Stimulated Deep Learning for Robust ASR (2015-2016) Goal of this project is to develop a stimulated deep learning framework that allows contextual information, such as the speaker, background noise and channel information, to be explicitly incorporated into the multi-style training process. In particular, algorithms and feature representation are being explored to jointly learn the different contextual information for adapting the DNN model parameters.
Speech-based Conversation System for Indian Languages (2009-2012) Goal of this project is to develop a system to facilitate illiterate/semi-literate people to access information about the agricultural products sold in the markets across Andhra Pradesh state, India. In this project, a framework for a telephone-based spoken dialogue system is developed which includes building signal processing components for speech in noisy environments, dialogue state tracking, caller personalisation and logging, and integrating the ASR and the TTS components.
Digital Library of India (2007-2008) Goal of the project is to capture books in digital format and make it available on the Internet. Worked as a research engineer in developing modules for handling meta data of the books, search and retrieval, and system monitoring. This work has also was extended in developing a search interface for searching local news paper articles.