead of just what they say. Blurred sources stay invisible to it.

   This is a beta feature. If you don't see the toggle, contact your account team to enable it on your organization.

10. Customize tone and translation (optional)

    Three free-text fields shape the moderator's voice without touching the guide:

    • Affirmation tone: describe how the moderator acknowledges answers between questions ("neutral and brief, never enthusiastic"). Replaces the default acknowledgement style.
    • Custom behavioral instructions: guidance the moderator follows for the whole interview ("never suggest answers", "refer to the product as Acme One"). Won't override the built-in greeting or closing.
    • Custom translation instructions: steer how the moderator translates your guide ("use formal Dutch", "keep brand names untranslated").

    All three are optional. Defaults work for most studies.

The next interview will use these settings. Active interviews already in progress finish on their original settings.

Tips

• You can't replace the greeting or the goodbye. Both are built into the moderator. To add framing in your own words, put a Statement as the first or last item in your guide.
• One voice and language config per project, not per question. If you need different audiences in different languages, allow multiple languages and let participants pick at the start of the call.
• Preview the voice before publishing. Sample text won't catch jargon problems. Take a full preview interview to feel the actual pacing.
• Custom behavioral instructions can fight the moderator's defaults. Keep them additive ("never suggest answers") rather than contradictory ("always tell them what others said").

Related