IISc Bengaluru leads Project Vaani to make speech recognition inclusive
IISc Bengaluru is on a mission to make speech recognition systems more inclusive with Project Vaani.
Instead of just focusing on "standard" voices, the team is collecting and studying speech from people with neurological, cognitive, or motor conditions: voices that most tech usually misses.
Teaming up with ARTPARK and Google's Project Euphonia, they've already built a special dataset using recordings from 40 speakers in three Indian languages.
Project Vaani gathers 31,000+ hours nationwide
Project Vaani has gathered more than 31,000 hours of speech from 165 districts, covering 105 languages like Kannada, Tulu, and Marathi.
They use creative prompts, like showing images, to get authentic audio samples that reflect real-life diversity in age, gender, and background.
Plus, they've transcribed and shared more than 2,000 hours of this data for researchers to help make voice tech smarter (and fairer) for everyone in India.