Talk details
Title: Vakyansh: An open source toolkit for developing speech technology in Indic languages
Speaker: Anirudh Gupta, Senior Research Engineer, Gan.ai, Noida
Date: 11 February 2023
Time: 6 PM (IST)
Abstract:
Vakyansh is an end-to-end toolkit for Speech Recognition in Indic languages. India is home to almost 121 languages and around 125 crore speakers. Yet most of them are low-resource languages in terms of data and pretrained models. Through Vakyansh, we introduce automatic data pipelines for data creation, model training, model evaluation and deployment. We create 14,000 hours of speech data in 23 Indic languages and train wav2vec 2.0 based pretrained models. These pretrained models are then finetuned to create state-of-the-art speech recognition models for 18 Indic languages, which are followed by language models and punctuation restoration models. We open source all these resources with a mission that this will inspire the speech community to develop speech-first applications using our ASR models in Indic languages.
Speaker’s bio
Anirudh is working as a Senior Research Engineer at gan.ai. He works primarily on problems related to speech and NLP. He is also the maintainer of Vakyansh, an open-source project which provides recipes for creating state-of-the-art Speech Recognition models for Indic Languages. Anirudh completed his BTech in Mechanical Engineering from Vellore Institute of Technology in 2013. He did his masters in applied mathematics and computational physics as an Erasmus Mundus scholar from several European universities. He wrote his master’s thesis at Freie Universitaet Berlin in 2015 in computational biophysics. He was also a MITACS Globalink scholar in 2012, for which he received a letter of appreciation from the Prime Minister of Canada for his research work. Before his job, he worked as a researcher at the department of theoretical physics at Freie Universitaet Berlin to understand hydration interaction between lipid membranes using large-scale molecular dynamics simulations.