An AI that can speak, feel, sing: IIT-BHU graduate makes model that’s more ‘human than machine’

Jaipur-based entrepreneur Sparsh Agrawal, 25, has introduced one of the first speech-to-speech foundational AI models capable of singing, whispering, pausing, and responding with emotional intelligence — all built without support from big tech or venture capital funding.

Developed under his startup Pixa AI, the model, named Luna AI, processes audio directly to produce lifelike speech rather than converting it into text and back. This approach enables faster, more natural, and emotionally aware conversations.

According to Agrawal, Luna’s design…