Skip to Main content Skip to Navigation

Détection et Reconnaissance des Sons pour la Surveillance Médicale

Abstract : From few years, the general concept of perceptive spaces or smart rooms that answers in different way to the human actors needs, demands or expectations is in a continuous developing. The perceptive spaces deals with speech recognition, video signals, environmental data, persons localization, gesture following and recognition, etc. The work presented in this thesis is set on the border of the perceptive spaces and telemedecine which has recently evolved to : telesurgey, medical telemonitoring, telediagnosis, etc. Telemonitoring as one of his branches, is used especially to follow the evolution of person with accident risk, that suffer of chronic diseases or persons exposed to critical situations. This can be applied not only at home for eardely but also in a dangerous professional environment. The sound analysis and information extraction are important aspects of perceptives spaces for the medical telemonitoring. Thus, this thesis analyzes and proposes solutions to sound processing problems for the perceptives spaces, generaly and for medical telemonitoring, particulary. From all the problems linked to perceptive spaces, the automatic classification of everyday life soundswas not explored toomuch until now. A two steps sound analysis systemfor avoiding the classification of a continuous audio flow is proposed in this work. The role of the sound event detection, the first step of the proposed system, is to extract the signal to be identified from the environmental noise. If the detection method is applied simultaneously to a sound sensors array that are distributed in an appartment, it allows also a first localization of the sound source. The state of the art algorithms are not efficient enough in our work conditions and thus, new algorithms, like those using the wavelet transform, better adapted to impulsive signals are proposed. Concerning the sound classification itself, the second step of the proposed system, a first approach was the use of the automatic speech recognition techniques. These techniques are, then, improved by adding better adapted acoustical parameters among which those determined with wavelet transform and those used to detect the musical signals. The performances of the classification method are determined in a noisy environment and a preprocessing solution is presented too. The problems concerning the coupling of detection and classification steps, as well as the system evaluation are also presented. In the last part of the thesis the evolution to a sound key recognition system is approached. A real-time implementation of the proposed algorithms was realized for a medical telemonitoring application beeing in a validation process in the test appartment. Same experimental results obtained in the test appartment are presented in this work.
Complete list of metadata
Contributor : Dan Istrate Connect in order to contact the contributor
Submitted on : Friday, April 9, 2004 - 1:36:21 PM
Last modification on : Friday, March 25, 2022 - 11:10:17 AM
Long-term archiving on: : Friday, April 2, 2010 - 8:45:01 PM


  • HAL Id : tel-00005830, version 1




Dan Istrate. Détection et Reconnaissance des Sons pour la Surveillance Médicale. Interface homme-machine [cs.HC]. Institut National Polytechnique de Grenoble - INPG, 2003. Français. ⟨tel-00005830⟩



Record views


Files downloads