Languages & voices

An agent has one primary language and can understand additional ones.

Language#

• Primary language drives how the agent listens and, by default, the language of call summaries.
• Secondary languages are additional languages the agent will recognize if a caller switches.
• To keep recognition accurate (so a Greek caller is not misheard), the platform locks recognition to the languages you declare.

Time zone#

The Time zone field (next to the languages) sets the clock the agent uses whenever it mentions dates and times. It lists every time zone with its current offset from UTC; type part of a city or region, such as athens or new york, to find one. New agents start on Europe/Athens, and an agent with no time zone set also runs on Europe/Athens (the field shows "Europe/Athens (default)").

Voice#

Choose from the voice catalog: each voice shows its name and, under it, whether it is a male or female voice, the language(s) it fits and a short description. Greek voices come first, male voices and then female ones, followed by all other voices; typing searches names, descriptions, languages and gender. You can play a sample before choosing.

Voice quality#

Voice quality picks the voice model behind the voice. New agents start on Lifelike, real-time (Eleven v4 Turbo) with expressive delivery on; agents created before keep the quality they had, and Automatic lets the platform choose. Lifelike, real-time (Eleven v4 Turbo) is the newest and most natural voice for live calls: it starts speaking as quickly as Balanced and covers 90+ languages, Greek included. Lifelike, studio quality (Eleven v4) sounds best of all but takes a little longer to start speaking, so it suits agents where quality matters more than speed. Hear selected voice plays it at the quality you picked.

The Lifelike voices have two tuning sliders, Stability (how consistent the delivery stays from one line to the next) and Similarity (how closely it stays to the original voice). They have no Speed control: the slider is greyed out, its value is kept for the other qualities, and pacing follows punctuation instead (an ellipsis adds a pause, capitals add emphasis).

Expressive delivery (audio tags)#

On Lifelike and Most expressive a switch Expressive delivery (audio tags) appears, turned on when you pick one of them. The agent may then colour a reply with a short cue in brackets such as [laughs], [sighs], [whispers], [curious] or [excited], which the voice performs instead of reading out. It uses them rarely and never on prices, dates or instructions, and they never appear in the transcript, captions or webhooks. You can also write a cue into the first message. On any other quality cues are removed before speech, so a caller never hears the word "laughs".

Call summaries in multiple languages#

If you serve customers in more than one language, you can request the post-call summary in each, useful when your team reads in one language and your customers speak another.