Seminar by Prof. Wonyong Sung, Bio-Signal Processing - Standing on the Shoulders of Speech Recognition Technology (12 Dec 2023)
You are cordially invited to the following Seminar hosted by
Professor Gan Woon Seng (EEE)
Date and Time:
12 Dec 2023 (Tue) | 10:00am – 12:00pm
Venue:
NTU EEE Executive Seminar Room (S2.2-B2-53)
Title:
Bio-Signal Processing - Standing on the Shoulders of Speech Recognition Technology
Speaker:
Prof. Wonyong Sung
Abstract:
Over the past decade, we have witnessed remarkable strides in automatic speech recognition technology. Speech recognition technology has been transformed from statistical models to sophisticated deep neural network-based end-to-end algorithms. Today, our attention turns to bio-signal processing, a field with vast potential for enhancing human health. Several applications within bio-signal processing, such as ECG, PPG, and EEG can be seen as sequential classification algorithms, with automatic speech recognition technology being one of the most well-studied examples.
This presentation shows the possibility of advancing bio-signal processing by harnessing automatic speech recognition technology. To illustrate this concept, I will provide a case study focused on sleep stage classification using electrocardiogram (ECG) signals. Traditionally, ECG-based sleep classification relied heavily on manually selected features. In contrast, our approach leverages deep neural networks inspired by the speech recognition models. Specifically, we introduce a sleep model, which resembles the language models used in speech recognition. Furthermore, I will share preliminary results for CTC-loss-based deep neural network training, including its implications for data collection.
Biography: