Language resources and methods for under-resourced speech systems

Abstract – Building Voice AI for under-resourced languages is still a challenging scenario, particularly while addressing multilingual societies like India. In this talk, we will discuss available language resources, more specifically open and free-to-use resources, for Speech Recognition (speech-to-text) and Speech Synthesis (text-to-speech). We will also discuss data augmentation and multilingual acoustic modelling strategies to handle the scarcity of data for under-resourced languages. A short introduction to zero-resource speech processing will follow and how it can impact the development of speech technology in low-resource languages.

Speaker biography - Shekhar Nayak received his Ph.D. degree from IIT Hyderabad in 2019, M.Tech degree in Signal Processing from Centre for Applied Research in Electronics (CARE), IIT Delhi in 2011, and B.E. degree in ECE from CSVTU, Bhilai in 2009. He worked at Hewlett Packard Pvt. Ltd., Bangalore as a Technology Consultant from 2011-2013, as a research assistant at Institute for Infocomm Research (I2R), Singapore on multilingual speech recognition for low-resource Indian languages in 2015, and as Senior Chief Engineer at Samsung R&D Institute, Bangalore. He is currently working as an Assistant Professor of Speech Technology at Campus Fryslan, University of Groningen, Netherlands. His research interests include automatic speech recognition and zero resource
speech processing with applications to language identification, spoken term discovery, speaking rate estimation, etc.

Difficulty of the talk - Intermediate

Take away -

• Resources and Methodologies for low resource speech processing.
• Understanding zero resource speech processing.