About

Me inside the Flower Dome, Gardens by the Bay, Singapore in 2018

My name is Yash Shah, and I’m a final year undergraduate student at IIT Bombay pursuing B.Tech (with Honors) in Computer Science and Engineering and Minor in Applied Statistics and Informatics. I am also a member of CSALT Lab at IIT Bombay, which is headed by Prof. Preethi Jyothi. I have spent my summers at Verisk AI (with Dr. Maneesh Singh) working on accent adaptation for ASR during May-July 2019, and at SINAPSE Institute (now N.1 Labs), National University of Singapore (with Prof. Alcimar Soares) working on 3D shape recognition using tactile sensing during May-July 2018.

I am broadly interested in learning in settings where labeled data is scarce but unlabeled data is abundant, with applications in the domains of speech recognition and NLP. In the past, I have worked on leveraging large amounts of unlabeled data to identify word segments for low-labeled-resource languages in an unsupervised fashion, and building language models cognizant of this information. Currently, as a part of my undergraduate thesis, I am exploring adversarial and semi-supervised methods for accent adapatation in end-to-end ASR (and seq2seq models in general) to achieve better error rates for multi-accent corpora having single-accent transcriptions.

I love exploring other domains/problems for inspiration and research exposure, and have forayed into topics such as latent variable models for human motion synthesis (with Prof. Arjun Jain), loss functions for student-teacher learning using statistical n-gram language models (with Prof. Preethi Jyothi). I also occasionally make short research excursions to other interesting topics in the form of course projects; you can learn more about them here.