Conversational ・ Friendly
News ・ Authoritative
Conversational ・ Consistent
Entertainment ・ Dynamic
Want to use this API?
New product launch. New compliance disclaimer. That means booking a studio, recording in Hindi, Tamil, Telugu, Bengali, and English, QA-testing every file, then deploying to production. With dynamic TTS, you update the text and it goes live instantly across all 11 languages.
Static IVR says 'Your account balance is...' and stops there. Dynamic TTS generates 'Suresh ji, aapke khaate mein 45,230 rupaye hain' in real-time. Name, balance, branch, transaction date, all personalized per caller.
IVR runs at 8kHz telephony quality, not studio-grade 24kHz. Bulbul V3 is optimized specifically for 8kHz output. What callers hear on an actual phone line sounds natural, because the model was trained for that bitrate.
Define your prompt templates
Map your existing IVR tree. Convert each pre-recorded prompt into a text template with placeholders: customer_name, balance, due_date, branch_name. Your existing IVR logic stays the same.
Detect or select caller language
Route to the correct language based on caller DTMF selection or automatic detection. One API endpoint handles all 11 Indian languages. No separate integration per language.
Generate audio in real-time
Fill templates with live customer data, send to Sarvam's API, get natural audio back in under 250ms. Latency low enough for real-time IVR playback without caller-perceptible delay.
Stream to your telephony stack
Standard WAV/PCM output at 8kHz. Works with Genesys, Avaya, Cisco UCCE, Twilio, Exotel, Ozonetel, and any SIP/VoIP system. No telephony migration required.
from sarvamai import SarvamAI client = SarvamAI( api_subscription_key="YOUR_KEY" ) # Dynamic IVR prompt with caller's data prompt = ( f"Namaste {customer_name} ji. " f"Aapke savings account mein " f"{balance} rupaye ka balance hai. " f"Pichhla transaction {last_txn_date} " f"ko {last_txn_amount} rupaye ka tha. " f"Aur jaankari ke liye ek dabayen." ) audio = client.text_to_speech.convert( text=prompt, target_language_code="hi-IN", model="bulbul:v3", speaker="shreya", pace=1.0, enable_preprocessing=True ) # 8kHz WAV for telephony playback with open("ivr_prompt.wav", "wb") as f: f.write(audio.audios[0])
“नमस्ते सुरेश जी। आपके सेविंग्स अकाउंट में 45,230 रुपये का बैलेंस है। पिछला ट्रांज़ैक्शन 12 मई को 2,500 रुपये का था। अकाउंट स्टेटमेंट के लिए 1 दबाएं।”
“Hello. Your policy number 8847291 is due for renewal on 15 June. Premium amount is Rs 12,400. Press 1 to renew, press 2 to speak with an agent.”
“உங்கள் மெயின் பேலன்ஸ் 147 ரூபாய். டேட்டா பேலன்ஸ் 2.5 ஜிபி மீதமுள்ளது. வேலிடிட்டி 18 ஜூன் வரை. ரீசார்ஜ் செய்ய 1 அழுத்தவும்.”
“మీ ఆర్డర్ నంబర్ 7721 డెలివరీ అవుతోంది. డెలివరీ కన్ఫర్మ్ చేయడానికి OTP 4839. దయచేసి ఈ కోడ్ డెలివరీ పార్ట్నర్కి చెప్పండి.”
End-to-end from API call to audio response, including network round-trip on India-hosted infrastructure.
Bulbul V3 is optimized for 8kHz output. IVR prompts sound natural on actual phone lines, not just in browser demos.
All data processed within India. No caller data stored post-processing. Audit logs for every API call.
Hindi, Tamil, Telugu, Bengali, Kannada, Malayalam, Marathi, Gujarati, Odia, Punjabi, and English from a single endpoint.
Starting at ₹30 per 10K characters. View pricing
Higher is better
ElevenLabs Flash V2.5
ElevenLabs V3 Alpha
Cartesia Sonic-3
Account balance inquiries, transaction alerts, loan status updates. Each prompt personalized with the caller's name, amounts, and dates, generated in their preferred language. No pre-recorded prompt library to maintain when products or compliance requirements change.
Policy renewals, claim status, premium reminders. Every call reads out policy numbers, amounts, and due dates correctly in the policyholder's language. Update prompts instantly when IRDAI guidelines change.
Balance inquiries, plan activations, complaint status. When tariff plans change, prompts update by changing text, not by re-recording across languages. Subscribers hear natural audio in their preferred language.
Aadhaar, DigiLocker, UMANG, and state-level portals need IVR in regional languages. Dynamic TTS serves citizens in 11 Indian languages from a single API, with all data processed within India.
RBI's data localization circular mandates that all payment system data must be stored and processed within India. For IVR systems that handle account balances, transaction details, and OTPs, this is not optional. Routing caller data to overseas TTS servers violates the circular. Sarvam processes all data within India, on Indian infrastructure, with no cross-border data transfer.
Latency matters for a different reason too. International API calls add 150-300ms of network latency before the TTS engine even starts processing. For IVR, where callers expect immediate audio response after pressing a key, this delay is perceptible and frustrating. Sarvam's India-hosted infrastructure delivers sub-250ms end-to-end latency, including network round-trip.
Enterprise IVR requires SOC 2 Type II certification, ISO 27001 compliance, no storage of caller data post-processing, audit logs for every API call, and SLA-backed uptime. Sarvam meets all of these. Sarvam does not store or train on any customer data passed through the API. Every request is processed and discarded.
Dynamic IVR prompts are step one. Samvaad takes your phone system from menu navigation to real conversations. Callers speak naturally, the agent understands intent, and calls resolve without transfers. Same infrastructure, same compliance, same 11 languages.
Natural, expressive voices for IVR systems in every major Indian language.
Modernize your IVR with a single API Powered by Bulbul V3
Modernize your IVR with a single API
Powered by Bulbul V3
