Navana.ai Launches Bodhi TTS at 60% Lower Price Than Sarvam

Navana.ai has launched Bodhi TTS, a text-to-speech model for Indian voice AI applications, with support for 10 Indian languages and more than 50 Indian voices. 

The model is priced at ₹12 per 10,000 characters, which the company says is about one-eighth the price of ElevenLabs and 60% lower than Sarvam AI’s Bulbul v3.

Bodhi TTS supports Hindi, Telugu, Tamil, Marathi, Bengali, Odia, Malayalam, Kannada, English and Gujarati. Navana.ai said the model begins generating audio in under 100 milliseconds and can run inside an enterprise’s own data centre.

The company said Bodhi TTS is built and trained in India, with Navana.ai owning the speech stack end-to-end. It is designed to keep customer audio, transcripts and personal data within an enterprise’s security environment, without routing information through a third-party model provider.

At ₹12 per 10,000 characters, Bodhi TTS is priced below competing services. ElevenLabs lists its Multilingual and v3 models at $0.10 per 1,000 characters, which is about ₹95 per 10,000 characters, while Sarvam AI lists Bulbul v3 at ₹30 per 10,000 characters, according to Navana.ai.

“India runs on voice, and most of it is not in English. Until now, a bank that wanted a voice agent in Marathi or Tamil had to choose between a global model that mispronounced everyday words and a price that only worked for a pilot,” said Raoul Nanavati, Co-Founder and CEO of Navana.ai.

“We built Bodhi from the ground up to work to represent every accent, so in a country where every region speaks a little differently, businesses get fine-grained control from day one to make their bots sound truly Indian. And we priced it for the call volumes Indian businesses actually run, with everything in the box. One price, every voice, every feature,” he added.

Bodhi TTS is built to handle Indian-specific text formats. Navana.ai said it can read ₹1,00,000 as “one lakh rupees” and correctly vocalise PAN numbers, phone numbers, dates, EMIs, scheme names and account numbers.

The model also supports pronunciation control through dictionaries. Enterprises can upload product names, branch names and customer names, while a language model can mark up specific names inline for Bodhi TTS to pronounce according to the supplied instructions. Navana.ai said pronunciation issues identified during a live call can be corrected without retraining the model.

The company said its 50-plus voices are tuned for conversational phone interactions. New voices can be created from a few seconds of reference audio without additional training. Voice cloning is available across all 10 languages, with reference audio cached so that custom voices do not add latency during calls.

“Everyone assumes quality in speech scales with model size. For Indian speech it does not,” said Jai Nanavati, Co-Founder of Navana.ai.

“We built Bodhi TTS about five times smaller than the nearest comparable model. A small, fast model is one a bank can run on a mid-tier GPU it already owns, inside its own data centre. That is why we can price Bodhi the way we have, and why no customer data ever has to leave the building,” he added.

Navana.ai said its voice agents currently support more than ₹1,000 crore in monthly loan disbursals across its banking and lending customers. The company has processed more than 100 million voice AI minutes for BFSI customers, including Bajaj Finserv, Protean, Ujjivan Small Finance Bank, Jana Small Finance Bank and Grihum Housing Finance.

Alongside Bodhi TTS, Navana.ai has opened Bodhi STT, its speech recognition model supporting 10 Indian languages and more than 40 dialects, to customers by invitation. The company said benchmark results for the model have been published on Hugging Face.

Bodhi TTS is available through Navana.ai’s platform, with ₹1,000 in free credits for new users. Banks, insurers and lenders can also apply for early access to both Bodhi TTS and Bodhi STT. The first 100 approved teams will receive 100 hours of usage credits valid for 90 days.