Sarvam AI rolls out AI voice model in 11 Indian languages
Sarvam AI's Bulbul lets developers fine‑tune the system to a specific speaker’s voice and is designed for low latency and cost-effective pricing.
Bengaluru-based AI startup has rolled out Bulbul v2, its new text-to-speech (TTS) model supporting 11 Indian languages, including Hindi, Tamil, Bengali, Kannada, and Gujarati.
The firm claims that the voices are delivered in authentic regional accents, unlike the typical robotic tones in other models.
“Natural, familiar speech in 11 Indian languages, with authentic accents that sound just like India. From lower latency and India-first pricing to wider language, Bulbul sets a new benchmark for Speech AI in India,” read the company’s LinkedIn post.
Unlike many global offerings, Bulbul lets developers fine‑tune the system to a specific speaker’s voice, and is designed for low latency and cost-effective pricing. This positions the product as a leaner alternative to global rivals such as ElevenLabs.
The launch builds on Bulbul v1, released in August, and introduced six preset voice personalities such as Amartya (expressive), Pavitra (dramatic), Amol (narrational), Maitreyee (informative), Arvind (conversational), and Meera (professional).
The release comes just days after the IndiaAI Mission selected Sarvam under its objective to build the country’s first sovereign large language model (LLM).
The startup will receive dedicated compute resources—including 4,096 Nvidia H100 GPUs for six months—to train the model from scratch.
Founded as a research lab by Vivek Raghavan and Pratyush Kumar in 2023, Sarvam has expanded into a full-stack artificial intelligence platform that provides generative AI solutions to governments, enterprises, and nonprofits.
It recently partnered with the Unique Identification Authority of India (UIDAI) to improve the user experience for Aadhaar services.
Edited by Kanishk Singh