Build agents

Voices & speech synthesis

Every voice agent has a speaking voice, and you pick it from the gallery. The voice you choose decides the synthesis behind it, so there is no separate engine to set. Play a few previews, find one that fits your brand, and select it.

The gallery lists every voice available to your workspace, each with a play button so you can hear it before choosing. Voices are grouped by family to make browsing easier, and each one is marked for where it can be used: the classic pipeline, realtime mode, or both. Some voices work in both, so the same voice can carry across a classic agent and a realtime one.

The gallery only shows voices from families your workspace has configured. If you were expecting a particular family and do not see it, ask an administrator to add it.

Voice families

Voices come in families, and each family has its own character. When a family also offers a choice of model, you make that choice in the bot's speech settings.

FamilyNotes
AuraClear, natural voices for the classic pipeline.
SonicExpressive voices with fine tuning for speed, volume, and emotion. The newer Sonic 3.5 model is selectable in the bot's speech settings. See Voice tuning.
ElevenLabsHighly expressive voices with style and speed controls, plus a low-latency mode worth using on phone agents.
AzureA broad set that includes high-definition voices and dedicated Urdu voices.

Importing more voices

Administrators can bring in additional voices from a provider's catalog. When a voice is imported, Hania records a short preview in the voice's own language automatically, so it is ready to audition in the gallery straight away with nothing to upload by hand.

Shaping how a voice sounds

Once a voice is chosen, its tuning card appears in the bot settings with only the controls that voice supports. Sonic voices expose speed, volume, emotion, and pronunciation; ElevenLabs voices expose style, speed, and a low-latency mode. Ranges, defaults, and examples are on the Voice tuning page.