Languages
Set the language with the language query parameter when you connect:
Supported languages
Each value also accepts the India region, for example hi_IN or en-IN. Other regions are rejected.
Automatic detection
If you set language=auto or omit the parameter, the model detects the language from the audio, choosing from every supported language.
To detect from a smaller set, pass a comma-separated list. The model then chooses only from the languages in the list:
A single value sets the language instead of detecting it.
Transcripts do not include the detected language. If your application needs to know it, for example to choose a matching voice for the reply, set language explicitly.
Set language explicitly whenever you know it.
Output script
Transcripts are returned in each language’s native script. For example, Hindi is returned in Devanagari:
Mixed-language speech
When a speaker mixes languages, for example Hindi with English words (Hinglish), each word is written in the script of its own language. The transcript therefore contains mixed scripts: Devanagari for Hindi words and Latin for English words.
Numbers
By default, numbers are written as words, not digits. For example, fifty thousand spoken in Hindi is transcribed as पचास हज़ार, not 50000. Convert to digits in your application if you need them, for example before passing an amount to a payment tool, or turn on text formatting for Hindi and English (India).
Text formatting
Set text_formatting=true when you connect to format Hindi and English (India) transcripts. Numbers and similar terms are rewritten in written form, so they are easier to pass to your application.
Text formatting applies only to Hindi and English (India). For other languages, and when the parameter is omitted or false, transcripts are returned without formatting.
Changing language
The language is fixed for the life of the connection. To transcribe in a different language, open a new connection.