Languages

Set language on the connection or the request. It defaults to hi-IN+en-IN.

Available

language	Name
hi-IN+en-IN	Hinglish, Hindi and English mixed within an utterance
hi-IN	Hindi (India)
en-IN	English (India)

In a query string the + must be percent encoded as %2B, or it is decoded as a space:

?language=hi-IN%2Ben-IN

Choosing

hi-IN+en-IN is the one to use for how people talk at work. Code switching happens inside the sentence, not between sentences, and this setting expects that:

Invoice share kar diya hai, meeting 5 baje start hoti hai.

Hindi words are romanised, English words are spelled as English. Numbers, product names and the vocabulary that never gets translated stay in the script you would type them in.

hi-IN for speech that is Hindi throughout, transcribed in Devanagari.

en-IN for speech that is English throughout, with Indian pronunciation, names and place names.

If in doubt, hi-IN+en-IN is the safer choice for the middle of the country’s working day. A single language setting on mixed speech is what produces the wrong kind of wrong: a confident transcript with the other language quietly dropped.

One language per session

The language is fixed for the life of a WebSocket session. To change it, open a new connection. Batch takes it per request.

In training

These are being trained. They are not accepted by the API yet, and asking for one returns unsupported_language with a 422. This list is what is being worked on, not a delivery schedule.

• Tamil
• Telugu
• Marathi
• Bengali
• Kannada
• Malayalam
• Gujarati
• Punjabi
• Tanglish, Tamil and English mixed within an utterance
• Benglish, Bengali and English mixed within an utterance

If one of these is what you need, tell us at [email protected], or book a call. Knowing who is waiting on which language changes the order they land in.

Unsupported values

{
  "error": {
    "code": "unsupported_language",
    "message": "Language 'ta-IN' is not available yet.",
    "request_id": "req_01JD4Z2Q8W6M"
  }
}

See Errors.