Text to speech

Routed across ElevenLabs, LMNT, and Deepgram Aura. Voice aliases (female, male) map to a comparable native voice on each provider, so the same request works wherever it lands. Billing is per input character.

Format	Providers
mp3	ElevenLabs, LMNT, Deepgram Aura
wav	LMNT, Deepgram Aura (ElevenLabs is skipped automatically)

Request

POST /v1/tools/speech
{
  "text": "Your build finished — three tests fixed, zero regressions.",
  "voice": "female",
  "format": "mp3",
  "routing": { "sort": "quality" }
}

Result

Audio comes back base64-encoded, with the provider-native voice the alias resolved to:

{
  "result": {
    "audio_base64": "//uQxAAB…",
    "format": "mp3",
    "voice": "Rachel",
    "characters": 58
  }
}