The Voice & Avatar tab controls what people hear, and optionally see, when they talk to your agent. Every agent needs a voice. An agent only needs an avatar when you turn on video.
Conversation mode: voice or video avatar
At the top of the tab, Conversation mode asks “How should users experience this agent?” and gives you two cards to choose between:
- Voice only: audio with a dynamic voice visualization, no camera-style face.
- Video avatar: a lifelike video avatar for face-to-face chats.
Pick Voice only for a lightweight embed or a phone-style conversation where a face adds nothing. Pick Video avatar when seeing a face builds trust, such as a receptionist widget or a product walkthrough.
Selecting a card saves the mode right away, no separate save step. Switching modes only changes which sections appear below it: your saved voice and your saved avatar are kept independently, so going back and forth between modes doesn’t lose either one. Choosing Video avatar adds an Avatar section above the Voice section so you can pick one.
Choosing a voice from the catalog
The Voice section is always visible, because every agent speaks even when it isn’t showing a face. Its subheading says plainly what it’s for: “The voice your agent uses to speak with users.”
Until you pick one, the card reads “No voice selected” with a Choose a voice button. Once a voice is set, the same card shows its name, its provider (voices come from providers like ElevenLabs and OpenAI), and a short description, next to a Change voice button.
Either button opens Select a voice, a dialog listing every available voice with a live count (“N voices available”). From there you can:
- Search by name or description.
- Filter by provider, gender, or language, or clear every filter at once with Reset all filters.
- Preview a voice’s sample clip with the play button on its tile, where one exists.
- See at a glance which voices are marked Premium.
Pick a tile and select Apply selection to save it. The change applies immediately, with an inline confirmation once it’s saved. Browsing and previewing before committing matters here: providers and voices vary a lot in tone, accent, and naturalness, so a name alone rarely tells you whether a voice fits your brand.
Switching to a video avatar
Turn on Video avatar mode and an Avatar section appears, described as “The avatar that represents your agent during conversations.” It works the same way as the voice picker: “No avatar selected” with a Choose an avatar button until you pick one, then a card with the avatar’s thumbnail, name, and category, plus a Change avatar button.
Opening the picker takes you to Select an avatar, where you can filter by All, Standard, or Premium, and toggle Recommended to see only the avatars Speak AI recommends first. Select a tile and confirm with Apply selection.
An avatar changes what people see, not what your agent sounds like: it still speaks with whatever voice you set in the Voice section above, video just adds a face to that same voice.
When your agent runs on a Live model
Voice and avatar aren’t the only way this tab can look. On the agent’s Details tab, the Model field groups choices under Standard and Live (speech to speech), with this explanation: “Standard models listen, think and then speak. Live models hear the caller and speak directly.”
A Standard model is a pipeline of separate steps: transcribe what was said, reason about a reply, then synthesize it in the text-to-speech voice you picked above. A Live model collapses that into one step, so it needs a different kind of voice: one built into the model itself.
Pick a Live model on the Details tab and this tab’s Voice section changes to show only that model’s own voices, with no search or filters, since there’s a fixed, usually small, set to choose from. Your text-to-speech voice isn’t lost. It’s kept in reserve for whenever you switch back to a Standard model. A few settings that only make sense for a text-to-speech step, like pronunciation rules, are grayed out while a Live model is selected, each with a short note explaining why right underneath it.
If you’ve just picked a Live (or Standard) model on the Details tab but haven’t saved it yet, this tab shows a banner saying it’s previewing settings for that unsaved model, and that you need to save on the Details tab before it takes effect.
Related pages
- Create and manage Speak AI voice agents: the API reference for the Voices and Avatars catalogs this tab reads from.