Voice lets a learner talk with a simulation's character, or with a coach, out loud instead of typing. Voice is not limited to particular plans, and anyone who can edit the simulation or coach can set it up. This page covers turning voice on, picking a voice, and the settings that go with it.
Before you start
- The simulation or coach must already exist. See Create a simulation.
- Voice is charged at 120 credits per minute of session time, on top of the credits for the AI's replies. A voice session only starts when the workspace balance covers a full-length session, so the session length you set below also decides how many credits must be available. See Credits, usage and low-credit alerts.
- Voice is off by default for a new simulation and for a new coach.
Steps
To set up voice for a simulation:
- Open the simulation. In the Design stage, select Simulation and find the Voice heading.
- Check Enable real-time voice mode. The rest of the voice settings appear.
- Choose a voice under Character voice. When the list has loaded and no voice is chosen yet, the first voice in the list is selected for you. Select Preview to hear a sample of the chosen voice.
- Drag Max session length to the number of minutes a voice session may last, from 1 to 25. The default is 5.
- Leave Show transcript by default checked, or clear it. It is checked by default.
- Optional: under Enable max turns, set Max Turns and check Also enforce this turn limit in voice mode. This checkbox only appears while voice and the turn limit are both turned on, and it is off by default.
- Select Save Changes.
To set up voice for a coach:
- Open the coach. In the Design stage, select Personality.
- Check Enable voice mode.
- Choose a voice under Coach voice, and set Max session length and Show transcript by default. They work the same way as for a simulation, with the same limits and defaults.
- Select Save Changes.
To try it, open the Test stage. Once voice is saved as on, the stage shows a Text and Voice picker, and Voice runs a voice session against the saved settings, including on a draft. If the background setup described below has never completed for this simulation, for example because it failed, voice is set up when the voice session starts. If that also fails, the session does not start: save again and retry. A test voice session uses credits like any other voice session.
Result
Saving stores the settings. When voice is on and a voice is chosen, the editor then sets up the voice session in the background, and shows "Syncing voice agent..." under the voice settings while it does. The save itself does not wait for this step.
In the Publish stage, the links and embed snippets now come in two forms, labelled Text variant and Voice variant. Back in the Design stage, the voice settings show a Voice test link once voice is checked. That link opens the same page as the voice variant of the standalone link, so it only works once the simulation or coach is published:
- Published: the link is shown with a Copy URL button.
- Draft simulation: the link and Copy URL are shown with the note "This simulation is a draft. The link opens once you publish. To try voice mode before then, use the Test stage."
- Draft coach: no link is shown. The note reads "Publish your coach first to get its voice test link. To try voice mode before then, use the Test stage."
Publishing a simulation with voice on also runs the background voice setup again with the published content.
Voice cannot be turned on without a voice. Saving a simulation that turns voice on with no character voice, or that clears the voice while voice is on, is refused with "Select a character voice before turning on voice mode." A coach cannot be saved with voice on and no voice chosen. The editor shows "Select a voice" instead of saving. A simulation saved with voice on and no voice before this check existed can still be saved, and cannot be published until a character voice is chosen: publishing stops with "Pick a character voice in the Simulation section of the Design stage before publishing."
What each setting does
- Character voice and Coach voice: the voice learners hear. The list shows voice names only. If it cannot be loaded, the editor shows "Couldn't load voices. Please refresh, and contact support if it keeps happening."
- Max session length: a voice session ends when it reaches this length, and the learner sees the limit next to the running timer. If the background setup after a save fails, a changed limit is applied again when the next voice session starts.
- Show transcript by default: when checked, the transcript panel opens by itself once the first line of the conversation has been transcribed. When cleared, the panel stays closed. Either way the learner can open it with "Show transcript" and close it with "Hide".
- Also enforce this turn limit in voice mode (simulations only): when off, a voice session ignores the simulation's turn limit. It still ends on an end condition or when the session length is reached. When on, the character wraps up and the simulation ends once the turn limit is reached. See How a simulation ends.
The switches for offering voice on text links and text on voice links, and for offering text when a voice session fails, are covered in Letting learners switch between text and voice.
Pronouns and the chosen voice
Each voice in the list can carry a gender label. On a simulation, the Pronouns field under the character settings uses it:
- If no pronouns are set and the chosen voice is labelled female or male, the editor offers a one-click suggestion of matching pronouns.
- If the pronouns are she/her or he/him and the voice is labelled the other way, the editor shows a note asking you to double-check the pairing. It does not block saving. Publishing still succeeds and shows a "Heads up" notice with the same point.
- They/them and custom pronouns are never flagged.
Language
A simulation's voice sessions use the simulation's own language, and a learner who opens the simulation in a language you added hears and speaks that language.
A coach has a separate voice language. It is English by default, and the coach editor has no control for it. The coach's voice sessions use it, and so do the voice screen's labels and the mode-switch and transcript labels on the coach page.
Where coach voice applies
Coach voice applies to debriefs and to direct conversations with the coach. While a linked simulation is still running, the coach's voice option is not offered and real-time guidance stays in text, so the coach never talks over the simulation.
If voice setup fails or is flagged
- "Voice mode could not be updated with your latest changes, so voice sessions may fail to start or use the previous settings. Text mode is not affected. Save again to retry, and contact support if it keeps failing." means the settings were saved but the background setup did not finish. Save again. Two shorter forms name the cause: "Voice mode was not updated because too many updates were sent in a short time. Wait a minute, then save again." and "Voice mode could not be updated because the connection dropped. Check your connection, then save again."
- If the setup fails while a simulation is being published, the simulation is still published and the editor shows a "Heads up" notice: "The simulation is published, but voice mode could not be updated with the published content. Voice sessions may fail to start or use the previous settings. Save the Simulation section of the Design stage to try again."
- A warning that an automated safety check flagged the character's prompt (for a coach, the coach's prompt) can appear in the same place, after a save or when you open the editor. One form of the warning says voice sessions will not connect. In that case a voice session is refused when a learner tries to start one, and text mode still works. The warning suggests rewording the scenario (for a coach, the coach's role or personality) or contacting support to request a review. The other form says the prompt was flagged for some voice types only, and suggests trying a different voice if sessions fail to connect.
Both messages sit inside the voice settings, which show only while the voice checkbox is on. When a save produces one of them, the editor also shows it as a notice wherever you are in the editor, titled "Voice mode was not updated" or "Voice mode needs attention". A safety warning that was already on record when you open the editor appears in the voice settings only.
Duplicates and version restores
Duplicating a simulation or coach that has voice set up also sets up voice for the copy. If that step fails, the copy is still created. For a simulation copy, voice is set up the next time you save it with voice on, or when its first voice session starts, whichever comes first. For a coach copy, voice is set up the next time you save it with voice on. Restoring an earlier version of a simulation or coach runs the same background setup when the restored version has voice on and a voice chosen.
AI tools over MCP
An AI tool connected through the MCP server can list the available voices, by id and name, with list_voices. It can then set up voice on a simulation with update_simulation: the voice, the session length, the transcript default and the voice turn limit. Turning voice on this way is refused unless the simulation ends up with a voice id, and the same session length limits apply. A voice id that is not in the voice list is refused with "Unknown voice. Choose a voice from the voice list.", so use an id returned by list_voices. If the voice list cannot be read at that moment, the voice is accepted without the check. The tool cannot change a coach's voice settings. Do that in the editor.
update_simulation also accepts a field named voice_input_enabled. It is off by default, the editor has no control for it, and it does not turn on voice sessions. Use realtime_voice_enabled.