Hindi text to speech for voice agents
Bolna voice agents speak Hindi on phone calls using Hindi voices from Sarvam, ElevenLabs, Cartesia, Soniox, Maya Research and Kalpa Labs. The same agent transcribes Hindi callers with Sarvam, Deepgram and six other speech-to-text providers.
How Hindi text to speech works in a voice agent
In a voice agent, Hindi text to speech is the step that turns the agent's written reply into spoken Hindi on the call. The reply comes from the LLM as text, and the synthesizer streams it as audio so the caller hears the first words before the whole sentence is ready.
On Bolna you pick the Hindi voice per language in the Languages tab. A multilingual agent can use Sarvam for Hindi and a different provider for English, and each language keeps its own voice, transcriber and prompt.
Hindi voices you can use on Bolna
Sarvam
bulbul:v4-flash, bulbul:v3
bulbul:v4-flash has 129 Hindi voices, 13 of them Hindi-English code-mixed. bulbul:v3 adds 16 voices that work in every Bulbul language.
ElevenLabs
eleven_v4_turbo, eleven_v3_conversational
eleven_v4_turbo and eleven_v3_conversational support 74 languages, Hindi among them. Each language can have its own pronunciation list.
Cartesia
sonic-3.6
sonic-3.6 covers 44 languages, including Hindi, and is the Cartesia model Bolna recommends for production.
Soniox
tts-rt-v2
One model speaks 63 languages, including Hindi, so a multilingual agent can keep the same synthesizer and change only the voice.
Maya Research
Maya 2 Native
Two voices, Ananya and Arjun, each speak all 11 Maya languages. Bolna switches Maya's language mid-call without reconnecting.
Kalpa Labs
kalpa-tts-1
Speaks English, Hindi and code-switched Hinglish from one model, with no language parameter to set.
Azure and AWS Polly are also Bolna voice providers, but their Bolna docs do not list Hindi, so check the voice list for your account before relying on them.
Hindi speech to text for voice agents
Hindi speech to text is the other half of the call. The transcriber turns the caller's Hindi into text in real time so the LLM can answer. Set the transcriber's language to hi explicitly, because auto-detection adds latency.
For agents serving Indian callers, the Bolna docs call Sarvam usually the strongest transcriber choice. Sarvam, Soniox and Gladia also handle Hinglish, where callers mix English words into Hindi sentences.
Sarvam
saaras:v4
Saaras models are trained on Indian speech and handle callers who mix English into Hindi. saaras:v4 is the recommended model and the only Sarvam model that reads keywords.
Deepgram
nova-2
Use nova-2 with language set to hi. Nova-3 is the better pick for English.
Gemini 3.5 Transcribe
gemini-3.5-transcribe-live
Google's real-time model. Bolna hints Hindi for the whole call, and the model still copes with the odd word from another language.
Soniox
stt-rt-v5
One multilingual model with code-switching. Pick Hindi as a single language, or the multilingual option for callers who mix in English.
Smallest AI
pulse
Pulse is built for sub-100ms first-token latency and transcribes calls that move between English and regional languages.
AssemblyAI
universal-3-6-pro
universal-3-6-pro code-switches across 32 languages.
Gladia
Gladia streaming
For Indian markets Bolna sets Gladia to recognize the primary language and English together, which suits Hinglish callers.
OpenAI
GPT Realtime Whisper
Pass hi as the language code.
Hindi voice features on Bolna
Set up a Hindi voice agent
Add Hindi to the agent
In the Agent Tab, click + Add Language and pick Hindi. Use the crown icon if Hindi should be the default language.
Write the Hindi prompt in Devanagari
Write नमस्ते, not Namaste. Hindi gets its own prompt tab, and switching instructions stay out of it because Bolna handles switching itself.
Choose a Hindi voice and transcriber
In the Languages tab, select Hindi and pick a voice, for example Sarvam bulbul:v4-flash, and a transcriber such as Sarvam saaras:v4. On Sarvam the voice language field is required, so set it to hi.
Start calls in Hindi
Pass agent_data.language as hi on the Call API, or add an agent_language column to a batch CSV. Calls without it start in the default language.
curl https://api.bolna.ai/call \
-H "Authorization: Bearer $BOLNA_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"agent_id": "<your-agent-id>",
"recipient_phone_number": "+919876543210",
"agent_data": { "language": "hi" }
}'Hindi and the other supported languages are covered in the multilingual docs, and ready-made templates are on the voice agents page.
Frequently asked questions
It depends on the voice you want and your latency budget, so test two or three on real calls. The Bolna docs recommend Sarvam for Indian languages, with 129 Hindi voices on bulbul:v4-flash. ElevenLabs eleven_v4_turbo, Cartesia sonic-3.6, Soniox, Maya and Kalpa Labs also speak Hindi.
For Indian callers, the Bolna docs name Sarvam saaras:v4 as usually the strongest choice. Deepgram nova-2, Gemini 3.5 Transcribe, Soniox, Smallest AI Pulse, AssemblyAI universal-3-6-pro, Gladia and OpenAI GPT Realtime Whisper also transcribe Hindi.
Yes. Sarvam has 13 Hindi-English code-mixed voices on bulbul:v4-flash, with _enhi_ in the voice id, and Kalpa Labs speaks English, Hindi and Hinglish from one model with no language setting. On the listening side, Sarvam, Soniox and Gladia transcribe code-mixed speech.
Yes. Add both languages and give each its own voice. When the caller says a full sentence in Hindi or asks for Hindi, the agent switches its transcriber, voice and replies together. A single English word inside a Hindi sentence does not cause a switch.
You can clone a voice from an audio sample on ElevenLabs or Cartesia in the Languages tab, and you set the sample language when you upload it. The cloned voice then appears in the voice picker for every agent in the workspace.
Pay-as-you-go starts at $0.06 per minute (about ₹5.52), with speech-to-text, the LLM and text-to-speech included when the agent uses the bundled models. Telephony and the platform fee are added per minute, and every new account gets $5 in free credits.
Hear your agent in Hindi
Walk through your Hindi call flow with the Bolna team, or add Hindi to an agent in Agent Studio and pick a voice.