Development of a robust, low-complexity infant cry classification system

This proposal aims to develop a robust, low complexity infant cry deep learning classification based on various babies’ responses to physiological needs such as hunger or to discomfort and pain. The significance of this research lies in its potential to enhance early detection of needs and moods in newborns, contributing to improved infant care, early intervention and augmented infant-parent communication.
The novelty of the proposed research lies in applying methods to improve performance when small datasets are available and to reduce complexity in deep learning classification systems for infant cries. In this project we will also select and benchmark multiple datasets for training and testing, evaluate and compare different methodologies for feature selection and scaling, and implement a model suitable for real-time applications.
Future research will be extended in subsequent years to include the detection of additional emotional responses, and include babies who are diagnosed with medical conditions, but the emphasis of this one-year research proposal is on machine learning, engineering, and computer science aspects, and it will be performed using existing public domain datasets.

Faculty Supervisor:

Martin Bouchard;Hilmi Dajani;Helly Goez

Student:

Partner:

CRYNOSTICS

Discipline:

Engineering

Sector:

Professional, scientific and technical services

University:

University of Ottawa

Program: