Skip to main content
GET /v1/voices returns the voice engines, the named voices and style presets each offers, and the languages an agent on it can hold a conversation in — the menus the dashboard’s voice picker renders.
curl
string
the value for ttsConfig.model.
string[]
values for ttsConfig.voice. only mulberry has named voices.
object[]
values for the agent’s language; disabled ones are listed but not yet selectable. defaultLanguage is used when language is unset.
object[]
curated voice styles for ttsConfig.description (mulberry). you may also write your own one-sentence description.
boolean
whether agents on this engine can take phone calls. the speech-to-speech engine cannot.

setting a voice

ttsConfig on an agent (or a version) is:
string
muga — our more expressive engine, Hinglish-first. mulberry — our faster engine with named voices and a styleable description. spider — speech to speech; browser and realtime calls only.
string
mulberry only. one of voices; defaults to the first.
string
mulberry only. a preset’s description, or your own sentence following the mulberry prompting guide.
sampling knobs, all optional. leave them alone unless you know why.
a muga agent needs only { "model": "muga" }. the voice and language are checked when a call starts: a voice name outside the roster is refused with 400 agent_speaker_not_supported, a language the engine does not offer with 400 agent_language_not_supported.

language

language is a property of the agent, not of the voice: it selects the script and register the agent speaks in and how speech is recognised. hinglish is Hindi in Roman script mixed with English, hindi is Devanagari. next: variables and tools.