Voices - Waterr.ai

Set the voice once per session in session.update:

{
  "type": "session.update",
  "session": { "voice": "charles_uk", "instructions": { "type": "constant", "text": "..." } }
}

Voice ids are case- and separator-tolerant ("Charles UK" resolves to charles_uk). An unrecognized id falls back to the default voice, watercooler.

Preset catalog

Voice id	Gender	Character
`watercooler`	F	Warm, casual — the default
`anna`	F	Clear, neutral
`mary`	F	Bright, energetic
`eve`	F	Calm, measured
`jane`	F	Soft, conversational
`paul`	M	Neutral, professional
`george`	M	Deep, steady
`michael`	M	Friendly, upbeat
`charles_uk`	M	British accent

All presets are conversational seeds — chosen to sound like a person on a call, not a narrator. Expressive/narration voices were tested and cut for this release.

Custom voices

Enterprise preview partners can bring reference audio for a custom voice. Contact us — consent and licensing review are required for any custom reference.

Licensing

Preset voices derive from the CSTR VCTK Corpus (University of Edinburgh, CC BY 4.0), synthesized through Kyutai’s open TTS foundation (CC BY 4.0). If you build a product on this API, surface these attributions in your own docs or About page — details and ready-to-copy credits on the voice attributions page.