Étude comparative des différentes méthodes d'estimation de la fréquence fondamentale des cris des nouveaux-nés

Abstract

Over the last thirty years, various methods have been developed to improve the estimation process of the fundamental frequency. Indeed, detection and time domain analysis of changes in the fundamental frequencies of the ewborn infants’ cries are necessary to obtain reliable information that helps pediatrician to make an accurate diagnosis. We present in this report a comparative study of the three fundamental frequency extraction methods and their performances on the baby’s cry signals. The aforementioned algorithms are AMDF «Average Magnitude Difference Function», ACF «Auto-Correlation Function» and frequency algorithm HPS «Harmonic Product Spectrum» which are implemented in the MATLAB environment. In addition, the modified SIFT algorithm «modified Simple Inverse Filtering Tracking» and the Praat software are used for comparison study only. We have studied several cry segments: phonic, high-pitch, hyperphonic, unvoiced and undefined. In order to evaluate the performance of different algorithms, we have considered five types of error; sound error, hyperphonic error dysphonic error, unvoiced error and undefined error. We also present the classification procedure of different cry segments depending on the error type. We have observed that the ACF is not capable to estimate the hyper-phonic frequency. Also, we have observed that the HPS estimate the fundamental phonic frequency with a higher precision than other method. In addition, the AMDF and the modified SIFT are not so precise for estimation of the fundamental phonic frequency. Because of this low precision, we have developed a new algorithm named CAMDA «Combined Average Magnitude Difference and Autocorrelation», which is a combination of the modified SIFT and the AMDF. The goal of this new algorithm is to correct the error proceeded during the detection of the phonic segments. The CAMDA has showed its capability to correct the problem given by the AMDF and the SIFT modified during the detection of the phonic segments.

Date	12 Dec 2013
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Chakib Tadj (Supervisor)