FR2533513A1 - Method and system for communicating, on board a motor vehicle, complex information relating to the vehicle and its environment - Google Patents

Abstract

This system comprises a receiver of acoustic information, such as a microphone 1, a speech recognition device 2 coupled to the receiver, a speech synthesis device 12 coupled to a loud-speaker 13, a device 3 for actuating devices A1, A2,... An of the vehicle, a set of sensors C1, C2,...Cn of parameters of the vehicle connected to an interface module 9 and a central processing unit 11 coupled to the speech recognition device 2, to the speech synthesis device 12, to the actuating device 3 and to the interface module 9 and controlling the actuating device 3 and the speech synthesis device 12 according to instructions transmitted and decoded by the speech recognition device 2.

Description

METHOD AND SYSTEM FOR COMMUNICATING ABOARD A MOTOR VEHICLE 3aES COMPLEX INFORMATION RELATING TO THE VEHICLE AND ITS ENVIRONMENT
The present invention relates to a system for communicating, on board a motor vehicle, complex information relating to the vehicle and its environment, as well as a method of operating this system.

 It makes it possible to control more or less simple organs, to access information or to provide information, such as, for example, to control secondary-order functions (windshield wipers, headlamps, lights, start-up). engine, etc ...), to select a particular car radio station, to obtain precise information by means of speech synthesis or light displays (tank capacity, distances traveled, etc ...), d. emitting information (programming of an on-board computer, etc.), by means of recognition and synthesis of speech, obeying a man-machine speech communication scheme , in addition to visual and tactile assessments.

As a general rule, the driver of a motor vehicle uses visual and tactile means (eyes, hands and feet) to move his vehicle and to order organs that are of primary order (steering wheel, brake, accelerator, etc.), or secondary order (flashers, wipers, headlights, etc ...)
Beyond a certain visual space, the driver can no longer appreciate all the information that his eyes receive and thereby, among other things, selects a part of it, it is acquired much less quickly. It is the same with the manual action on the keys, switches, etc ..., 01, beyond a certain amount, the system becomes partly untapped. For the reception of acoustic waves, the problem will also arise in the same way as soon as a number of useful or disruptive sound sources will be passed.

If the eyes and the ears are considered commercial receivers where a complex information circulates and is then treated, by our brain, the hands are rather felt like controlled actuators as a result of decisions taken by our brain With regard to the actions like the arcs-reflexes For example, it is not easy to judge whether the information emitted by our brain is complex or not. On the other hand, to ask our tactile actuators to place orders, search switches, etc ..., to perform more or less complex actions (ranging from the simple order of contact to the passage of useful information for the programming of a system called "intelligent"), appears tedious, complex and sometimes inappropriate.

The manual interface, in these cases, appears as a means that can not quickly render the services expected.

The evolution of the motor vehicle tends to provide a lot more pres tatios than in the past, especially in terms of information, and sometimes requires the driver to a more active phase, the latter is already very much solicited by the inner environment and outside.

For the service rendered to be optimal, it is necessary to access the information as quickly as possible and with the greatest flexibility allowed.

So this communication support must have a very large information rate, each information entity being compressed as much as possible.

Indeed, for a medium having a given information rate, the more the coding of each entity is reduced, the more entities are passed in a given period of time. This is allowed by the speech (uncoding passage from a few tens of? <Ilobits / s to a few tens of bits / s).

This means of communication must be seen as a complement to existing ones (visual and tactile) and to be adjusted in the most rational and harmonious way. The vehicle will be perceived with different aspects, going in the direction of its evolution, revealing possibilities previously unauthorized.

To this end, the subject of the invention is a system for communicating, on board a motor vehicle, complex information relating to the vehicle and its environment, comprising an acoustic information receiver such as a microphone, a device for speech recognition coupled to the receiver, speech synthesizer coupled to a loudspeaker, vehicle organ actuator, set of vehicle parameter sensors connected to an interface module and a central unit coupled to the recognition device, the synthesis device, the operating device and the interface module and controlling the actuating device and the speech synthesis positive as a function of instrucions transmitted and decoded by the device. speech recognition device.

Preferably, the communication system also comprises an on-board computer and / or a car radio and / or a terminal provided with a keyboard and a display screen controlled by the central unit according to transmitted instructions. and decoded by the speech recognition device.

Other features and advantages of the invention will emerge from the description which follows, with reference to the appended drawings in which: - Figure i is a block diagram of the whole communication system according to the invention - the FIG. 2 is a block diagram of an exemplary speech recognition device used in the system of FIG. 1; FIG. 3 is a block diagram of the organ actuator used in the system; FIG. FIG. 4a is a block diagram of a car radio used in the system of FIG. 1; FIG. 4b represents the volume control circuit of the car radio of FIG. 4a; FIG. block diagram of an on-board computer used in the system of FIG. 1 - FIG. 6 is a two-level tree diagram illustrating the operation of the system of FIG. 1 in speech recognition mode - FIG. is a block diagram of a synt device hess the role used in the system of Figure 1; FIG. 8 is a block diagram of the central unit of the system of FIG. 1; FIG. 9 is a block diagram of an output parallel expansion system used in the central unit represented in FIG.
Fig. 8 - Fig. 10 is a logic signal diagram explaining the operation of the system of Fig. 9; FIG. 11 is the main flow diagram of the central system log of the system of FIG. 1; FIG. 12 is a flowchart illustrating the "vocabulary learning" mode of the system of FIG. 1; FIG. 13 is a flowchart; illustrating the loading or the saving in the memory of the central unit of the system of a reference vocabulary - FIG. 14 is a flowchart illustrating a mode of operation of the system making it possible to modify or add acoustic references to the vocabulary stored in the speech recognition device - Figs. 15a and 15b together represent a system operation algorithm in speech recognition mode, and - Fig. 16 shows an algorithm used to manage the expansion system illustrated in Fig. 9.

Referring to FIG. 1, the communication system comprises a central unit 11 to which are coupled a speech recognition device 2 provided with a receiver microphone 1 a speech synthesis device 12 equipped with a receiver. 13, a device 3 for actuating a number of A1-A2-An organs of the vehicle, a display system 4 and manual dialogue, a car radio 6 equipped with a speaker 5 and an antenna 7, an onboard computer 8 receiving information from a number of sensors Col, Co2 -Con, an interface module 9 with other sensors Ci, C2-Cn of vehicle operating parameters, as well as a set of 10 dead and alive memories.

The microphone 1 is used to pick up the acoustic information coming, in general, from the driver. This microphone is required to use the speech recognition system or device 2. For some time now, speech recognition systems have been offered for a reasonable price, with a problem-free integration and opportunities to perform speech recognition. precise applications.

These systems differ from one another by several criteria, the main ones being the method of analysis (temporal, frequential, pseudo-temporal, etc ...), the level on which the recognition algorithm carries (acoustic level, phonetic, etc. ...), the principle of information processing, etc ...).

If some systems use a method related to a given language structure, other systems have the possibility to overcome this criterion using a method operating at the acoustic level It is this second method which is preferably used in the system according to the invention, for various reasons which can be stated the main ones.

The first point is that the criterion acoustic level offers a form of universality for the use This universality exists because the system requires to have done at least once the characterization phase before the use of the recognition phase .

This characterization phase, for example, may disappear when using the phonetic criterion by offering a multilocutor orientation, but restricts the use of the application because the study is generally based on a given language.

A second point is that currently, it is these types of speech recognition systems that offer the best compromise between price and quality.

It is interesting to give a brief description of this type of speech recognition system, to better understand some of the software managing the system and described later.

System 2 is shown in FIG. 2 where 20 designates the acoustic signal
tick (from microphone 1 or magnetic tape or any other
which can reproduce an original acoustic signal
re acknowledge the recognition). This signal attacks an amplifier-correct stage
21, a bank of filters 22 extracting a spectrum of energy over a range of
from 200 / hz to 7khz, a multiplexer 23 clocked by a clock
10 to 20 ms approximately, a digital analog converter 24, a processor 25, of the ROM 28 containing the al
gorithm, rules and overall system management, memory
RAM 29 containing the dictionary (i.e., the characterizations
acoustic references created at the time of learning) as well
that programmable parameters intervening on the algorithm of recon
birth, and an input / output communication system 26 with the central unit II which is a superior decision-making system allows
both to exploit the entire speech recognition system for the intended application.

The two important phases of operation of the recognition system
speech session 2 are learning and recognition but
other phases are adjoined and are processed by the software of the u
central unit, allowing a richer exploitation of the sys
An example illustrating this point is that of the analysis, by the lo
central software, of the spoken word sonogram in order to deduce
conclusions as to the nature of the parasitic noises, the more or less good validity of the recognized acoustic form for a classification
in a sense of evolution of the speaker, etc.

The learning phase consists of creating the reference dictionary
acoustic. This dictionary created, the recognition phase consists
to pronounce a reference among those learned to be compared
speech recognition system 2 then deduces a set of
recognition criteria that are analyzed by the central software.

A set of criteria or a set of criteria sets is associated
to a decision, with an interactive character. This set of criteria
is composed of a recognized word indication or not. In the case of the unrecognized word there are cases of acoustic forms too short, too
strong or too weak in energy, bad detection at the start.

In the case of a recognized word, there is the indication of the number of the first reference best recognized with its confidence level, as well as the second reference with its confidence threshold. It is understood that following this set of criteria, the central software can program the speech recognition system 2 in another configuration for the acoustic references that will be pronounced.

This programming can, for example, limit or increase the field of the references of the dictionary to make the search, to play on the value of the confidence threshold for the recognition, to decide that the reference pronounced can serve as temporary reference in supplement in the dictionary , remove from add some of the references in the dictionary. More specific examples will be cited later
Systems that can be used are VRM-102 from IEC (USA) and MOISE from VECSY.S. (La France). Of roughly identical structures (see Figure 2), they have significant differences.The first system uses a statistical algorithm (10 to 15 training passes to obtain a recognition rate of 99 fo, in a medium not too much noisy and with a body of references from the dictionary of a few tens). Each reference in the dictionary uses a memory zone of 34 bytes of RAM. The second system, for the same conditions of use, requires a single learning curve because the algorithm works on the variations of the information. The memory space for a reference is 100 bytes. Both systems are of the monolocutor type.

Block 3 constitutes the device for actuating the system. This part transforms electrical signals emitted by the central unit into mechanical or electrical power actions. Indeed, this block replaces the manual actions usually performed by the driver. More specifically, these actions are grouped under the term secondary acceleration as opposed to so-called primary actions. The primary acceleration is represented by an analog variable, namely the maneuvering of the steering wheel (due to its nature of change of direction), the acceleration, braking and clutch pedals. Depending on whether it is a vehicle with an automatic transmission or not, the differences are with the clutch pedal and the gearshift system. In the case of the automatic, it is possible to install an actuator for the vi change system
tesses.

FIG. 3 represents the different parts of this block 3. It comprises an interface adaptation module 31 to which a signal is applied.
electrical system 30 from the central unit 11, a security part
32 mechanically reconstructing those usually performed mechanically (for example, the action of
left, on the right, while leaving the superior possibility of
run the hazard warning lights), a decoupling device
33 of the opto-electronic type, an amplification portion 34 and an actuator portion 35 (relays, power transistors, thyristors,
etc ...> connected to different organs A. to order These different
Ai organs (Al, A2, --- An in Figure 1) include: the three posi
lighthouses (code lights, crossing lights and road lights, which may be coupled with fog or
road regulations. The call of lighthouses is done by logi
sky In fact, each time which driver wants to make a call
of headlights manually, it acts three to four times on a command
having a reminder. Each action lasts a certain time and is separated
another action at another time. When the driver utters
the acoustic form corresponding to the call of lighthouses, the cen software
He will make the same order, taking into account the times of
time delay, using the high beam control. The part of the lo
software processing this function is described later); the indicator of
direction (the mechanical system stopping the indicator after a certain
steering angle is replaced by an electronic sensor placed on
the steering column but a time delay maybe made by the
central software); the hazard lights the different modes of operation
windshield wipers (normal, fast and clock speed
although this last case can be realized by the central software in
using timers); the window washer (each jet of water is tem
porisé and accompanied by a few strokes of wipers, made them
also by software); climb controls. and down the windows
equipped with time security made by software (this time
porisation is slightly greater than the maximum time that a glass pane passes from one limit position to the other. This makes it possible to avoid deterioration of the electric motor when the window is locked in one of the limit positions); the electromagnetic condemnation of the doors which is an exclusive command; defrosting the rear window; the control of the reading lights and the ceiling lamp; starting and stopping the vehicle engine under certain conditions; the horn with certain combinations made by software; fan control; windshield wipers on the rear window.

This list is not exhaustive in itself and it is easy to understand the ease of use that precedes the speech recognition when the number of commands to be filled becomes more important and this notion of facility takes on even more importance. when it is known that, in normal times, the implementation of the various hand controls on the dashboard varies a lot from one type of vehicle to another, with each change of vehicle it follows a discomfort at the beginning that says - appears partly in time1 because of the driver's compatibility with the vehicle. The habit is usually done with the aid of arcs reflexes for controls that are used with great frequency (direction indicator, lighthouse call and horn) But unsolicited orders are often the subject of some research time
Thus, by this voice control system, it is possible to circumvent the difficulty of the problem of non-universality of the implementation of manual controls. It is possible, through an easily transportable information storage medium (badge, magnetic card, etc.).

to load its own reference dictionary into the vehicle system which it then wishes to use with this command mode. We can estimate that an order of magnitude of 30 to 50 words or acoustic forms (words to command, cancel and redundant) suffice for this part of application, which represents a few kilobytes of information to be transported. For the application, the reference dictionaries are saved in RAM or dead, programmable (part of block 10, Figure 1). Before each use, the driver loads his own dictionary, either manually or by voice.

Of course, this system can be multiuser, that is to say that each has its own dictionary, the only limitation from the size of the memory.

Another important feature that emerges is that you can customize your vehicle. Indeed, each desired action is associated with an acoustic form or a group of acoustic forms, but several acoustic forms or groups of distinct acoustic forms can lead to the same action. For example, to put the direction indicator in the deaie position, we can take the acoustic form "flashing- right" pronounced in full or "flashing" then "right" by marking a silence, between the two acoustic forms, a few hundred of ms, or to associate one or several other acoustic forms corresponding in finality to the same action Moreover the silence can be omitted when one uses a system of recognition of the speech operating with a method of connected words. This last aspect, making appear a tree, will be discussed later So the user chooses the words he wants to associate with the desired action. The redundant nature (2 to 3 acoustic forms for the same state) offers, in general, more flexibility for the user. This notion is optional and left to the free arbitration of the driver
Another important aspect for this part is that of security
Indeed, most recognition systems are monolocuteur, that is to say that a user, operating in the recognition phase on a reference dictionary previously created by another speaker, gets worse results. So using the confidence threshold and a string of two or three acoustic forms to be pronounced, so a vocal key is made. In general, two or three drivers use the same vehicle. When one of the drivers, or an intruder, comes to the system, he must pronounce a keyword that can be either its name or a particular code, etc. , left to choose. The acoustic form thus pronounced is compared with a reference dictionary said common. This dictionary contains all the keywords of the users and is loaded automatically by the block 11, from the memory 10, into the speech recognition system (block 2) each time the general system is powered up. this comparison operation, we obtain a result of the speech recognition system indicating whether a word has been recognized or not with respect to a confidence threshold initially set. After a few attempts, the system stops its recognition phase If, on the other hand, a word is recognized, we examine whether it responds to a keyword and in this case we load the reference dictionary corresponding to the speaker thus verified. Then the speaker can either pass the voice commands related to the operation of block 3, or request the engine start.

It pronounces two acoustic forms ("to start" then, if this form is well recognized1 the "motor" form, but other keywords can be chosen), whose confidence threshold to reach is of the same order of magnitude as for the first keyword and these acoustic forms are compared only with the references of the common dictionary belonging to the speaker whose first keyword has been recognized.After a few attempts leading to a "non-recognition" of these two other words key, the system stops its recognition phase and returns to the initial position.The results showed that this principle of security was satisfactory.In the case of the recognition of the channe keywords, the engine starts by the means of an actuator ensuring the correct starting of an engine (start-o-matic, start-stop, etc.).

Then the vehicle can evolve normally, with the voice command.

To stop the engine, it suffices to pronounce two corresponding keywords, subject to certain conditions on the speed of the vehicle (this security already exists in the starter actuator).

The speech recognition system software performs a fixed acoustic pattern by eliminating those that are too short (less than 300 ms, ie short spurious noise), those that are too long (greater than 2s), those which are too weak or too strong in spectral energy and those having a bad start (it takes 100 ms of silence or a very weak energy before each acoustic form to be pronounced and 200 ms of silence at the end of acoustic form).

The use of this part can be very important for specific applications. If the aspect of simplification and security already exists for a normally constituted driver, it becomes essential and even necessary for applications with drivers having a form of manual impairment, or even a lack of manual control. If, under normal conditions, the manual function accepts a form of multiplexing for the control, this is not the case for the function performed by the foot (at most it allows to defring one or two analog values multiplexed as for example, control the brake pedal and acceleration) These physically disadvantaged cases then require a very rich means of communication, with a certain level of security, conditions fulfilled by the voice command.

A final point that emerges from the use of the voice command is its description state aspect. Indeed, to each entity or association of pronounced acoustic entities is linked an action. This action is, in a way, fairly well described by this acoustic image With the manual action, it often happens that, to define the state of a command, one needs a certain spatial repre- triggering a state transition analysis.To set the ideas, we can cite a typical example: often, when the external brightness in which the vehicle evolves leaves ambiguity on the lighting position of the headlights, it happens that the driver proceeds, switching traffic lights - crossing road, with an external visual inspection and interior possible, which allows him to determine if he was in the right position or not. To make the same command by the voice appears simpler because it does not need to know the previous state, even in the worst case, because it is only a confirmation of an already existing state.

In the case of an evolution of the vehicle with a multiplexing system, whatever the mode of organization of the management of the various electronic modules managing the functions of the vehicle, the fact of putting a voice control system connected to this network and doubling or not the manual control system has, by its principle, little influence on the total cost of the equipment already installed because its insertion is easy.

Block 4 of FIG. 1 represents the entire display and manual dialogue part with the general system. The link with the main block It is of the series type and parallel pork. This block consists of a pocket terminal and some control lights. This calculator-sized pocket terminal consists of 40 keys, of which a large part is multiplexed on 4 levels and a display part of 8 alphanumeric characters. This system makes it possible to broadcast messages or commands relating to the use. This block represents the voice-graphic communication but can be reduced to one or two buttons and some indicator lights. The broadcast messages contain information relating to the operation of the block 2 (words recognized or not, the different phases, etc ...) and to the overall condition of the vehicle. But too much visual information to monitor by the driver causes discomfort and a.

form of insecurity for driving the vehicle.

Block 6 represents a car radio with the adjacent interfaces of its speakers 5 and its antenna 7. The apparatus used is a PIONCER KE 5 300, most of which manual controls can be operated by voice. FIG. 4a shows the actual car radio 47, provided with its antenna 48 (or 7 of FIG. 1) and its interface with the electrical signals 40 coming from the main block 116, the power-up control 41, the selector 42 (FM, GO, PO), the control of one of the 5 stations programmed 439 the commands for programming and searching stations (Seek, Scan and memorizing) 44, the control of the volume, of the balance and the tone of the two tracks 45
Figure 4b describes the principle of volume control with a signal
BF 50, a potentiometer 51 actuates habitTeme2 / a cursor 522 a
a resistor switching system 53 for decreasing more or less the cursor signal, and a binary switch control signal 54 of the resistor switch 53. Then the output signal 55 drives a power stage which is connected to a high loudspeaker.

For balance and tone control, the principle is the same.

With a signal 54 of four bit weights, 16 possible values for the signal 55 are obtained. A dynamic of the curve is fixed with two limits. One of the limits will set the resistive ratio of the potentiometer (when the xedrive of the group 53 goes to infinity), the other limit will be taken into account for the safety (when the impedance of the group 53 is zero). This resistance ratio of the potentiometer fixed, one deduces the values of the different resistances taking into account the dynamics of the desired curve at the output 550
Two aspects emerge from the auto-radio function: control (increase or dimillution of the flywheel, change of ranges, etc ...) and programming (search for a particular station and memorization on one of the 5 programmable positions) . This storage lasts as long as the RAM portion of the car radio is powered, while having the OFF position of the start. Turning the car on or off (ON / OFF) is done by successively pronouncing two acoustic forms. A range change (FM, PO or GO) can be made by pronouncing an acoustic form (i ' auto-radio is already in operation) or two acoustic forms as if it were to start it although it is already. Each time the radio is switched on, the system programs an average value, by default, for the volume,
The driver can always increase or decrease the volume, change the tone and balance. To program a station on one of the Spositions, it pronounces two acoustic forms to indicate the "programming" mode. Then he pronounces the acoustic forms corresponding to the appropriate orders to stop the car radio on the desired station and memorize it on one of the Spositions.

In case of error, the dialogue can be done visually trace to the terminal), or by the synthesis described below. It is also possible to program any radio with access possibilities to send a series of values to select a particular frequency. If most important stations have well-known frequencies (often installed on the radio as pre-set keys), then this kind of data can be established in a table operated by the software. But the problem becomes more complex in the F.M. band because of the large number of stations that reside there and are broadcast over a small area. In general only a few interest the driver.

The possibility of associating the acoustic form of one's choice with the desired station and to obtain it thus avoiding a fastidious search suggests a certain approval.

The flexibility of programming by the vocal means depends on the decisional power of the central software. This power is related to the tolerance granted for the dialogue through block 2 and synthesis 12 (described below).

Block 8 is an on-board computer equipped with various COI sensors,
C02 ----, COn. The computer used may be the ADAC marketed by the Régie Nationale des Usines Renault or any other computer or similar onboard computer oriented in the direction of driving assistance.

The computer has two modes of operation. One is geared for information on time, outside temperature, speed, mileage and fuel consumption. The second concerns the programming of the route and delivers forecasting information given what has been programmed and data in progress.

FIG. 5 shows this calculator associated with temperature sensors 60, flow rate 61 and speed sensors 62. The actual calculator with its display system and its keyboard is designated 63. Most of the keys (23 in total) usually serve to the programming are replaced by a connection leading to the central unit or main block ll.

In operation, the time is permanently displayed and, following a voice command, the corresponding information is provided on the display device (consumption and speed La can be average or instantaneous, remaining capacity in liters of the tank, automatic , etc.).

The information is thus displayed for a certain time (determined by software) and then disappears to leave the time display again.

Thus, the driver, judging the opportune moment to look, vocally requests the information rather than doing either a search for a button among several, or a cyclic scanning of the button if it contains more information. This last case is realized, saros modify the principle of the calculator, by the software which supports this research, without the driver perceives it.

For the software, it is just a table of data, knowing the position of each piece of information in the cycle, for a given button, and their correspondence with the pronounced acoustic forms. If this aspect is already very interesting, that of the programming of the calculator becomes almost primordial. Indeed, the programming consists of giving information, by means of the keyboard, relating to the course that the driver will achieve. The route can be divided into sections (road, motorway, city) with the corresponding mileage, the estimated duration of crossing cities, the capacity of the tank, etc.

The outboard calculator is a fairly small size and the keys
Keyboard programming have the dimensions of those of a calculator.

to use
The results showed that many drivers refuse / the system
because of the programming difficulty (search for keys and com
prehension of the programming principle leading to many errors).

The fact of programming each function by an acoustic form or two,
pronounced while driving, with control by visual feedback
or acoustic (blocks 12 and 13) possibly, demonstrates an improvement
certain as to the operation of the on-board computer. Some information may even be implicitly known to the software (distance to
km between major cities, for example, where should the vehi go
cule).

Of course, the more programming of the onboard computer will be rich,
the more the number of acoustic forms to use will be important, but
in progress less fast. Indeed, the decision-making system organizes
the control or programming of certain functions by asking for pro
denominate one, two or more chained acoustic forms.
as for other functions, reduce the total number of shapes
acoustic and reduce the number of acoustic forms to compare for a given level of the tree If the first point offers a
interest (simplification, for the driver, of memorizing all
acoustic forms, faster response time of the recognition system
because there are fewer acoustic ambiguities, etc.,.), the second point
is still fine because, for a given tree level, we limit
the field of candidate acoustic forms while knowing their
past and their future, for certain levels, and from a base of
assumptions made in the CPU software 2.

Figure 6 illustrates an example to specify these two points
It represents two levels of tree 90 and 91, the entities cli
gnoting (92), climb (93), descent (94) and stop (95) for 90 and right
196) and left (97) for 91. The entities 92,93,94,96 and 97 allow,
in two steps, to create the functions flashing-right (92-96), blink
left-hand (92-97), right-hand climb (93-96), left-hand climb (93-97),
cente-right (94-96) and descent-left (94-97). We realize 6 functions
using 5 chained acoustic entities. The stop entity (95) loops back to the same level because it announces an inhibitory function specified by the second entity. Level 91 is not required in this case.

Indeed, if the driver has put his direction indicator in one of the positions, and he wants to stop it he will pronounce 95-92. By cons, to stop a side window, the driver must specify the left or right, and therefore pronounce a third entity. Functions performed by pronouncing two acoustic entities separated by a silence can also be realized by pronouncing a single entity, thereby creating a redundancy on the starting tree level.

This creates an increase in the vocabulary of reference but leaves a little more flexibility to the driver by reducing the constraints This notion disappears if the software of the block 2 uses a recognition algorithm on the connected words (software MOZART on the system of the company VECSYS) . A word detection algorithm in sentences offers a lot of interest for the programming phase
This tree structure is based on decisions taken on the validity of the first entity but security, within a given limit, can be joined, requiring a visual or aacoustic dialogue in addition.For this, we set two thresholds for the rates. confidence of acoustic forms a threshold where all lower acoustic forms are refused, another where all higher acoustic forms are accepted. Between the two thresholds, the driver is asked, by visual or acoustic way, if the acoustic form that he has pronounced is this one. The driver is then asked for an affirmative or negative answer, and the recognition is made on two acoustic forms with a threshold a little stronger. We have a very good recognition to distinguish one of the two answers. Cases unrecognized but close to the threshold imply asking the driver the question. For the rest of the cases, one avoids this loop resting the question while waiting for a new acoustic form for the answer. This waiting is limited by a temporal security (realized at the level of the central software) in order to avoid the blocking of the system. In this case the driver will have to repeat (or not if he wishes to suspend his order) the acoustic form which created the ambiguity by being between the two thresholds for the confidence rate
The sound dialogue (return by synthesis) seems more interesting because
it disturbs the driver less when he drives his vehicle. This
principle is not systematically applied to all forms
but only on those corresponding to the functions im
load bearing (programming, stopping the engine, etc ...).

 Block 9 represents the interface of all the other sensors C1, C2, --- whose system needs to give information to the driver.

These sensors are complementary to those already connected to the calculator
The interface, in conjunction with block 11, consists of a
formatted part and signal processing, of conventional type, and a
data buffer with control signals. One of the con signals
control triggers a cyclical interruption so that block 11
data acquisition. But we can very well group all the
COI, C02, ---, COn and Cl, C2 Cn sensors on the same module in order to
see a better integration of the system. The system uses heading
providing information on engine speed (number of revolutions /
mn), the engine temperature, the oil, but this list may well
heard to be more or less important.

Block 10 represents all system resources exploited by
block 11. These are the read only memories (containing the main software
pal, fixed data, etc ...) and memories of which a part
is in C-MOS technology to save information in
having a very low consumption. So we can cut the power
of the general system for a long time. Blocks 2,8 and 12 have
their own software and therefore their own memories. The memory part
live is used for the storage of the acoustic event and the dictionary of
references for block 2.For blocks 8 and 12, this memory serves
when passing programming parameters
Block 12 represents the synthesis module, associated with a high-bettor
13 to restore the sound message This module is used, like this
has already been mentioned before, in the sense of the man-machine dialogue
complete part, that is to say the part completing the voice communication of the
driver with his vehicle. Two strategies can be used.

 Either we limit the space of the possibilities of the communication of messages, but by using products of synthesis of good quality (method by LPC, by formants, etc ...), or we widen the horizon of the possibilities of the communication of messages. messages, but using a different method. For this last case one can, either proceed with a support of important data ~, or use a synthesis from the text. The advantage of using a synthesis from the text is its ease of programming (set of alphabetic characters of the language provided with some grammatical and prosodic markers). The disadvantage is the very average quality effected to the listening of the driver.

The products that can be used are ICOLOG from the company VECSYS or SPARTE CNET The architecture is similar and an illustration is given in Figure 7. The signal 110 represents the string of characters, from the text that you want transmit in sound form, provided with its punctuation parameters. This signal, coming from block 11, is processed in text spelled by block 111. This text is then translated into a phonetic and prosodic string by a block 112. From this chain, a list 113 control parameters that are sent to a block 114, which generates an analog signal. This analog signal is then filtered, smoothed and amplified using a module 115 for providing the sound wave through a speaker 116 (or 13 in Figure 1). Both of the systems mentioned above also have grammar rules acting at the level of the module 112. This software is also called grapheme-phoneme conversion and its importance depends on the language to be treated (language whose spelling is not or a lot of phonetics). This software solves problems such as "chickens convent convent" or "president presidents".

Both strategies were tried and the low importance of the quantity of messages to be announced allowed the use of the synthesis by L.P.C.

because of its quality. The S.D.P. of C.N.E.T. offers a very interesting tax advance insofar as this circuit can be used at the same time with data of synthesis by L.P.C. and data from the text. Examples of announced messages are "What is your code?", "Repeat, please", "Did you say XXX?", XXX being the acoustic form on which the ambiguity resides, etc.

Figure 8 shows the architecture of the CPU II of Figure l.

This unit consists of a clock portion 120 used for time delays, timebases of serial communications and processor, part of R.O.M. 121, part of R.A.M. 122, a microprocessor 123, a processing circuit 124 interrupts 125 from the sensors, the clock, etc ..., a serial communication system 126 with a serial interface 127 for the handheld terminal 128 (or 4 of FIG. 1), a serial interface 129 with the speech recognition system 130 (or 2 of FIG. 1), a serial interface 131 with the speech synthesis system 132 (or 12 of the FIG. 1), a parallel port communication system 140 with an entire parallel port expansion system at output 133 (described in detail below) allowing the connection with the car radio 134 (or FIG. 1), with the actuating device 135 (or 3 of FIG. 1) and with the on-board computer 136 (or 8 of FIG. 1).

Similarly, 137 is an input port expansion system with signals from sensors 138 (or CI, C2, ---, Cn of block 9 of FIG. 1) and contact indicatess 139 (for example, wear brake pads, badly closed doors, automatic or manual mode, etc ...).

FIG. 9 shows in more detail block 133 of FIG. 8. This block 133 makes it possible to expand two parallel 8-bit output ports 71 and 72 (programmed from the processor 123 of FIG. 8 by the link 70 ) into seven parallel output ports of 8 bits, represented by blocks 79 to 85. Chaeun of these blocks 79 to 85 consists of two parts. One part is an 8-space shift register to transform a serial message in parallel, while the other part is an 8-cell port for copying the data from the shift register. All outputs are followed by three-state amplifiers.

The signal 73 represents the data common to all the blocks 79 to 85. Seven validation signals 74 inverted at 75 make it possible to select the block in which it is desired to carry out the parallel transfer of the data coming from the shift register. Seven clock signals 76 make it possible to select the block into which the 8-bit data is to be entered in the shift register. A signal 77 inverted at 78 and common to all the blocks allows the output validation (three-state logic). These output signals drive interfaces 86 for the radio 6.87 for the actuating device 3 and 88 for the on-board computer 8.

FIG. 10 represents the diagram of the logic signals that enter the blocks 79 to 85 of FIG. 9. The signal 100 (or 73 of FIG. 9) represents the 8-bit data which will evolve in series from D1 T to D8. This group 107 is called QNT, each Di (D1 to D8) becoming
T
Qi when he is out. The signal 101 is that of clock allowing a series shift of a box in the register at each positive transition (indicated by the arrows). Thus, the binary value present on the data wire is recorded. The reference 102 designates the QT-1 group which is the output data, generated at the time
N from before. This output data becomes Q (or 103) when the enable signal 105 makes a negative transition. The signal 106, which
T is the output validation, transforms the QN group (103) into a high impedance state ZN (104) This high impedance state is generated on the low state of the signal 1060 All control is by a logic level "O" and all wires connected to the smB commands polarized by resistors attached to the positive terminal of the power supply In case of wire breakage or destruction of an output (in addition to the existing software security, ie say the level "1" imposed for the output or the high impedance state possibly), the command is not carried out.

The block 137 of FIG. 8 obeys the same organization structure as the block 133, but i3: operates as input (data read).

Figure 11 shows the main flowchart of the software of the central unit 11 managing the entire system. When the system is turned on
150 by means of the ignition key, it runs the routine 151 which is the initialization at power on. This ensures the smooth running of the software (initialization of the stack pointer, registers, interrupts, etc ...) as well as the good initialization of the peripheral systems (generally the peripheral systems already having a clean software containing a routine If this is not the case, the central software will redo the initialization by programming.) Then routine 152 organizes all the initialization of the hardware of block Il (FIG. 1), namely time bases, links series and parallel as well as their programming in security mode, the system of interruptions, etc. Then, the modes of connection thus assured, one initializes all the peripheral systems with the block 153. One tests first to know if each peripheral system responds correctly by its previously established link, before sending it the initialization parameters, For the recognition system of the paro the 2, we check its initialization and stil if it is waiting for an order (recognition, learning, etc ...).

For the hand-held terminal 4, after the test of the link, the window of 8 characters, from 32, is positioned on the right, with a refreshing of the R.A.M. and a cursor position and a short sound signal. For the on-board computer, it is ensured that there has been no loss of its parameters. Indeed, the calculator is defined for a whole range of vehicles and requires characterization parameters of the vehicle in which it is installed (engine flow curves, mounted wheel diameter, manual or automatic gearbox, etc ...).

Now this computer is always powered by the battery, having a very low consumption, even if the general system is not itself powered or if the vehicle does not circulate In the case of a loss of parameters, it resets the computer and, in the case of a new answer synonymous with error, one tests to see if one of its sensors is deficient or detect any other anomaly. If the information persists, we abandon this routine and the display flashes, indicating an anomaly Moreover, whatever the case, we display the time (even erroneous) by default. If there has been a reset, the driver will have to reprogram certain parameters (the time, the tank capacity, the mileage of the sections, etc ...).

The speech synthesis system 2 is programmed to be in "silence" mode (analog signal leaving the block 115 of FIG. 7 having a zero amplitude).

Once this phase 153 has been completed, phase 154 of initializing block 11 and the parameters is executed. Some parameters are specific to block 11 (zeroing of the counters for timers, security for secondary functions, etc.), others are sent to peripheral systems, such as, for example, the generation of a device. sound signal by the handheld terminal 4 to signal that the general system is ready, or the need for the driver to load his own vocabulary of references by the keyboard of the handheld terminal, or to go directly through the voice command to load vocabulary and start the vehicle.

In the latter case, the speech recognition system 2 is programmed to receive the reference dictionary common to the various users of the vehicle, to test the correct loading of this vocabulary, to send the recognition threshold and to program the system in recognition mode. Then, if the driver satisfies the conditions of this voice security, the block It programs the recognition system to load the vocabulary of use of the driver and then goes into the recognition phase. This mode, which can be called automatic mode, goes directly from phase 154 to phase 159.

The flowchart of Figure ii represents the other mode, called the manual mode. The difference between these two modes can be made, for example, by the positioning of a switch read at power up (read through block 137, Figure 8). For normal use, it is the automatic mode that operates, but having done at least once learning to create the dictionary references. To complete this phase 154, one can use either the pocket terminal to indicate that there is no anomaly (display + sound signal), or use the synthesis to send a message in the same direction (for example " ready system "," hello "," pronounce your code1 ?, etc ...).

Once this phase is completed, phase 155 is executed, where the driver is asked, by means of the display of the handheld terminal 4, to send a character on the keyboard in order to select a mode. These different modes are, by entering a character on the keyboard 157, A for learning 158, R for recognition 159, M for help in understanding the system 160, S for saving a vocabulary 161, C for the loading of a vocabulary 162, M for the modification 163, P for the parameters 164. Any other character is forbidden and, in this case, an error message appears on the pocket terminal. This routine is represented by phase 165.

The transition from phase 151 to phase 155 represents the general initialization 156 of the system and appears at each power up (by means of the ignition key). The transition from phase 157 to one of the modes, then the return, represents the main loop 1660 This return is made when one of the modes ends. However, the recognition mode 159 can be looped on itself and terminate either by cutting off the power of the system with the ignition key, or by pronouncing a particular acoustic form.

The understanding mode of the system 160 allows, through the display on the pocket terminal 4, to provide all the information useful for the operation of the general system. This mode, making use of a user manual, is not mandatory in the context of a current application.

The parameter mode 164 makes it possible, by means of the display-keyboard dialogue on the handheld terminal, to read or write values specific to the operation of the system. For example, the recognition rate can be known.

birth (number of recognized or unrecognized words) after a period of use, bring out the acoustic forms that are more or less well recognized, which allows to decide to change one acoustic form by another to improve performance of the system,
load data, etc ... This mode, too, does not have a character
mandatory but is very useful for establishing results of experiments,
statistical codes according to various criteria (time of use,
urban, road or highway, quality of microphones, etc ...).

Figure 12 describes the learning mode. The phase 170 represents the passage of parameters (pointers for the symbolic command chain of the speech recognition system) as well as the test to know if the learning that one wishes to execute will realize a new vocabulary creation, it that is, if the RAM of the recognition system contains a reference vocabulary or not
(It is thus determined whether a creation or a change of a vocabulary of references has already been made before.This makes it possible to avoid false maneuvers.In the case of a non-presence of vocabulary, one does not execute by the phase 171). Phase 171, through the pocket ter minal screen, raises the question of whether or not to delete the vocabulary present in the recognition system. The answer can be either positive (overwriting the vocabulary), or negative (we leave the learning mode), or an error indication and in this case we re-execute the phase 171.

Phase 172, via the keyboard, asks the driver-the total number of words that he wants to constitute for this vocabulary as well as the number of learning passes. With this information, phase 173 realizes the programming the recognition system 2 and launch Ba command to start it At each acoustic form that the driver must pronounce, we associate a graphical symbolic reference that appears on the screen of the terminal This is a very useful help and this graphic form is in fact the writing of the acoustic form which is much more preferable than a simple number At phase 174, the central software sends the graphic form to the handheld terminal and is waiting for the response of the recognition system making the acquisition of the signal acoustic When this acquisition is ensured according to the conditions of the recognition system 2 algorithm, the driver receives a response indicating whether there has been a problem th, (acoustic form too short or too long, with too much or not enough.

of energy, with a bad start, etc.,.,) o The read-and-test are executed at the phase 175 which loops back with the phase 174 in the case of an error. Otherwise, we go on to the next acoustic form and, in phase 176, we test if we have already recorded the last acoustic form.

In the negative case, we loop back to phase 174 Otherwise we go to phase i77 which tests if we made the second learning curve (a learning curve consists in pronouncing acoustic forms from the first to the last word constituting the vocabulary). We go back to phase 174 until we have completed the last learning pass. Phase 178 tests whether any anomaly in programming, data transmission with the recognition system or an inherent error in the recognition system has occurred. In this case, the phase 179 generates an error message, visual and sound, on the terminal and the training routine is abandoned by reinitializing, in advance, the recognition system 2. In the case of normal operation, it It is indicated that a vocabulary creation has been carried out and parameters used by the other subprograms presented by phase 180 are passed, and then the training routine is terminated.

This flowchart describes that used with the V.R.M. I.E.C. The one used with the MOISE system, from the company VECSYS, differs in the programming mode, the nature of the errors and the test on the number of learning passes (indeed the latter system has an algorithm working on a single pass of 'learning).

Figure 13 shows the flowchart for loading or saving a reference vocabulary. The backup consists in reading the information characterizing the acoustic forms in the recognition system 2 and then writing them in the central memory 10. The loading mode operates in the other direction and therefore the flowchart of FIG. 13 corresponds to the two modes , with small differences. Phase 190 represents the passing of parameters and the preparation of data and indexes for the tables. This makes it possible, knowing the structure of the characterization data of an acoustic reference, to arrange them in the right order and to reduce the occupation size of the central memory. The phase 191 acquires the number of the file, either to load, or in which one wants to make the safeguard. The phase 192 tests if in the case of a loading, one has the file corresponding to the existing number and if the system Recognition does not yet have a vocabulary in its memory, or if, in the case of a backup, we will not overwrite a file already present; in this case, the user is asked the question. The phase 193 is triggered, in case of error following the tests of the phase 192, and one sends an adequate message of error on the terminal before giving up the routine. Phases 191-192 are not executed when the software evolves with the recognition routine (phase 159) according to particular conditions. For example, when the driver turns on the system, or after he has pronounced his code to start the program. car, or after decisions have been made at the level of the tree linking the acoustic forms, etc ..., the algorithm, which has tables
of data, retrieves or stores vocabularies of acoustic references, or parts of vocabularies of acoustic references, in the central memory. Phase 192 executed normally, we go to phase 194 consisting of programming the recognition system 2, to transmit the command and to process the information as and when circulating. Then phase 195 tests whether a programming or information transmission error or an error specific to the recognition system has taken place. Certain errors lead to the rerun of phase 194 but this loop is limited to a few tests in order to avoid a blocking. In this case, and with the other types of errors, an error message is sent and the recognition system is reset before abandoning the routine In the case of normal operation, the parameters are passed. and the data specific to the mode used (phase 197) and the subroutine is finished
FIG. 14 represents the flowchart of the modification mode 163 of FIG. 11.This mode, having passed the parameters to the phase 200, makes it possible to modify or add (test made at the phase 201) acoustic references (with their graphic symbols associated with) the vocabulary residing in the recognition system 2 (which can then be saved). In the case of a modification, phase 202 is tested whether it is symbolic or acoustic. The symbolic modification is made in phase 203. For the acoustic modification, the information necessary to know the acoustic forms to be modified (phase 204) is acquired. If the -201 test results in the addition mode (or additional creation to an already existing vocabulary), we execute the phase 205 where we acquire the number of symbolic forms.

In phase 206, symbolic creation and functional attribution are realized (that is, we associate an acoustic form with an action or other acoustic forms for an action knowing that we freeze starting from the space of the actions described by tables of data.Of course, these tables are closely related to the application that one wants to realize). The phases 204 and 206 lead to the phase 207 where it acquires the number of learning passes (Phase Deleted with the MOISE recognition system) o The phase 208 carries out all the programming of the recognition system 2 according to the course which has been done before in the flowchart and starts the command. The phase 209 makes it possible to display the symbolic form associated with the acoustic form on the pocket terminal and is waiting for the acquisition of the signal via the recognition system.

Phase 210 tests whether an error has occurred and loops back to phase 209 in the positive case. The phases 211 and 212 successively test the passage of the last word and the second learning pass.

Phase 214 performs a reset if a programming, transmission, or recognition system-specific error is detected in phase 213. Phase 215 passes all parameters and data to complete the routine.

FIGS. 15a and 15b together represent the flow chart of the recognition mode 159 of FIG. 11. phase 220 makes it possible to pass the parameters and the data (manual mode or automatic nature of the recognition system, etc.). Phase 221 tests whether the general system is operating in automatic mode (after power-on using the ignition key and all the initialization and safety procedures that have been carried out, we go directly to the recognition mode) or in manual mode (manual manipulations of the driver's share with the keypad of the handheld terminal). In the latter case, phase 222 is executed where it acquires the reference vocabulary number, the value of the confidence threshold, the whole or partial tree, etc.

If the driver goes into recognition mode without having previously loaded the recognition system with a vocabulary of references, then we execute the phase 245 representing the sending of an error message (visual and audible) on the pocket terminal and a reset of the recognition system before abandoning the recognizable routine.

ciency. In the case of a good data acquisition, we go to phase 232.

If the test 221 indicates that one evolves in automatic mode, one loads the system of recognition of the vocabulary common to the users and containing the key acoustic references, used as means of safety. Phase 224 then sends information about the smooth running of the previous operations to the handheld terminal. The synthesis system 12 delivers a message inviting the driver to pronounce his code in order to load the appropriate vocabulary. In the same period of time, the reconnaissance system 2 is programmed in recognition mode on this common vocabulary and the command is issued. Phase 225 is waiting for the acquisition of the acoustic signal. Phase 226, in the event of an error, loops back to phase 225 (the same type of error as during the learning or modification phases) .These types of errors arrive as much to the good users as to the possible intruders. ).

In the normal case (acoustic form pronounced according to the criteria of the test 226), one executes the phase 227 carrying out the tests of threshold of confidence, of validity of word and chaPnage. If the confidence threshold is not reached (therefore less than that programmed) or a word having reached or exceeded this threshold is not valid, the phase 228 called accounting is executed. Finally, the third criterion indicates the position of the acoustic form in the tree structure of the voice command string used as initial security. In phase 228, we keep updated the number of acoustic forms pronounced since the passage of phase 221 in automatic mode as well as the scores of the thresholds achieved. Phase 229 tests If, in these accounting tables, we have crossed or not the safety threshold For the first criterion, beyond a certain number of tests (ranging from about 5 to 10), phase 230 is performed which resets the recognition system and repositions the general system waiting for a mode principal (1572 figure 11) in the opposite case, it is examined whether the acoustic form having exceeded the confidence threshold corresponds to the level of the tree structure.

It may be, for example, that the driver has pronounced the first acoustic form (level 1 of the tree) and continues to pronounce it, having forgotten to look at the indication of good recognition with the handheld terminal, while the The general system is waiting for other acoustic forms corresponding to level 2 of the arborecence.

In the normal case of an affirmative answer to the tests 227, og executes the phase 231 which loads the vocabulary of the references corresponding to the information contained in the chargeable account. At the same time, we send visual and audio information indicating that the security has been crossed (with the pocket terminal and the syn thesis system). The correct vocabulary is loaded, the phase 232 programs the recognition system and starts the command. This phase is also performed in manual mode, after executing phase 222.

In phase 232, we update other accounting tables specific to the progress of the rest of the program (structure of the tree structure, number of candidate acoustic forms per tree level, different thresholds, etc ...)
Phase 233 is comparable to phase 225 (waiting for good acquisition of the acoustic signal). Phase 234 loops back with phase 233 in case of error. Once the signal acquired according to the criteria of phase 234, phase 235 is executed which tests the confidence threshold achieved, according to the nature of the acoustic form. Certain acoustic forms have their confidence threshold determined with respect to a given value. In the case where the threshold is not reached, we loop back to the phase 233, otherwise we continue in sequence. However, other acoustic forms have their confidence threshold determined with respect to two given values. In the case where one is smaller than the smaller of the two, one loops back with the phase 233. In the case where one is greater than the greater one of the two, one continues in sequence. In the third case, we ask the question of the validity of the acoustic form pronounced through the synthesis of speech (with a display on the pocket terminal). In this case, the driver must either confirm the validity (yes or no answer) or repeat the acoustic form that left the ambiguity. In both cases, we go back to phase 233. The first solution is more interesting. If the answer is negative, we stay at level 1 of the tree1 otherwise we continue in sequence leaving a trace in the accounting table.

The sequence continuation leads to the phase 236 where it tests, as the case may be, the validity of the word in the tree (same scenario with the procedure security at startup). In a case of invalidity, it is looped back to the phase 233. In the case of validity or positioning in a control chain, the phase 237 carries out the display of the associated symbolic shape on the pocket terminal. the accounting table As a result of the positioning of the acoustic form in the tree structure, phase 239 is executed or not to load partial vocabularies of references (for example, if the driver asks for the car radio, we will load the vocabulary by tiel of references containing the acoustic forms making it possible to choose a station, to increase or decrease the sound, to change ranges, etc ...). Of course we also remove partial vocabularies of references (for example, if the driver asks to stop his car radio).

Phase 240, according to the information contained in the accounting table, tests whether the action, associated with the acoustic form or the string of acoustic forms, is valid or noh. In the case of non-validity, we go back to phase 233. If the action to be triggered is valid, we execute phase 241 which is a superior safety test
For example, to stop the engine, even if the driver complies with the voice command provided with security (dialogue with the summary), this action will only be performed below a certain speed of the vehicle (a few km / h). The only superior possibility that the driver has is to turn off the ignition with the key.

In phase 241, actions and inhibitions of actions are executed using data tables. These tables are organized with electrical controls to pass, equipped with security reproducing those usually existing. There is therefore, in addition to electronic security contained in block 3 of Figure 1 (or more precisely the block 32 of Figure 3), the same security in the software. The action (or inhibition) being carried out in phase 241, phase 242 is executed. Phase 242 makes it possible to stop the recognition mode by a voice command. If the test is negative, it loops back with phase 233. In the opposite case, it is ensured, by the synthesis, that the driver actually wants to exit the recognition program, in automatic mode. In manual mode, the driver acts by means of a key or a special code entered by the keypad of the pocket terminal.

Phase 243 resets the recognition system. Phase 244 is used to pass parameters and recognition mode data before dropping the subroutine. It is also possible, in phase 242, to leave the general system in a standby state, ie to remain in recognition mode which will be reactivated, either by pronouncing a particular acoustic form or by actuating a fugitive contact.

FIG. 16 represents the subroutine used to manage the block 133 of FIG. 8 which is detailed in FIGS. 9 and 10. In FIG. 10, the clock signal 101 is at rest in state "1" and performs the shift of the register on a positive transition while the enable signal 105 is at rest in the state "0" and performs the transfer in parallel on the outputs on a negative voltage. As a result, it suffices to pass 2 values to the subroutine, V1 (8-bit data to be output on one of the blocks 79 to 85) and V2 (clock-enable state to select one of the blocks 79 to 85) .

V1 is represented as follows:
D8 D7 D6 D5 D4 D3 D2 D1
bit 8 bit 7 bit 6 bit 5 bit 4 bit 3 bit 2 bit 1 - V2 is represented as follows
State
high impedance block 79 block 80 block 81 block 82 block 83 block 84 block 85

<tb> bit <SEP> 8 <SEP> bit <SEP> 7 <SEP> bit <SEP> 6 <SEP> bit <SEP> 5 <SEP> bit <SEP> 4 <SEP> bit <SEP> 3 <SEP > bit <SEP> 2 <SEP> bit <SEP> I <SEP>
<tb><SEP> clock <SEP> or <SEP> validation
<Tb>
Phase 250 initializes the counter C to 8 to send in series the 8 bits of the value V1. Phase 251 makes a temporary backup of
Phase 252 isolates bit 8 from the value V1. The phase 253 compares, using the value V2, the state "0" of the clock input of the block to be selected (79 to 85) with the bit 8 of VI, then sends this data (block 254) on the link 70 then 7i: :( fig 9). Then, without affecting the bit-8 to have its stable state output, it passes the clock input of the selected block to the state "1" (transition to the state 1 bits 7 to I of the data ) to enter the bit of V1 in the selected block (79 to 85) and to shift the register (phase 255> .The value VI (phase 256) is then recovered and a shift to the left of a box (the value of bit 1 passes in bit 2, the value of bit 2 in bit 3, etc ..., the value of bit 8 in bit 1) and it is tested in 258 if the counter C If C is different from 0, we loop back to phase 252, which sends the value of V1 (D81 then D7, ..., and D1) If C is 0, we execute phase 259 which compares the bit 8 with the normal state (opposite of the high impedance state) with the inverted V2 state at 75 (transition to the state 1 of the validation signal for the selected block where the value VI has been sent) Then, the transfer (phase 260) is carried out in parallel Releases the value Vi on the outputs by changing to state "O" on. signal validation of the selected block (transition to the "O" state of the bits 7 to 1 of the data sent in 70 then 72, Figure 9) .The routine is then complete
Of course many modifications can be made to the example described without departing from the scope of the invention Ceest and, for example, as indicated above, the pocket terminal can be replaced by any device for introducing information ( for example choice of operating mode) or to display information (word recognized or not, signal / noise ratio, etc.) Regarding the word information recognized or not, it should be specified that, preferably, the system ensures in all cases a function "acknowledgment of receipt, that is to say it indicates by any means sound (speech synthesis), or visual (graphic or symbolic display or indicator lights) the position that has taken relative : the acoustic form emitted, ie word recognized word not recognized "where" word rejected (not giving rise to any recognition phase for the reasons indicated above: word too short, too long, etc ...).

Claims (8)

A system for communicating, in a motor vehicle, complex information relating to the vehicle and its environment, characterized in that it comprises an acoustic information receiver such as a microphone (1), a device for speech recognition (2) coupled to the receiver, a speech synthesis device (12) coupled to a loudspeaker (13), a device actuating device (3) (A) A2, ... An) of the vehicle, a set of sensors (C1, C2 of the heaters connected to ufl interface module (9) and a central unit (11) coupled to the speech recognition device (2), to the speech synthesis device (12), the actuating device (3) and the interfacing module (9) and controlling the actuating device (3) and the speech synthesis device (12) in accordance with instructions transmitted and decoded by the speech recognition device (2). 2. System according to claim 1, characterized in that the actuating device (3) successively comprises an interface module (31) connected to the central unit (11), an electrical safety module (32), a an opto-electronic electrical decoupling circuit (33), an amplifier circuit (34) and a power circuit (35) for supplying said devices (A, A, Oo, Ai).  3. System according to any one of claims 1 and 2, characterized in that it comprises a terminal (4, 128) provided with a keyboard and a display screen controlled by the central unit (11> in Teaching function transmitted and decoded by the speech recognition device (2). 40 The system according to any one of claims 1 to 3, characterized in that it comprises an on-board computer (8, 63, 136) controlled by the central unit (11) according to instructions transmitted and decoded by the speech recognition device (2).  System according to any one of claims 1 to 4, characterized in that it comprises a radio (5-7, 40-48, 134) controlled by the central unit (11) as a function of instructions transmitted and decoded by the speech recognition device. System according to Claim 5, characterized in that the radio itself (47) is connected to the central unit (11) via a power-up control module (41). a frequency range selection module (42), a preprogrammed station selection module (43), a station programming and search module (44) and a volume control module , tone and balance (45). 7. System according to claim 6, wherein the car radio (6) comprises, for each of the volume, tone and balance controls, a manually operated potentiometer (51, 52), characterized in that this tool comprises in addition, in parallel with the output of the potentiometer (51, 52), a resistance switching system (53) controlled by a binary signal (54) generated by the volume, tone and balance control module (45). 8. System according to claims 1 to 5, characterized in that the central unit (here) comprises a microprocessor (123) connected to the speech recognition device (2, 130), to the speech synthesis device ( 12, 132) and to the terminal (4, 128) by a serial communication system (126) and serial interface modules (127, 129, 131), the actuating device (3, 135), the edge (8, 63, 136) and auto-radio (5-7, 40-48s 134) by a port / parallel communication system (i40) and an output parallel port expansion system. (133), and to the interface module (9) with the sensors (C1, C2, ... Cn) by said parallel port communication system (140) and an input parallel port expansion system ( 137), said central unit also comprising a clock (120) for determining the time bases of the serial communications and the microprocessor (123), an integrated processing circuit (124). ruptures (125) from among others sensors (C1, C. C) and the clock (120), a read-only memory -  n 2, n 121) containing the microprocessor software (i23) and fixed data and a random access memory (10, 120) for storing variable data. 9. System according to claim 8, characterized in that the parallel output port expansion system (133) comprises n blocks (79 85) each comprising a shift register of n + 1 boxes for the conversion of a signal. serial data into a parallel data signal and a port of n + 1 cells1 for copying the data of the shift register, whose outputs are followed by three-state amplifiers, a first circuit (71) controlled by the microprocessor (123) and transmitting on the one hand a series data signal of n + 1 bits (73, 100) to all blocks (79 - 85>, and on the other hand n enable signals (74) to select the block in which to be performed the parallel transfer of the serial data signal previously entered in the shift register of this block, and a second circuit (72) controlled by the microprocessor (123) and generating on the one hand an n-bit clock signal ( 76) to select said block in the shift register of which the serial data signal of n + 1 bits is to be written and, on the other hand, an output enable signal (77, 106) applied to all the blocks (79 - 85) for the application of the parallel data at the output of the block selected at one of the interfaces (86, 87, 88) of the car radio (6), the actuating device (3) and the on-board computer (8). 10. System according to any one of claims 1 to 9, characterized in that it comprises sound and / or visual means for indicating whether an acoustic form emitted has been recognized, unrecognized or rejected. 11. A method of operating the system according to any one of claims 1 to 10, characterized in that, following the electrical power of the system is selected the operation of the system in automatic or manual mode1 in that, in both systems, the system provides a set of initializations in a given sequence, in that the system then goes directly into speech recognition mode when the automatic mode has been selected and issues a query as to the mode desired operation when the  Manual diet was selected, and that, in response to the said  interrogation is selected by means of said terminal (4) said  desired mode from among several modes of operation comprising  a "learning" mode (158) for the introduction of a diction  number of acoustic references in the system and the mode "recon  birth of speech "(159) according to which the system compares  acoustic forms pronounced by a speaker to a diction  reference number previously stored.  Process according to Claim 11, characterized in that the  operating modes that can be selected in manual mode  also include  - a 'steep way to understand the system' (160) next  which the system delivers, via the terminal (4),  useful information for its operation  a "backup" mode (161) of reading information  characterizing acoustic forms stored in a me  of the speech recognition device (2) and to  write them in the random access memory (10, 120) of the central unit (11); ;  a "loading" mode (162) of reading information  characterizing acoustic forms stored in memory  of the central unit (11) and to write them in a memory of the  speech recognition device (2)  a mode "modification" (163) allowing to modify references  Acoustic vocabulary stored in the device  speech recognition (2) or to add new  acoustic references to this vocabulary  a mode "parameters" (164) allowing, via the  terminal (4), to read or write values specific to ltex-  operation of the system; and  an "error" mode (165) delivering an error message in case of  wrong operation of the terminal (4) when selecting one  said modes of operation of the manual mode