Infant Cry Recognition System using Autoregressive Model Coefficients

Volume 6 Issue 2 April - June 2018

Research Paper

Infant Cry Recognition System using Autoregressive Model Coefficients

S. R. Fatimah*, A. M. Aibinu**

* Department of Electrical Engineering, Nile University of Nigeria, Abuja, Nigeria.

** Department of Mechatronics Engineering, Federal University of Technology, Minna, Nigeria.

Fatimah, S. R., & Aibinu, A. M. (2018). Infant Cry Recognition System using Autoregressive Model Coefficients. i-manager's Journal on Digital Signal Processing, 6(2), 9-16. https://doi.org/10.26634/jdp.6.2.15591

Abstract

Understanding infants’ needs through crying is a skill acquired by health care givers as well as parents from training and experiences. However, errors may evolve due to variations in judgment and limitations on the human sensory system. Various approaches have been proposed to mimic the classical human based method which also tied results to system dominant errors. This work uses the Autoregressive (AR) model coefficient as features for recognizing infant cry. First, a dataset of infant cry consisting of Hunger, Pain and Normal cry was obtained. Each cry was framed and widowed with overlap to enable the processing of the rapidly changing cry signal. Then AR model coefficients (features) were extracted from the trained Artificial Neural Network (ANN). The extracted features were then used to train an Artificial Neural Network recognition system. The performance of this system was tested using three different activation functions, sampling frequencies and various threshold values. Results show the appropriateness of this new approach.