Set up a voice simulation (the Delivery Mode page)
A voice simulation is spoken end to end: the learner talks to the AI character out loud and hears them answer. You turn it on and shape it in one place — the Delivery Mode tab, under Simulation Display in the simulation editor. It is an ordinary authoring page that any creator can use.
Who can use it
Voice is on by default for every organization, including a new one, and for every individual (personal) account — you no longer have to ask Cesura to enable it. Any creator can switch a simulation to voice from this page.
An organization admin can switch voice off for the whole organization with Voice simulations on the organization’s Preferences page. The Delivery Mode tab then stays visible, with its controls greyed out and the line “Voice simulations are switched off for this organization. An organization admin can turn them back on in the organization’s preferences.” A simulation that was already set to voice keeps running as voice.
If your organization runs on its own AI provider, voice calls need a Google Cloud (Vertex AI) connection and run in your own Google Cloud project; an organization on a plain API key (OpenAI, Anthropic, Google AI Studio or OpenRouter) cannot run voice calls. See Run simulations on your own AI provider account (BYOK).
⚠ Because no administrator sits between a creator and this switch any more, the doubled run cost below is now something any creator can turn on. It is worth saying so to your team before the first invoice.
What a voice run costs
A voice run is charged as two runs instead of one. It is not a separate price: whatever your organization’s Workspace runs rate is, a voice run costs double it. The page says so next to the choice: “Enabling audio changes what a run costs: each run of this simulation is charged as 2 runs instead of one.”
When learners choose between text and voice, the balance is checked against the voice price before the run starts, so a learner who picks voice can always finish. The actual charge follows their choice: voice costs two runs, text costs one.
Voice also counts double against your organization’s fair-use allowance, so a member reaches that ceiling in half as many runs. See Pricing & Limits.
Turn it on
Under How this simulation runs, choose one of three options:
- Text chat — “The learner types. The transcript stays on screen throughout.”
- Voice call — “The learner talks with the character out loud. There is no text composer, and ending the call ends the run.”
- Let the learner choose — “The learner picks text or voice before the run starts, and stays in what they picked for the whole run.”
Whichever mode the run starts in holds for the whole run. There is no way to switch a voice run into a text chat halfway through, or the other way round. The voice settings below (“These apply only when the run is a voice call”) matter once voice is possible.
Choose a voice
Every voice in Select Voice comes with a pitch (higher, middle, lower-middle or lower) and a handful of words describing it — warm, gravelly, informative, breezy. Press the play button beside a voice to hear it speak a sample sentence before you commit.
If your organization uses its own AI provider, the play button is unavailable: “Voice previews are unavailable: your organization uses its own AI provider key (BYOK), and previews run on Cesura’s account.”
The list is filtered by your character’s gender: with a gender set on the Persona page you see only matching voices, and the page says how many. A persona set to neutral sees every voice.
Describe how the character sounds
Voice and delivery, directly under Select Voice, is an optional description of how the character sounds: “Used only in voice conversations. Describe how the character sounds at their everyday level and how their voice changes under pressure.” For example: “Speaks quietly and a bit flat. When pushed about home, gets shorter and faster, never loud.”
The field appears only when the simulation’s delivery includes voice. Learners never see it.
Keep the voice and the character’s gender matched
When you change the persona’s gender, Cesura switches a voice of the other gender to the default voice for the new gender. If the voice list cannot load at that moment, a message says the voice was not updated, and you choose one on the Delivery Mode tab yourself.
If a simulation’s voice does not match its character, the tab shows a warning such as “The selected voice, Kore, is a female voice, but the character is male.” Click Use followed by the suggested voice’s name to switch, or leave the voice as it is if the mismatch is deliberate. Nothing changes until you click. Simulations built by Professor Cesura get a voice that matches the character’s gender.
Age comes from the persona, not the voice
The character speaks at the age you gave them. Set Age on the Persona page and an eight-year-old will sound like a child rather than an adult reading a child’s lines — the pitch and delivery shift, while the voice keeps its identity. The preview reflects it too, so what you hear when choosing is what the learner gets.
The preview speaks your template’s language
Every language Cesura ships in — English, Hebrew, Arabic, French, German, Spanish, Italian, Portuguese, Russian and Chinese — is spoken by the preview and by the simulation itself. Choose a voice on a Hebrew template and the sample sentence comes back in Hebrew.
If a template is set to a language the voice service does not speak, the preview falls back to English and the play button says so: “Preview this voice in [language] — this template’s language has no preview voice. The simulation itself runs in the template’s language.” If you see that notice, check the language on the General page.
Choose the experience
Choose Audio Experience picks how the run looks on the learner’s screen. All three are full-screen, share the same controls and cost the same — they differ only in what the run is pretending to be.
| Experience | What the learner sees | Good for |
|---|---|---|
| Phone Call | A phone handset, with a ringing screen before the conversation starts | Anything genuinely conducted by phone — support calls, cold calls, triage |
| Portrait | The character’s face on a panel, with their words captioned underneath | Face-to-face conversations: interviews, feedback, difficult news |
| Video call | A video-meeting layout, the character on the stage and the learner as a participant who can turn their own camera on | Remote conversations, and any scenario where being on camera is part of the exercise |
If a Cesura administrator has set this simulation to Lifelike video avatar, it appears in the list too. Creators cannot choose it themselves, and a creator who switches away from it cannot switch back. Video avatars are not available to organizations that use their own AI provider. See Video avatar mode.
Immersive Avatar is retired. Simulations already set to it keep working and still show it in the list, but it is no longer offered for new ones — and, as the page warns, if you switch away from it you will not be able to switch back.
Give the character a face
Character photo decides the face the character wears in a voice call, and beside their messages in a text chat. If you choose no photo, the character wears the simulation’s own image (the cover image on the General Info tab). A simulation with no image shows a plain silhouette in a voice call, and the character’s initial beside their messages in a text chat. Cesura never picks a stock photo of someone else for you.
⚠ If your simulation image shows more than one person, the learner may read the wrong one as the character. Choose a photo or the silhouette instead.
Click Choose a photo to open Choose a character photo. From there you can:
- Pick a ready-made photo. The gallery is filtered to the gender and age you gave the persona, and you can search by name or description or widen the age filter.
- Upload your own photo. Say which gender and age it pictures so it lines up with the character.
- Pick a tile: Silhouette (a plain outline), Initials (the character’s initial), or Simulation image (offered when the simulation has an image).
Adjust framing crops a chosen or uploaded photo: drag to move, slide to zoom. What you see in the circle is what the learner sees.
How long a voice run can last
Every voice run is capped at 20 minutes and ends itself there. If the activity’s own Max Duration is shorter, the shorter one wins; if you set it longer, the field tells you it will not help: “Voice activities are capped at 20 minutes — a voice run ends there whatever you set here.” Plan the exercise to fit inside that.
Before you share it
Run it yourself once from a testing space. A voice simulation reads very differently out loud than it does on the page, and testing runs are free for organizations. See What is a Testing Space?
Show captions
Show captions decides whether the learner starts the run seeing the character’s words written on screen as they are spoken.
- On — “Learners see the character’s words as they are spoken, and can hide them mid-run.”
- Off — “Nothing is written on screen, but learners can turn captions on mid-run.”
Your setting seeds the learner’s, it never locks it. Captions are an accessibility affordance, so a template that starts with them off must not shut out a learner who needs them — there is a captions button in the run’s control bar either way. The learner’s choice lasts for that run only; nothing is stored.
A simulation saved before this setting existed keeps captions shown, so nothing changed underneath your existing templates.
The page shows you what you are building
The Delivery Mode tab is split: the settings on one side, and a miniature of the run they produce beside them. Each overlay mode wears its own frame, so the mode picker explains itself rather than asking you to imagine the result, and the caption line shows your character’s real opening sentence — so the captions switch has a visible consequence before you ever start a run.
In the run itself, the character speaks that opening line out loud as soon as the call starts, so the learner does not have to speak first.
Three things that used to be on this page are gone: the selected-voice alert (the preview button now sits under the picture), the portrait thumbnail, and the generated image filename beneath it. The cost warning moved next to the switch that causes the charge.