Sarvam AI launches Bulbul V3, wins praise for the Indic text-to-speech model

Per Sarvam cofounder Pratyush Kumar, Bulbul V3 is the fifth of 14 planned launches. In a study, Bulbul V3 topped the charts for 8 kHz audio, setting what Kumar called a new benchmark for speech synthesis for voice agents, with listeners tagging real failure cases to test stability, and the model recording the lowest average error rates.