Indian Language Support
Vomyra is built for India-first voice experiences. This page covers which languages are supported, how to configure them in your assistant, how to handle Hinglish and code-switching, and ready-to-use prompt patterns for multilingual callers.
Supported languages
| Language | Best transcriber | TTS voices | Notes |
|---|---|---|---|
| Hindi (hi-IN) | Azure (recommended), Deepgram | Azure Neural, Vomyra AI | Best-supported Indian language. Handles code-switching well. |
| Tamil (ta-IN) | Azure | Azure Neural | Good accuracy on standard Tamil. Regional dialects may need custom keywords. |
| Telugu (te-IN) | Azure | Azure Neural | Solid performance on standard Telugu. |
| Kannada (kn-IN) | Azure | Azure Neural | Reasonable accuracy on standard Kannada. |
| Bengali (bn-IN) | Azure | Azure Neural | Covers West Bengal Bengali well. Bangladeshi dialect differs. |
| Marathi (mr-IN) | Azure | Azure Neural | Good coverage. Common in Mumbai and Pune deployments. |
| Gujarati (gu-IN) | Azure | Azure Neural | Supported. Test with business-specific terms. |
| English (en-IN) | Azure, Deepgram, Whisper | Multiple providers | Indian-accented English. Use en-IN, not en-US, for Indian callers. |
Configuring language
Language is configured per assistant under Assistants → Speech Input.
- 1Open your assistant.
- 2Navigate to Speech Input.
- 3Select your transcription provider.
- 4Choose the language locale.
- 5Save your configuration.
Recommended locales
hi-INen-INta-INte-INkn-INmr-INgu-INbn-INCan Myra handle multiple languages?
Yes. A single assistant can handle multilingual conversations, including Hindi-English code-switching (Hinglish). For best results:
Set a primary locale
Configure the primary language as hi-IN or en-IN depending on your audience. The locale hint guides the transcriber when the first few words are ambiguous.
Instruct Myra to follow the caller
Add a prompt instruction like: "Speak in the caller's preferred language. If they switch, switch with them."
Allow code-switching explicitly
Include a line in the system prompt that permits mixing. Without it, the model may resist replying in a different language than the prompt was written in.
Hinglish and code-switching
Code-switching — switching between two languages within a single conversation or sentence — is a default behaviour for many Indian callers, not an edge case. A caller might say:
The transcriber sees this as a mix. Configure it with hi-IN or en-IN depending on the predominant language, and tell the model to follow the caller's lead.
Prompt patterns for code-switching
Why: Without explicit permission, the model defaults to whichever language was used in the system prompt.
Why: Indian callers vary widely in formality. Matching register makes Myra feel less like an IVR.
Why: Translating 'Spice Garden' or 'FC Road' into Hindi sounds unnatural and confusing.
Why: A bilingual opener shows capability in both languages and lets the caller choose.
Regional accents and dialects
Regional accents can affect transcription accuracy. Always test with callers from the same region as your target audience. Local terms, business names, landmarks, and colloquial phrases are the most common sources of transcription errors.
Test with real callers from your target region
If you are building for a Madurai clinic, test with Madurai Tamil speakers — not just colleagues in a different city.
Add regional terms as custom keywords
Locality names, markets, landmarks, and business names are where transcription accuracy drops fastest. Custom keywords recover most of this.
Account for formality variation
A restaurant caller from Pune may use Marathi terms even when the call is nominally in Hindi. Allow for this in your system prompt.
Recommended configurations
Starting points for common India deployments.
- Speech Input
- Azure
- Locale
- hi-IN
- Voice
- Azure Neural or Vomyra AI
- Speech Input
- Azure
- Locale
- hi-IN
- Voice
- Azure Neural
- Prompt tip
- Follow the caller's language and allow Hindi-English mixing.
- Speech Input
- Azure or Deepgram
- Locale
- en-IN
- Voice
- Azure Neural
Common mistakes
Using en-US instead of en-IN for Indian English callers.
Translating business names, product names, or landmarks into Hindi.
Not allowing code-switching in the system prompt.
Testing only with internal team members who may not reflect real caller speech.
Forgetting to add custom keywords for local locations, landmarks, and product names.
Related pages
Speech-to-Text
Full transcriber configuration — language codes, denoising, and fallback.
Prompting Myra
Voice-specific prompting rules including language and code-switching instructions.
Custom Keywords
Boost transcription accuracy for brand names, dish names, and localities.
Indian Phone Numbers
TRAI, DLT, and Truecaller — the compliance side of Indian telephony.