Built on Sarvam's speech stack — tuned for how India actually speaks, including code-mixed Hinglish and Indian-accented English.
Full STT and TTS coverage with natural prosody for India's most-spoken language.
Sarvam handles natural Hindi-English switching mid-sentence — the way Indian callers actually talk, not textbook Hindi.
Native speech recognition and synthesis for West Bengal and Bangladesh-facing callers.
End-to-end Tamil voice calls across transcription, understanding, and spoken replies.
Telugu speech understanding and natural synthesized voices for the Telugu-speaking market.
Marathi STT and TTS tuned to how the language is spoken across Maharashtra.
Gujarati-language voice agents for commerce, support, and appointment workflows.
Kannada speech recognition and synthesis for Karnataka-first deployments.
Malayalam voice support for Kerala-facing call flows and customer service lines.
Punjabi-language calling with Indian-accented voices that sound local.
Odia speech support so agents can serve Odisha callers in their own language.
Indian-accented English voices and recognition tuned to Indian names, places, and number formats.
40+ global languages through your configured STT and TTS providers — including:
The stack is configurable per agent — pick the STT, LLM, TTS, and telephony providers that fit your language mix, region, and budget.
| Layer | Providers | Notes |
|---|---|---|
| STT (speech-to-text) | Sarvam saaras:v3, Deepgram nova-3, Azure Speech, Google STT | Sarvam for Indian languages and Hinglish; Deepgram/Azure/Google for 40+ global languages. |
| TTS (text-to-speech) | Sarvam bulbul:v2, ElevenLabs, Cartesia | Streaming synthesis — audio starts as the LLM streams tokens, not after the full reply. |
| LLM | Streaming models (configurable per agent) | Token-by-token streaming so the TTS stage never waits for a complete response. |
| Telephony | Exotel, Vobiz, Twilio, Plivo, Telnyx, Vonage, Cloudonix, Asterisk ARI | 8 providers with health probing and failover; Indian routes via Exotel and Vobiz. |
How the layers stream into each other determines turn latency — see the breakdown on Benchmarks.
Real conversations don't stay in one language. A caller might start in Hindi, drop into English for product names, then switch back — and a dedicated language-switch processor in the pipeline tracks that automatically.
The agent follows the caller's language instead of forcing them to pick one upfront. For Hinglish speakers, code-mixed utterances are handled natively by the Sarvam stack rather than mistranscribed as pure Hindi or pure English.
Caller speaks Hindi
Agent detects Hindi and responds in Hindi.
Caller switches to English
Language switch processor detects the change mid-conversation.
Agent follows automatically
Responses switch languages without the caller repeating themselves or pressing a key.
No language packs, no tier locks. Whether you're on the Pilot plan or Enterprise, the full Indian and global language matrix is available — you choose providers per agent, not per plan.