What is vagdhenu?

Vagdhenu is a text to speech system built for one very specific purpose: turning written Sanskrit verses into chanted audio that sounds like a real person reciting them, complete with the rhythm and melody that traditional chanting uses, rather than a flat robotic read-aloud voice. The project comes from a professor at the Indian Institute of Science in Bengaluru, and the README reports it has already been used to produce many hours of finished audio and video content, including a full chant series and an audio version of a lengthy classical text with over sixteen thousand verses. Under the hood, the system combines a few existing AI models rather than building everything from scratch. It uses a speech generation model to turn text into sound, paired with a separate model that converts that into higher quality audio waveforms. The README describes real engineering effort spent on getting Sanskrit specifically to sound right, including routing the text through a different script internally to avoid pronunciation quirks, and detecting the traditional meter of each verse so the timing of the chant matches classical conventions. To run it yourself, you need a computer with an NVIDIA graphics card and a specific version of Python, since the AI models require GPU acceleration. A setup script installs the needed software and downloads the pretrained model files automatically. Once set up, you feed it a verse in Devanagari script along with its meter, and it renders an audio file of the chant. There is also a project page with a live demo if you want to hear it before installing anything. The code itself is released under the Apache 2.0 license, though it depends on other AI models and tools that carry their own separate licenses. The README is explicit that this is meant for respectful use in study, recitation, and accessibility, and asks users not to use the synthesized voice to impersonate anyone.