Voices

viora speaks English and Swahili with natural voices, female or male. Each sentence is read by the voice for its own language, so an English sentence inside a Swahili answer still sounds right.

Swahili

Name	Voice	Microsoft neural name
`rehema` (default, `female`)	Female, Tanzania	sw-TZ-RehemaNeural
`daudi` (`male`)	Male, Tanzania	sw-TZ-DaudiNeural
`zuri`	Female, Kenya	sw-KE-ZuriNeural
`rafiki`	Male, Kenya	sw-KE-RafikiNeural

No Workers AI model speaks Swahili yet, so viora uses Microsoft's neural voices: through Azure AI Speech when a key is set (official, with a free tier of half a million characters a month), otherwise through the free voices of the Edge browser's Read Aloud service. That service isn't an official API: it is fine for personal use and pilots, but use Azure for a business. ElevenLabs also speaks Swahili with its eleven_v3 model.

English

Where	Female	Male
viora cloud (Deepgram Aura-2)	`luna` (default), `athena`, `asteria`, `hera`	`orion`, `arcas`, `apollo`, `zeus`
Your computer (Microsoft neural)	`ava` (default), `emma`, `sonia`, `libby`	`andrew`, `brian`, `ryan`, `thomas`

Everywhere, female and male (or mwanamke and mwanaume) pick that language's default voice.

Choosing a voice

Website and apps: speaker and speaker_sw in the assistant profile, or per conversation: viora.live({ voice: "orion", voiceSw: "daudi" }) in JavaScript, LiveSession(..., voice="orion", voice_sw="daudi") in Python. POST /v1/sites/:site/speak takes {text, language, voice}.
Your computer: viora setup asks female or male, then the names. Or set voice_en and voice_sw in desktop.yaml. viora voices lists them, and viora say "Habari!" plays one.

Other voice providers on your computer

`voice`	Notes
`edge`	Default. Free, no key, natural English and Swahili. One connection is kept open, so sentences start in under a second.
`elevenlabs`	`ELEVENLABS_API_KEY`. English uses the fast Flash model and Swahili uses `eleven_v3`. The free plan (about 10 minutes a month) is for non-commercial use with attribution; paid plans allow commercial use. Use any voice id from the ElevenLabs voice library.
`azure`	`AZURE_SPEECH_KEY` and `AZURE_SPEECH_REGION`. The same neural voices as `edge`, official.
`cloudflare`	Aura-2 for English, Microsoft neural for Swahili.
`openai`	`OPENAI_API_KEY`.
`system`	espeak-ng, macOS `say` or Windows SAPI. Robotic but offline.

Self-hosted viora servers choose voices with VIORA_SPEECH (English) and VIORA_SPEECH_SW (Swahili: auto, edge, azure, elevenlabs, openai or off).

MCP Next: Use from software