Voices & Languages

A voice is a person, recorded across one or more languages. Each voice has a human-readable name and a language code. The same person — say, Archana — can speak in several languages (Archana in ta-IN, Archana in hi-IN).

Supported languages

Code	Language
`en-IN`	Indian English
`hi-IN`	Hindi
`ta-IN`	Tamil
`te-IN`	Telugu
`kn-IN`	Kannada
`mr-IN`	Marathi
`gu-IN`	Gujarati
`bn-IN`	Bengali

Model

Every synthesis request must include a model_id. There is a single model:

Model	Description
`raaga-v1`	Production text-to-speech model.

Voices

Every voice speaks all eight supported languages, so you can pair any voice with any language code. Pass the voice's name (or a vc_ custom voice ID) as the voice parameter, with a matching language, in synthesis requests.

Custom vc_ voices come from voice cloning, which is included on the Growth plan only.

Voice	Gender	Languages
Aakash	Male	All eight
Amit	Male	All eight
Anand	Male	All eight
Archana	Female	All eight
Bhamini	Female	All eight
Jay	Male	All eight
Kavya	Female	All eight
Kiran	Male	All eight
Meena	Female	All eight
Mohan	Male	All eight
Neha	Female	All eight
Nisha	Female	All eight
Priya	Female	All eight
Vidya	Female	All eight
Vikram	Male	All eight
Yash	Male	All eight

All sixteen public voices support every language, so any voice/language combination is valid. GET /v1/voices is always the source of truth for the current roster.

A voice's fluency can vary from one language to another. The same voice may sound more natural in some languages than others, so it's worth sampling a few voice + language pairs to find the best fit for your use case.

Listing voices

Use GET /v1/voices to discover available voices. Each entry's voice_name value is what you pass as the voice parameter in synthesis requests. Results are grouped by language by default.

12client.voices.list(group_by="language")
client.voices.list(group_by="voice")

Grouped by language

Each key is a language code, and the value is the list of voices available for it:

1234567891011121314{
  "ta-IN": [
    { "voice_name": "Archana", "gender": "female" },
    { "voice_name": "Kavya", "gender": "female" },
    { "voice_name": "Kiran", "gender": "male" },
    { "voice_name": "Aakash", "gender": "male" }
  ],
  "hi-IN": [
    { "voice_name": "Priya", "gender": "female" },
    { "voice_name": "Vidya", "gender": "female" },
    { "voice_name": "Mohan", "gender": "male" },
    { "voice_name": "Vikram", "gender": "male" }
  ]
}

Grouped by voice

Each key is a person, with their gender and the languages they support:

12345678910{
  "Archana": {
    "gender": "female",
    "languages": ["bn-IN", "en-IN", "gu-IN", "hi-IN", "kn-IN", "mr-IN", "ta-IN", "te-IN"]
  },
  "Priya": {
    "gender": "female",
    "languages": ["bn-IN", "en-IN", "gu-IN", "hi-IN", "kn-IN", "mr-IN", "ta-IN", "te-IN"]
  }
}

Selecting a voice in synthesis

For HTTP requests, send model_id, voice, and language together. The server validates the voice + language pair against the active voice catalog.

The parameter is now voice. The older voice_name parameter is deprecated and will be removed after a migration window — it still works for preset voices during the transition. See POST /v1/tts/generate for details.

If the voice + language combination does not exist, the request fails with 422 and the code voice_language_not_found. Always source valid pairs from GET /v1/voices.

For WebSocket sessions, these fields are set once in a config message rather than on each text message. See the realtime guide.

Ask AI about this page