07-05-2025
Sarvam AI launches AI text-to-speech model with support for 11 Indian languages
Sarvam AI has launched a new text-to-speech AI model called Bulbul v1, available in 11 Indian languages including Hindi, Marathi, Punjabi, Oriya, Tamil, Bengali, Telugu, Kannada, Malayalam and Gujarati. The Bengaluru-based AI startup has said that the voices are in 'authentic accents' that don't sound robotic or rehearsed.
The model can also be trained in the user's voice while promising lower latency and cost as compared to rivals like Eleven Labs.
Users can also choose between six distinct personalities where each caters to a different communication need for a different industry. For example, Amartya is expressive and distinct, Pavitra is dramatic and engaging, Amol is narrational and mature, Maitreyee is engaging and informative while Arvind is conversational and articulate and Meera is professional and articulate.
A blog posted by Sarvam shared that their training data for the AI model combined high-quality, diverse audio from multiple speakers and languages. There were code-mixed inputs, proper nouns and abbreviations included as well as voices that had both professional and conversational tones for different use cases.
Last week, the government selected Sarvam as the first startup to build India's first indigenous AI large language model from among 67 applications. The startup will receive support from the government with regards to compute to build a foundational AI model from scratch.
