Kalpa Labs text-to-speech for Bolna AI voice agents

What is Kalpa Labs used for in a Bolna voice agent?

Kalpa Labs is a text-to-speech provider whose kalpa-tts-1 model speaks English, Hindi and code-switched Hinglish in the same utterance, with no language parameter to set. In Bolna it turns your agent's replies into speech over a persistent WebSocket, so audio starts while the LLM reply is still arriving.

Pick it when your callers mix Hindi and English and you want one voice to handle both.

How do I set up Kalpa Labs with Bolna?

1. Open your agent in Agent Studio and go to the Languages tab, or build the agent through the Create Agent API.
2. Set the synthesizer provider to kalpa.
3. Set the model to kalpa-tts-1, the default.
4. Set voice to a name from Kalpa's catalog. The default is Arjun. You can pass an explicit voice_id instead.
5. Turn on stream: true and keep buffer_size at the default of 400 characters.
6. Optional: tune chunk_length_schedule. The default [50, 80, 120, 150] suits most agents.

For every setting, see the Kalpa Labs setup guide in the Bolna docs.

Which Kalpa Labs models does Bolna support?

Model	Languages
kalpa-tts-1	English and Hindi, including code-switched Hinglish. Default. No language parameter
kalpa-tts-multilingual-beta-v0.1	Superseded by kalpa-tts-1, which Kalpa now serves under this name, so existing agents need no change
kalpa-tts-beta-v0.1	English only

Optional tuning knobs are temperature, acoustic_temperature, max_new_tokens and audio_quality. Bolna only sends the ones you set, and anything left out uses Kalpa's default for that model.

What can you build with Kalpa Labs and Bolna?

Hinglish sales and collections calls. Callers in India often switch between Hindi and English mid-sentence. Kalpa speaks both from one model, and you can pair it with Soniox or Sarvam transcription for code-switched input.

Faster first audio on long replies. Kalpa segments text on its side as it streams in, so the first sentence starts playing before the full reply is done.

Clean barge-in. When a caller interrupts, Bolna cancels the open response and stops the audio, so the agent is ready for the next turn.

One voice for phone and web. The same agent sounds right on a phone call and in a browser widget, since Bolna converts Kalpa's 24 kHz audio for telephony and leaves it native for web.