Supported languages

The Instant Voice Cloning API supports synthesis in 12 Indian languages. Pass any of the following codes as language_code / target_language_code, or omit the field to detect the language from text.

Language codes

Codes follow the BCP-47 standard. If you need a language not listed here, contact your account manager.

Urdu (ur-IN) is not supported. An omitted or explicit Urdu request returns 400.

Omitting language_code

When you omit language_code (REST / HTTP stream) or target_language_code (WebSocket), the service detects the language from text:

• Unique Indic scripts (Tamil, Telugu, Kannada, Malayalam, Gujarati, Odia, Punjabi) map directly.
• Latin text is English if it is mostly dictionary English; romanized Indic such as Aapka order ship ho gaya hai. returns 400.
• Hindi vs Marathi (Devanagari) and Bengali vs Assamese share a script, so those two pairs use a small detector.
• Short Assamese without the letters ৰ or ৱ is often detected as Bengali. Pass as-IN for Assamese.
• Detection can be wrong on short or code-mixed text. Pass the code when pronunciation has to be exact.

On the WebSocket, the first detection is locked for the rest of the session. Send a new config with an explicit target_language_code to change language.

Cross-lingual support

Every supported language can be synthesized from a reference clip in any other supported language. The reference clip’s language does not need to match language_code. See Clone Across Languages for details.

Script support

The text field accepts native scripts for each language. Always prefer the native script for the most accurate pronunciation:

Code-mixed text

Code-mixed input (e.g. Hindi with embedded English words written in Roman script) is supported, and frequently appears in real-world Indian content. The model handles common mixing patterns natively:

from sarvamai import SarvamAI
client = SarvamAI(api_subscription_key="YOUR_SARVAM_API_KEY")
response = client.voice_cloning.convert(
    voice_id="svc-your-voice-id",
    text="मैं office जा रहा हूँ, meeting at 10 AM है।",
    language_code="hi-IN",
)

For best results with code-mixed text, set language_code to the dominant language (Hindi in the example above).