An Ann Based Real Time System for Classification of Normal and Abnormal Cries of Pre-Term and Neonates 
Punith Kumar M B1, T Shreekanth2, Anupama M13, Sarsawath S4
1Punith Kumar M B, Associate Member, Institution of Engineers (AMIE), Member of IEEE Bangalore (Karnataka), India.
Manuscript received on 25 August 2019 | Revised Manuscript received on 01 September 2019 | Manuscript Published on 14 September 2019 | PP: 133-138 | Volume-8 Issue-5S3, July 2019 | Retrieval Number: E10330785S319/19©BEIESP | DOI: 10.35940/ijeat.E1033.0785S319
Open Access | Editorial and Publishing Policies | Cite | Mendeley | Indexing and Abstracting
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC-BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Abstract: Infants communicate with the external world through cry. Most of the problems in the infants can be explored through their cry within first year. Variations in cry can sometimes indicate the neurological disorders, genetic problems and many more. Classification of the infant cry as normal and abnormal at the early stages can reduce the course of action or any casualty. Hence this work proposes a computational approach for the early diagnosis of pre-term and neonates’ infant cry. The previous works include various algorithms for classification, however the novelty in this work can be attributed to processing only voiced part of the cry signal. The cry signal is first preprocessed by decomposing it into three levels using db13 wavelet in order to remove any noise that has been inherited during signal acquisition. This signal is further processed to extract only voiced part of the speech by identifying the endpoints through Zero Crossing Rate and Energy. Then the MFCC features are extracted, as this kind of signal envelop is best estimated eventually using these kind of features and are used to train feed forward neural networks based on back propagation algorithm. In order to train the network 100 normal and 100 abnormal samples were used. The database has been obtained from the neonatal ward of JSS Hospital, Mysuru. The algorithm has been tested on the test dataset consisting of 50 samples. The performance of the proposed method has been evaluated on only voiced part of the cry signal using the diagnostic test measures and the efficiency is found to be 98%as compared to 90% efficiency if the same procedure is applied on the entire cry signal.
Keywords: Back Propagation, DCT, FFT, MFCC, STFT, Neural Network, Pre-term, Neonates, Hamming Window, Wavelet.
Scope of the Article: Real-Time Information Systems