We learn from our alumni in this interaction series, often technically, sometimes semi-technically.

Nauman Dawalatabad CSAIL, MIT

Nauman Dawalatabad is a postdoctoral associate at the Computer Science and Artificial Intelligence Laboratory (CSAIL), Massachusetts Institute of Technology, USA. He obtained his Ph.D. (with institute research award) in Computer Science and Engineering from the Indian Institute of Technology Madras, India. During his PhD, he was also a visiting research student at Mila - Quebec AI Institute, Montreal, Canada, and a core developer of the open-source SpeechBrain toolkit. He was also a Lead Engineer at Samsung Research, Bangalore, working on on-device speech recognition. He serves as a reviewer for various speech and NLP conferences and journals. His current research interest includes robust speech recognition, speaker diarization, and multimodal processing of conversations.

In this talk, I plan to cover two different topics somewhat related to robust speech recognition. In the first part we will see an application of speech in healthcare. Speech can be used to identify various cognitive conditions. We propose a three stage pipeline approach to identify Alzheimer’s/Dementia from the speech interview conversations. In the second part, I will talk about the speech pseudo-label filtering method using uncertainty in the model. This is useful for unsupervised domain adaptation in automatic speech recognition systems. We will see the breakpoint of this algorithm and various approaches used to improve the filtering process to generate more reliable target domain pseudo-labels.

Organizers