Fig 3 - uploaded by Michel Vacher
Content may be subject to copyright.
Source publication
Medical Remote Monitoring needs human operator assistance by smart information systems. Physiological and position sensors give already numerous informations, but sound classification can give interesting additional informations about the patient and may help to the decision-making. A Real-Time implementation of a multichannel smart sound/speech sy...
Context in source publication
Context 1
... graphical user interface is the Second Parallel Task. It is used for setting up the application and to display signal and results on the screen as they are available (see Figure 3): chan- nel calibration, wave visualisation, choice of features, detec- tion/classification results and recognized sentences. ...
Similar publications
This paper presents #FIVE (Framework for Interactive Virtual Environments), a framework for the development of interactive and collaborative virtual environments. #FIVE has been developed to answer the need for an easier and a faster conception and development of virtual reality applications. It has been designed with a constant focus on re-usabili...
Citations
ABSTRACT
This paper presents text-independent speaker identification system for the Amharic
language. For the identification process, speech signals are collected from different
speakers including both sexes as well as different age groups. Each speech signal is
taken with a sampling frequency of 16KHz and 16 bit. Different speech pre-processing
techniques such as end point detection, pre-emphasis filtering, frame blocking and
windowing have been used to process the speech utterances. Then MFCC, ∆MFCC and
∆∆MFCC have been extracted. After these features are extracted from each speech
signal, Vector Quantization and Gaussian Mixture Models have been used for training
and identification purpose. The first one is template matching algorithm where as the
second one is stochastic. This is to see which approach is better for text-independent
speaker identification. For a total of 50 speakers, 74.2% accuracy is achieved when
vector quantization approach is used where as 84.3% accuracy for the Gaussian
Mixture Models. We got better results when we increased the number of features used.
After doing the experiment for the total number of speakers, we then tried speaker
identification based on gender. 25 male and 25 female speakers were considered.
From the experiment, we got 86.2% and 85.9% accuracy for male and female
speakers. Finally, we have carried out an experiment based on age groups. We
considered 20 speakers whose age is 40 years or younger and another 20 speakers
whose age is above 40. From the experiment, we got 86.8% for those whose age is 40
or younger accuracy and 86.3% accuracy for those whose age is above 40 years. The
result shows that speaker identification is feasible for the Amharic language. Moreover,
speaker identification is age and gender dependent.
The average age of the population in industrialized countries is steadily increasing. Seniors living alone are more numerous, either because they prefer to live independently, or because of a lack of space in the specialized institutions. We must find solutions allowing them to continue to stay at home comfortably and safely. Smart housings can constitute one of these solutions. One of the biggest challenges in Ambient Assisted Living (AAL) is to develop smart homes that anticipate the health needs of their inhabitants while maintaining their safety and comfort. It is therefore essential to facilitate interaction with the smart home through systems that respond naturally to voice commands, using microphones but no tactile interfaces. This thesis defines the concept of smart home and describes some interesting projects. It then explains how home assistance can take advantage of this concept thanks to sound analysis. The acceptability of a voice interface as part of the intelligent home has been studied through an experiment that showed what are the wishes, expectations and fears of older users, their families, and social workers. Audio analysis in smart homes is a still unexplored research area, the interest and the methods to analyze sound in- formation in a smart home have been studied here in an experiment which helped to highlight the challenges and technological obstacles to be removed in order to use sound information in addition to other modalities and, in the case of speech, distant speech recognition (ASR in remote recording conditions). A practical solution using several microphones is then presented. The intended purpose is to achieve a voice control system which will allow the user to control their environment not only by conventional switches and remote control devices, but also by voice. The advantage of the audio information combined with that of home automation sensors is then revealed through a multimodal analysis making it possible to locate a person in a smart home or to determine their activity. Localization is necessary, for example to determine the context in which a home automation command has been issued. The activity can be used to observe changes in the habits of a person to assist in diagnosis. Finally, the thesis presents the perspectives of research and future projects of the author. It is accompanied by the reproduction of four scientific papers published in refereed selective conferences.
