Signal Processing guided Machine Learning

The buzzword for building applications of practical relevance is “big data” today. Big data is required primarily to account for the variability in natural signals. Large amounts of data enable neural networks to learn the underlying statistical characteristics of signals. Such models are massive but simple to build since they use the raw signal without preprocessing. A consequence is that they consume significant compute resources for training, and testing.

While deep learning has revolutionized machine learning, in this talk we focus on the use of signal processing to preprocess or mine existing data, so that accurate or “smart data” is presented to machine learning models. Knowledge-based signal processing is primarily used to identify domain-specific events. Owing to significant variability in events, they are best characterised by machine learning models. We draw examples from speech, music, and brain signals to show that signal processing and machine learning should work in tandem.

About the Speaker:
Hema A. Murthy received a bachelor’s degree from Osmania University, Hyderabad, India, in 1980, a master’s degree from McMaster University, Hamilton, Canada, in 1986, and a Ph.D. degree from the Indian Institute of Technology (IIT) Madras, Chennai, India, in 1992. She is currently a Professor at the Department of Computer Science and Engineering, IIT Madras. Her research interests include speech processing, music information retrieval, and computational brain research, where her primary goal is to use machine learning and signal processing in tandem. She is a Fellow of ISCA (International Speech Communication Association), 2022 and a Fellow of INAE (Indian National Academy of Engineering), 2017.   She received the IBM faculty award in 2006 for work on speech synthesis in Indian languages.   She has been involved in various projects funded by the Government of India (GoI). She was the Chief Investigator of a consortium project to build text-to-speech (TTS) systems in 13 Indian languages. She is currently leading the Speech Technology effort for the Natural Language Translation Mission funded by GoI.