A MULTIMODAL PLATFORM FOR DATABASE RECORDING AND

ELDERLY PEOPLE MONITORING

Hamid Medjahed, Dan Istrate

RMSE, ESIGETEL, 1 Rue du Port de Valvins, 77 215 Avon-Fontainebleau Cedex, France

Jerome Boudy, Jean-Louis Baldinger, Bernadette Dorizzi, Imad Belfeki, Vinicius Martins

EPH,INT, 9 rue Charles FOURIER,91011 Evry, France

Franc¸ois Steenkeste

U558, INSERM, Toulouse, France

Rodrigo Andreao

Departamento de Engenharia Eltrica, Universidade Federal do Espirito Santo, Vitoria, Brazil

Keywords:

Medical Signal Acquisition, Biomedical multimodal database, Healthcare, Wearable Sensors and Systems,

Acoustic Signal Processing, Telemedicine.

Abstract:

This paper describes a new platform for monitoring elderly people living alone. An architecture is proposed,

it includes three subsystems, with various types of sensors for different sensing modalities incorporated into

a smart house. The originality of this system is the combination and the synchronization of three different

televigilance modalities for acquiring and recording data. The paper focuses on the acquisition step of the

system, usage and point out possibilities for future work.

1 INTRODUCTION

As the society is increasingly aging there is an im-

portant need to ﬁnd an intelligent support system able

to facilitate the maintenance at home of the disabled

and/or old people with safety and providing their au-

tonomy. The maintaining at home in safety of elderly

people is a new major challenge to social and health

government services: given limited resources, more

and more elderly people living alone at home are par-

ticularly prone to accidents and falls in the home and

can often lie injured and undiscovered for long peri-

ods of time. A statistical study indicates that 7% of

elderly people have a home accident due to every-

day life activity and in 84% of cases a fall occurs

(B.Th

elot, 2003). In practice all the industrialized

countries are affected by this phenomenon.

Very few systems that support the home life and

healthcare of elderly persons have been developed

to improve quality of life and the alleviation of

risk. Among established systems we can mention,

the TelePat project (French RNTS Program) (Boudy

et al., 2006) where certain physiological data and the

person’s activity are measured by different sensors

connected to a microcontroller based computing unit,

are sent through radio connection to a remote central

server application for exploitation and alarm decision.

Now, within the Tandem project (French RNTS Pro-

gram), accelerometer sensors are added to this sys-

tem for the detection of falls. In the framework of

DESDHIS project a medical home monitoring sys-

tem which use an accelerometer based sensor, infra-

red sensor, an oxymeter and a blood pressure device

has been developed at Grenoble (G.L.Bellego et al.,

2006). A system of multi-channel sound acquisition

is presented in (D.Istrate et al., 2006a), to analyze in

real time the sound environment of the home to detect

abnormal noises (i.e., call for helps or screams).

In this article a new multimodal platform for a home

remote monitoring is proposed, using a large num-

ber of sensors in order to reinforce the secure de-

tection of abnormal situations, in particular patient’s

385

Medjahed H., Istrate D., Boudy J., Baldinger J., Dorizzi B., Belfeki I., Martins V., Steenkeste F. and Andreao R. (2008).

A MULTIMODAL PLATFORM FOR DATABASE RECORDING AND ELDERLY PEOPLE MONITORING.

In Proceedings of the First International Conference on Bio-inspired Systems and Signal Processing, pages 385-392

DOI: 10.5220/0001065803850392

 SciTePress

fall event. Our software implementation gathers three

subsystems which have been technically validated

from end to end, through their hardware and soft-

ware. This speciﬁc platform is multimodal since it

allows us to record physiological data via the RFpat

(J.L.Baldinger et al., 2004) subsystem, audio infor-

mation via Anason (D.Istrate et al., 2006a) subsys-

tem and patient’s localization through infra-red sen-

sors via Gardien (S.Banerjee et al., 2003) subsystem.

An additional simulation process is added and will be

integrated to our platform as a way to overcome the

lack of experimental data required to design the deci-

sion part of the system, such as the cardiac frequency

during distress situations.

2 MONITORING SYSTEM

HARDWARE ARCHITECTURE

We deﬁne an intelligent environment as one that is

able to acquire and apply knowledge about its inhab-

itants and their surroundings in order to adapt to the

inhabitant and to improve its comfort and efﬁciency.

To record the multimodal medical database our ﬁrst

aim is focused on providing such an environment. We

consider our system as an intelligent agent, which

perceives the state of the environment using sensors

and acts consequently using device controllers. The

ﬁrst part of this intelligent environment was realized

within the framework of TelePat project, in order to

study the secure detection of patient’s fall event. The

present work is developed in the framework of the

Tandem project.

Our platform is a surface of 20 m2 in our laboratory

which is arranged in two rooms with a technical area

in order to evaluate and to supervise the experiments.

It integrate smart sensors (infra-red, audio, physiolog-

ical,) linked to a smart PC .The two microphones for

audio monitoring are linked to the PC through an ex-

ternal sound card, and can be interpreted as a single

smart audio sensor for the Anason software. Eight

infra-red sensors are ﬁxed on speciﬁc places of the

house (walls and ceiling) and connected to an acqui-

sition card (ADAM) (F.Steenkeste et al., 1999), which

is linked to the serial port of the PC. The card output is

RS485 which is converted in RS232 in order to allow

Gardien software to acquire the patient position at any

time. The RFpat subsystem is composed of two main

components: (1) a wearable terminal carried by the

patient, continuously recording his physiological data

and urgency call, (2) an in-door reception base station

linked to the PC via RS232 serial link providing the

information usually every 30 seconds. The layout of

our house environment is shown in Figure 1.

Figure 1: Layout of the house environment.

3 MONITORING SYSTEM

SOFTWARE ARCHITECTURE

The multimodal system has three main subsystems

like in Figure 2 and provides a general user inter-

face which encapsulates the Anason subsystem. It is

implemented under LabWindows/CVI software and

communicates with RFpat subsystem and Gardien

subsystem by client-server model using TCP/IP and

appropriate application protocols. Gardien is imple-

mented under C++ and recovers data every 500 mil-

liseconds. RFpat is also implemented under C++ and

receives data from receiver every 30 seconds. The use

of the inter-module communication through TCP/IP

socket allows each module (subsystem) to be run on

a different computer, and to synchronize each televig-

ilance modality channel. The user can interact with

the system via internet navigator and supervises the

different applications. For example, we use this web

server to communicate with the person, who inter-

BIOSIGNALS 2008 - International Conference on Bio-inspired Systems and Signal Processing

386

Figure 2: Software architecture of the system.

prets a patient’s activity by displaying a reference sce-

nario on the monitoring screen. This feedback can

signiﬁcantly help the system manipulation. The sys-

tem ﬂexibility obtained through TCP/IP socket com-

munication allows to add others sensors like heart

monitoring sensor (ECG).

Currently these three modalities work individually,

we investigate multimodal data fusion methods by ex-

ploiting the measurements coming from the platform

in order to increase reliability and to reinforce the se-

cure detection of patient’s distress events.

3.1 RFpat

Patient

Reception

base station

Data

Wearable

Terminal

A Smart PC

Figure 3: Architecture of RFpat system.

RFpat subsystem is composed of two fundamental el-

ements (Figure 3):

• Wearable terminal carried out by the patient con-

tinuously recording his physiological and activity

data.

• A reception base station (receptor connected to a

PC), which receives signals from the patient’s ter-

minal, analyses data in order to generate an alarm

after identiﬁcation of an emergency situations.

All the sensors data are processed within the wireless

portable device by using low consumption electronic

components in order to face autonomy problems

which are also crucial in that application. The circuit

architecture is based on different micro-controllers

devoted to acquisition, signal processing and emis-

sion. The wearable terminal includes chain of various

physiological signals, their possible pre-processing in

order to eliminate the power-line interference signal

(50 Hz) and the various measurement noises, such as

generated by the displacements of the sensors ﬁxed on

the patient’s body. The latter type of noise is generally

a factor limiting the use of such systems in ambula-

tory mode because the patient is often moving, even if

slightly. In this system, the noise problem was solved

in the acquisition stage of the portable device, by ap-

plying a digital noise reduction ﬁlter to the different

sensors signals, movements, attitude and namely the

pulse signal (heart rate). The performances of sig-

nal acquisition could be substantially improved when

the patient performs movements. The noise reduc-

tion processing (J.L.Baldinger et al., 2004) reduces

the variations of pulse measurement lower than 10%,

even 5%, which remains in conformity with the rec-

ommendations of the health professionals.

The design of sensors and embedded processing has

led to the realization of a remote wearable monitor-

ing terminal, equipped with actimetry and physio-

logical sensors, indicating the attitude of the patient

(vertical/horizontal positions, activity) and his heart

rate (pulse measurement); these speciﬁc sensors to

recorded physical data type are, either integrated in

the terminal, or externally ﬁxed .

Data generated from the different sensors are trans-

mitted, via an electronic signal conditioner, to a

micro-controller based computing unit, embedded in

the mobile terminal ﬁxed on the patient’s waist. Cur-

rently, a fall-impact detector sensor is added to this

system for robustizing the detection of falls.

3.2 Anason

The sound remote monitoring subsystem analyzes the

acoustical environment in real time and is made up

of four main modules which are presented in the Fig-

ure 4 (D.Istrate et al., 2006b). The ﬁrst module M1

continuously, supervises the sound environment in

order to detect and extract useful sounds or speech

from environmental noise. The signal extracted by

the M1 module is classiﬁed like sound or speech

by the M2 module. In the case of sound label, the

sound recognition module M3.1 classiﬁes the signal

between eight predeﬁned sound classes, while in the

case of speech label, the extracted signal is analyzed

by a speech recognition engine in order to detect dis-

tress sentences. For both cases, if an alarm situation

has been identiﬁed (the sound or the sentence belong

to an alarm class) this information is sent to the data

fusion system.

A MULTIMODAL PLATFORM FOR DATABASE RECORDING AND ELDERLY PEOPLE MONITORING

387

Figure 4: Sound monitoring architecture.

Sound Event Detection Module (M1). The sound

ﬂow is analyzed through a wavelet based algorithm

aiming at sound event detection. This algorithm

must be robust to noise like neighborhood environ-

mental noise, water ﬂow noise, ventilator or electric

shaver. Therefore an algorithm based on energy of

wavelet coefﬁcients was proposed and evaluated in

(D.Istrate et al., 2006a). This algorithm detects pre-

cisely the signal beginning and its end, using proper-

ties of wavelet transform.

Sound/Speech Classiﬁcation Module (M2). The

method used by this module is based on Gaussian

Mixture Model (GMM) (D.A.Reynolds, 1995) (K-

means followed by Expectation Maximisation in 20

steps). There are other possibilities for signal clas-

siﬁcation: Hidden Markov Model (HMM), Bayesian

method, etc. Even if similar results have been ob-

tained with other methods, their high complexity and

high time consumption prevent from real-time imple-

mentation. A preliminary step before signal classiﬁ-

cation is the extraction of acoustic parameters: LFCC

(Linear Frequency Cepstral Coefﬁcients)-24 ﬁlters.

The choice of this type of parameters relies on their

properties: bank of ﬁlters with constant bandwidth,

which leads to equal resolution at high frequencies

often encountered in life sounds.

The BIC (Bayesian Information Criterion) is used

in order to ﬁnd the optimal number of Gaussians

(G.Schwarz, 1978). The best performances have been

obtained with 24 Gaussians.

Sound Recognition Module (M3.1). This module

is based, also, on a GMM algorithm. The LFCC

acoustical parameters have been used for the same

reasons than for sound/speech module and with the

same composition: 24 ﬁlters. The method BIC has

been used in order to determine the optimum num-

ber of Gaussians: 12 in the case of sounds. A log-

likelihood is computed for the unknown signal ac-

cording to each predeﬁned sound classes; the sound

class with the biggest log likelihood is the output of

this module.

Speech Recognition Module (M3.2). For Speech

Recognition, the autonomous system RAPHAEL is

used (M.Akbar and J.Caelen, 1998). The language

model of this system is a medium vocabulary statisti-

cal model (around 11,000 words). This model is ob-

tained by using textual information extracted from the

Internet as described in (D.Vaufreydaz et al., 1999)

and from ”Le Monde” corpora. It is then optimized

for the distress sentences of our corpus. In order to in-

sure a good speaker independence, the training of the

acoustic models of RAPHAEL has been made with

large corpora recorded with near 300 French speakers

(J.L.Gauvain et al., 1990): BREF80, BREF120 and

BRAF100 corpora.

3.3 Gardien

Figure 5: The Gardien system.

The subsystem knows as Gardien (Figure 5) consists

of passive infra-red sensors placed in a residence and

connected to a remote computer. All the sensors are

connected through cables to an Input/Output parallel

card (ADAM 4053) which is connected to a master

PC. The computer automatically captures and regis-

teres data obtained from the different sensors, with

the help of Gardien software. Data corresponding to

movements are collected twice per second, and stored

with the time of the event in a speciﬁc ﬁle. When

several consecutive data are identical, only the ﬁrst

instance is stored.

The sensors are activated by passage of person un-

derneath, and remained activated as long as there is

movement under that sensor and for an additional

time period of 1/2 seconds after the movement end.

The results from the automatic processing of this data

are displayed in the form of list with all movements

noted together with the time and each movement’s

BIOSIGNALS 2008 - International Conference on Bio-inspired Systems and Signal Processing

388

duration. Gardien is also able to display the data

either in the form of graph (activity duration versus

days) or as three-dimensional histograms (each sensor

activation versus time). To validate the system, the

results from the automatic processing are compared

with manual analysis by an expert.

4 GENERAL INTERFACE OF THE

SYSTEM

Figure 6: Main windows of the system during the acquisi-

tion step.

Figure 6 shows the front panel of the software sys-

tem, where we can supervise the multimodal data ac-

quisition step. The user must ﬁrstly select the modal-

ity to record and to conﬁgure its parameters. RFpat

and Gardien need only to select the IP address and

the TC/IP port number, while Anason requests the se-

lection of the sound card (if two are present), the sam-

pling rate and the location of the backup ﬁle.

5 THE BIOPHYSICAL SIGNALS

SIMULATION STAGE (BSS)

The aim of this stage is to create pathological or crit-

ical situations for the patient at home. Indeed most of

actual signals recorded on domotic platform are gen-

erally and hopefully in normal conditions. The sim-

ulator is based on the existing RFpat sensors device.

The ﬁrst main goal was to simulate cardiac patholog-

ical proﬁles such as in particular bradycardias: the

design was done with the helpful collaboration of

SAMU-92 (French emergencies service. In its im-

plementation, are also foreseen functional stages for

the actimetry simulation: patient’s inclination (hori-

zontal or vertical position), his body movement and

Start

INCLINATION status choice:

Data already recorded (a)

Or Create a profile (b)

MOVEMENT ratio choice:

Data already recorded (a)

Or Data simulated (b)

Or Create a profile (c)

PULSE (cardiac frequency)

Data already recorded (a)

Normal profile (b)

Abnormal profile : Bradycardia (c)

File

MESOR pattern (a)

Or Cardiac cost model (b)

Correcting Vectors dimension

Creating the storage file

Results

(a)

(b)

Use normal file (a)

Or Generate a

profile (b)

(c)

Figure 7: The Biophysical Signals Simulation.

in a larger extent patient’s fall situations. The simula-

tor software architecture is summarized in the Figure

7. For the cardiac frequency generation, three cases

were proposed:

• A ﬁrst normal cardiac category, based on the

COSINOR method (F.Halberg, 1969), providing

a global pulse variations trend within one day;

this formula gives the cardiac frequency or pulse

Fc in quiet situation under the following form:

Frest(t) = Fmoy + Asin(2pi/24 ∗t), where Fmoy

(around 70 bpm), A are respectively the average

pulse or MESOR value and its maximal ampli-

tude variation (about 6 bpm) along one circadian

cycle; the Akrophase or maximal amplitude is sta-

tistically located around 16 hour.

• A second normal situation, called ”Cost model”,

providing a pulse variation model, still denoted

Fc, depending on the patient’s activity; the for-

mula is based on the pulse in a quiet situa-

tion (Fcrest) and the delta-variation due to pa-

tient efforts, namely representing the cardiac cost:

Fc(t) = Frest(t) + DFc(t). This part is develop-

ing stage.

• Bradycardia model corresponding to a situation

met with elderly persons, by assuming no speciﬁc

medication having a cardiac impact: the model is

either artiﬁcially inserted inside an existing pulse

signal sequence by taking into account the actual

pulse variance, or is completely substituting the

actual pulse sequence.

This tool is still open to other simulation process,

namely for the actimetry where presently are per-

A MULTIMODAL PLATFORM FOR DATABASE RECORDING AND ELDERLY PEOPLE MONITORING

389

formed investigations on the potential correlations be-

tween the Cardiac cost model and the body moveme-

nent. The BSS stage has been designed to be fully

interfaced to the multimodal patient database.

6 APPLICATION

6.1 Recording of a Multimodal Medical

Database

Most of monitoring systems use some form of learn-

ing method to discriminate between different types of

normal and abnormal events. This methodology re-

quires large amounts of training data that can be dif-

ﬁcult to obtain especially data describing abnormal

events that are by deﬁnition rare occurrences. An im-

portant issue for this problem, is to record a multi-

modal medical database which is the ﬁrst application

of our platform.

Data acquired from the patient are stored on the Mas-

ter PC in a folder named with a code number corre-

sponding to the patient. Each recording is composed

from ﬁve ﬁles corresponding to the different subsys-

tems.

The ﬁrst one, named ”personnel.xml”, contains the

patient’s identiﬁer and some personal information like

age, native language, usual drugs treatments, etc. The

second, named ”scenario.xml ”, describes the refer-

ence scenario. All these data relative to the tester are

protected for his privacy and let to his agreement.

The sound data is saved in real time, in a wav ﬁle with

16 bit of resolution and a sampling rate of 16 KHz, a

frequency usually used for speech applications.

The clinical data acquired from RFpat are saved in

a separate ﬁle which contains information about pa-

tient’s attitude (lied down or upright/seated), his agi-

tation (between 0% and 100%), his cardiac frequency,

fall events and emergency call in a binary type. The

acquisition sample rate is 1/30 seconds.

The data acquired every 500 milliseconds by Gardien

subsystem are saved in a separate text ﬁle, fully re-

specting the original storage format of the GARDIEN

application. Each line of this ﬁle contains the infra-

red sensors which are excited (they are represented

by hexadecimal numbers from 1 to D) plus the corre-

sponding date and hour.

To tackle the problem of the variety of each data sam-

pling rates, a synchronization prototype between the

three subsystems is obtained through TCP/IP proto-

col. RFpat is the master and supervise Gardien and

Anason by TCP/IP commands.

Thus, our multimodal database acquisition software

provides a very helpful and well-targeted application

to elaborate and assess the data fusion-based decision

methods. The low level data recorded by our system

will be useful for the development of processing algo-

rithms of each modality.

Figure 8: Sound ﬁle (*.wav) and its corresponding SAM

ﬁle.

In order to index our multimodal database, we have

retained the SAM standard indexing ﬁle (D.Well

et al., 2004) generally used for Speech Databases de-

scriptions. The SAM labeling of a sound ﬁle is shown

in Figure 8; it indicates information about the sound

ﬁle and describes this ﬁle by delimiting the useful part

for analyzing and processing. For each modality of

the database a corresponding indexing ﬁle is created,

we have adapted this type of ﬁles to the speciﬁcities

of each modality, and we have added another indexing

ﬁle for the entire database.

6.2 The First Approach of Fusion:

Bimodal Fusion between RFpat and

Gardien

This work is a ﬁrst step for a multi-modal experiment.

Indeed, this was performed with only two of the tele-

vigilance modalities presented in this paper: the ﬁxed

infra-red sensors based Gardien system and the mo-

bile sensors RFpat device. Its conclusions have mo-

tivated the extension of these modalities to the com-

bination with the sound detector AnaSon through the

current investigations led by ESIGETEL, INT and IN-

SERM.

In this ﬁrst step we have used a PCA analysis in or-

der to preliminary determine potential correlation be-

tween the combined data and in order to obtain a data

reduction. After preliminary evaluations with the K-

Nearest Neighbors algorithm, the Gaussian Mixtures

Models and Neural Network on RFPAT data, a Bi-

modal fusion was carried out by using the Neural Net-

work .

BIOSIGNALS 2008 - International Conference on Bio-inspired Systems and Signal Processing

390

6.2.1 Principal Components Analysis

The data resulting from RFpat subsystem and Gardien

subsystem were analyzed simultaneously through

their respective variables: posture, activity, cardiac

frequency, exposure time under the infra-red sensor

C3 (sensor indicating the input/output of the person in

the home), and exposure time under the infra-red sen-

sor C5 (sensor indicating the fall of the person). The

analysis of the PCA algorithm results made it possible

to propose a set of decision rules on several levels:

• To deﬁne an estimator in two levels: a ”physio-

logical” distance between two parameters (cardiac

frequency, activity) normally correlated: normal

state if they are close or pathological state if they

are distant. Then, a ”actimetric” distance (Slope,

C5): normal if distant or pathological if close.

• There are a correlation between the cardiac fre-

quency and the activity which will allow the fu-

sion system to avoid a malfunction of one of the

two sensors.

6.2.2 Application of the Neural Networks

The Neural Networks (NN) consist in an input layer,

the sensors signals, several transition layers (denoted

as hidden layers) and of an output layer delivering

the classiﬁcation of the data observed in situation ei-

ther ‘Normal’, or critical ‘detected Fall’. A classi-

cal NN structure was implemented by using a Multi-

Layer Perceptron (MLP) based on only one hidden

layer consisting in eight neurons after an optimal tun-

ing.

Each neuron realizes a scalar product between its

input vector and the weight vector, where a deviation

is added, then operates an activation function in order

to generate its output value y: y = f (x.w + b).

The activation function must be strictly crescent and

bounded. A classical function used in our experiment

is the standard sigmoid function whose equation is re-

minded hereafter: f (x) =

tanh(x)+1

Two types of networks were compared, with respec-

tively as input vector of the MLP ﬁrst layer:

• Single actimetric data of RFpat in entry of the

network, giving a rate of recognition of the order

84%.

• The actimetric data of RFpat and horizontal infa-

red sensors of Gardien, providing a rate of 86%.

The improvement nevertheless remains quite limited.

One improvement track is to increase the data cor-

pus used for the learning phase, namely by recording

more speciﬁc actual and simulated emergency situa-

tions thanks to the multi-modal recording tool previ-

ously described in this paper. Another main improve-

ment track will be investigated by adding the AnaSon

(abnormal sound detector) modality. Therefore that

is why the need of a new multi-modal recording tool

was considered as crucial for the follow-up. Thor-

ough investigations will also be performed again on

KNN and GMM techniques, namely by working on

the data pre-processing (normalized, transformed in-

put data).

7 CONCLUSIONS

This paper has focused on the technology used for

implementing the acquisition step of the platform.

Preliminary results are encouraging with the achieve-

ment a multimodal medical database including pa-

tient’s clinical data, usual environment sounds and pa-

tient localization. The platform enables us to have a

full and tightly controlled universe of data sets and to

evaluate the decision part of remote monitoring sys-

tems.

Our platform is in the research phase targeting a pro-

totype, the system will be completed and improved by

adding a data fusion-based decision element exploit-

ing the measurements coming from this platform in

order to propose new processes to reinforce the secure

detection of patient’s distress events. In particular the

fall situations are studied: indeed one or more televig-

ilance modalities might be out of order, or a particular

environmental situation (ambiant noise, bad wireless

conditions, sensors disabilities .. .) can hide one par-

ticular modality or more. This is a very challenging

issue for hospital emergency units such as for instance

SAMU in France or Telecare services providers in

general. Studies of usability are planned, in order to

test the satisfactoriness of patients towards this system

and to get a standardization prototype for our plat-

form. This constitutes indeed, a ﬁrst concrete step

before a prototype deployment. In actual situation,

evaluation and connection to smart home system are

also planed to be performed in the framework of a

new European project.

REFERENCES

Boudy, J., JL.Baldinger, F.Delavault, Farin, D., R.Andreao,

S.Torres-Muller, A.Serra, F.Rocaries, C.Dietrich,

F.Steenkeste, A., Dr.M.Baer, Dr.Ozguler, and D.Gaiti

(June 2006). Telemedecine for elderly patients at

home. In the TelePat project.ICOST conference ulster

Belfast.

B.Th

elot (June 2003). Resultats de l’enquete permanente

sur les accidents de la vie courante. In Reseau

A MULTIMODAL PLATFORM FOR DATABASE RECORDING AND ELDERLY PEOPLE MONITORING

391

EPAC,Institut de Veille Sanitaire,Departement mal-

adies chroniques et traumatismes.

D.A.Reynolds (January 1995). Speaker identiﬁcation and

veriﬁcation using gaussian mixture speaker models.

Speech Comm, 17:91–108.

D.Istrate, E.Castelli, M.Vacher, L.Besacier, and J.F.Serignat

(April 2006a). Information extraction from sound for

medical telemonitoring. Transactions on Information

Technology in Biomedicine, 10:264–274.

D.Istrate, M.Vacher, and F.Serignat, J. (Aug 30-Sept3,

2006b). Generic implementation of a distress sound

extraction system for elder care. In 28th IEEE

EMBS Annual International Conference,New York

City, pp.3309-3312. IEEE.

D.Vaufreydaz, J.Rouillard, and M.Akbar (December 1999).

Internet documents: a rich source for spoken language

modelling. Workshop ASRU’99, Keystone-Colorado,

pages 277–281.

D.Well, J.Barry, W.Grice, M.Fourcin, and A.Gibbon

(2004). Sam esprit project 2589-multilingual speech

input/output assessmen,methodology and standardiza-

tion. In Final report. Technical Report SAM-UCL-

G004, University CollegeLondon.

F.Halberg (1969). Chronobiology. In Ann Rev Physiol,

vol.31,pp.675-725.

F.Steenkeste, H.Bocquet, M.Chan, and B.Vellas (December

1999). Remote monitoring system for elders in a geri-

atric hospital. In Promoting Independence and quality

of life for older persons:an international conference

on aging Arlington.

G.L.Bellego, N.Noury, G.Virone, M.Mousseau, and

J.Demongeot (January 2006). Measurement and

model of the activity of a patient in his hospital suite.

TITB, 10:92–99.

G.Schwarz (1978). Estimating the dimension of a model.

Annals of Statistics, 6:461–464.

J.L.Baldinger, J.Boudy, B.Dorizzi, J.P.Levrey, R.Andreao,

C.Perpre, F.Devault, F.Rocaries, and A.Lacombe, C.

(Juillet 2004). Tele-surveillance system for patient

at home: The medeville system. In Congrs ICCHP

2004,Paris.

J.L.Gauvain, L.F.Lamel, and M.Eskenazi (1990). Design

considerations and text selection for bref, a large

french read-speech corpus. In ICSLP ’90, pp. 1097-

1100.

M.Akbar and J.Caelen (1998). Parole et traduction au-

tomatique : le module de reconnaissance raphael. In

COLING-ACL’98, Montral, Quebec, vol.2, p. 36-40.

S.Banerjee, F.Steenkeste, P.Couturier, M.Debray, and

A.Franco (Septembre 2003). Telesurveillance of el-

derly patients by use of passive infra-red sensors in a

smart room. In Telemed Telecare.

BIOSIGNALS 2008 - International Conference on Bio-inspired Systems and Signal Processing

392