FR2940497A1 - METHOD FOR CONTROLLING AN APPLICATION FROM A VOICE SIGNAL AND ASSOCIATED DEVICE FOR ITS IMPLEMENTATION - Google Patents

Abstract

The present invention relates to a method of controlling an application, for example a video game, by a voice signal (4) characterized in that it comprises the step of defining a control vocabulary, that is to say to say a set of control words, preferably onomatopoeia, each having a characteristic phoneme or several characteristic phonemes making it possible to distinguish each word from the other words of the control vocabulary. As soon as a control word contained in the voice signal (4) is detected, generating (21, 41) a specific action. This detection of the control word is carried out by means of at least one discrimination machine (38) making it possible to detect and discriminate the pronunciation of different phonemes while possibly excluding a set of other phonemes, this discrimination machine (38) assuring for example the distinction between at least two characteristic phonemes or the distinction between at least two characteristic phoneme groups. Then, the specific action is driven (17, 44, 45) as a function of at least one continuous magnitude (55) of the voice signal, such as duration, intensity, pitch, or timbre.

Description

Method for controlling an application from a voice signal and associated device for its implementation

The present invention relates to a method for controlling an application from a voice signal and the associated device for its implementation. The invention is intended in particular to increase the responsiveness of the application control, while allowing a reliable and accurate control.

The invention finds a particularly advantageous, but not exclusive, application for controlling actions in video game type applications from the voice. By action, we mean in particular moving, io jump-type actions, shooting, video game characters, and the game environment (scenery, music).

However, the invention could be used in other applications such as for the control of digital musical instruments, the control of an avatar, the control of a playful robot, domestic or military, or any other application capable of being controlled by a user, such as a mouse simulation software for disabled people or navigation in a database or in a list, such as the directory of a mobile phone.

Traditional voice control systems providing basic control by recognition of an entire word once the word has been spoken are known. We then have a discrete control, insofar as each word is recognized identically regardless of how it was pronounced. For example, the recognition of left or right words offers only two discrete choices: left or right and thus the generation of two discrete actions associated with these words. These systems necessarily wait for the end of the word before taking action.

The invention makes it possible to initiate an action corresponding to the detected word before the user has finished pronouncing the word in order to increase the responsiveness of the voice control method. For this purpose, the invention detects the presence of voices, then implements a recognition, not of the entirety, but of the first phonemes of the control words of the application, then makes it possible to control the application in function of the pronunciation of the continuation of the command word. In the case of multi-syllable control words, the detection / control process can be reiterated, so the control word contains several detection and control parts.

In the remainder of the document, a CONTROL word is understood to mean a word whose pronunciation makes it possible to control an application, by detection, then by control.

In the remainder of the document, DETECTION PHONEMES in a control word are understood to mean the phonemes used for the recognition of a part of a command word, making it possible to trigger an action in an application.

In the remainder of the document, PHONEMES DE PILOTAGE in a control word is understood to mean the phonemes used for the monitoring of characteristics of the voice making it possible to control an action in an application.

In the remainder of the document, the VOCABULAIRE DE COMMANDE is understood to mean all of the command words to be discriminated simultaneously in order to control an application.

The conventional implementation of detection in a command word makes use of voice recognition techniques known to those skilled in the art, based for example on Hidden Markov Models (or HMMs).

An optimized implementation of the detection in a so-called phoneme command word, consists in detecting at least one characteristic phoneme of each part to be detected in the control word, the first part to be detected being preferably located at the beginning of the word . This is done using voice analysis machines to discriminate these characteristic phonemes, called discrimination machine. For example, to detect Flip and Flap, we use a machine to discriminate the pronunciation of A and I.

In the remainder of the document, PHONEME CHARACTERISTIC of a word is understood to be the phoneme making it possible to distinguish it from other words in the control vocabulary.

However, in order to ensure the reliability of the detection of a characteristic phoneme, the discrimination machine must not produce false detection with one of the phonemes pronounced before the characteristic phonemes of the words to be detected. Thus, in our example, it must be ensured that the A / I machine does not produce false detection at the pronunciation of F or L.

In the remainder of the document, DISCRIMINATION MACHINE is understood to mean a voice analysis tool making it possible to detect and discriminate the pronunciation of different phonemes, while possibly excluding a set of other phonemes of the detection.

In addition, in order to enhance the detection reliability, it is possible to combine several discrimination machines in order to detect several phonemes characteristic of each control word. In this case, the detection is no longer a single, but a group of phonemes. It suffices simply to ensure that each discrimination machine does not produce false detection with each phoneme uttered before the characteristic phonemes of the set of words of the control vocabulary. This method is hereinafter referred to as the principle of non-detection of the previous phonemes of the control words.

This early recognition of the control word makes it possible to implement the second step of the invention, consisting in using the driving phonemes of the control word, that is to say the phonemes according to the detection phonemes, to perform additional controls of the application, called application control. This control is based on the prosodic parameters of the voice, such as, for example, the duration, the pitch, the energy or the timbre with which the driving phonemes of the control word are pronounced.

For example, the pronunciation of the left command word will have an influence on the control of the action, depending on whether it is brief (left) or longer (gauuuuuuche) depending on whether it is done with a deep voice, a high voice , or a voice whose pitch evolves, according as it is made with a loud voice, a soft voice, or a voice whose intensity evolves. These prosodic parameters make it possible to add additional voice controls, and can be applied to continuous values (such as a position or speed, for example), which is not possible with the discrete recognition of control words. These prosodic parameters also make it possible to drive a second action within a command word. Thus, the first phonemes (detection phonemes) are used to choose between two actions, and the following ones (control phonemes) are used to control the actions, the control being able to be reduced to a choice between two other actions.

More specifically, in the invention, a control vocabulary is chosen consisting of a set of control words each comprising a sequence of phonemes, a phoneme being considered as the smallest discrete or distinctive unit (that is to say to distinguish words from each other) that can be isolated by segmentation in the spoken channel. In the invention, the phoneme can correspond to a sound held or not, which has a height or not (for example the different vowels and consonants). The control vocabulary consists of a set of words, called control words, whose pronunciation is distinct, and which can therefore be reliably discriminated from one another, preferably before the end of their pronunciation. The control words chosen for the voice control are preferably onomatopoeia, that is to say short words, preferably monosyllabic, such as Paf, Fire, etc ... or longer words such as bazooka or A rifle having a characteristic phoneme detectable before the end of the word, in the case of a phoneme-optimized detection.

The voice control according to the invention has several steps. In a first step, the presence of any vocal sound is detected without being attached to its content, this detection being able to generate a generic action instantaneously. This step is preferably carried out even before the classification step defined below. However, this first step is not mandatory and could be omitted. Then, in a detection step, the system recognizes the detection phonemes of the command word for triggering an action, among the set of detection phonemes of the predefined command vocabulary for the control of the application. In the case of the phoneme-optimized detection, this detection is based on the recognition of at least one phoneme characteristic of the onomatopoeia using at least one discrimination machine. A specific action corresponding to the detected part of the control word is then engaged.

Several discrimination machines can be used in parallel, in order to improve the reliability of the detection, the initiation of a specific action then requiring the combination of several detection phonemes.

On the other hand, several discrimination machines can be used successively, so that a specific action can be taken when detecting each of the different detection phonemes. Once the system has recognized the detection portion of the control word and determined the specific action to be taken, the action is driven by a prosodic control over the pronunciation of the control word driving phonemes. This prosodic control makes it possible to control the specific action engaged as a function of a continuous magnitude extracted from the voice signal such as the duration, the intensity, the pitch or the timbre of the voice, each of these values being able to constitute an axis control of the action.

The specific action, and if appropriate the generic action, ends when the sound of the voice ends, or when a particular phoneme of the command word is recognized.

As a variant, the step of controlling the prosodic parameters is started if the piloting part is at the beginning of the control word, and a specific action is triggered when the detection part of the control word is recognized.

With the optimized method of detection by phoneme characteristic, based on the use of simple phonemes discrimination machines, the method according to the invention has an extremely low CPU consumption.

In addition, the discriminating machines used are very reliable so that by choosing characteristic phonemes far enough from a point of view of pronunciation, one obtains a system of great reliability.

Moreover, the method according to the invention is easily adaptable to several users. Indeed, phoneme recognition algorithms used by discrimination machines by default offer generic settings multi-speakers. In addition, to improve their reliability, these algorithms are easily adaptable to any new user by an extremely short learning: a simple recording of some 15 phonemes or each word.

The invention thus relates to a method for controlling an application, for example a video game, by a voice signal characterized in that it comprises the following steps: defining a control vocabulary, that is to say a set of control words, preferably onomatopoeia, each having a characteristic phoneme or several characteristic phonemes for distinguishing each word from the other words of the control vocabulary; - detecting a control word contained in the voice signal by means of at least one discrimination machine making it possible to detect and discriminate the pronunciation of different phonemes while possibly excluding a set of other phonemes, this discriminating machine ensuring, for example, the distinction between at least two characteristic phonemes or the distinction between at least two two groups of characteristic phonemes, 30 - generate a specific action corresponding to the command word de tected, then - drive the specific action generated according to at least one continuous magnitude of the voice signal, such as the duration, intensity, pitch or timbre of the signal.

According to one embodiment, the discrimination machine detects the phoneme 5 or the group of phonemes characteristic of the control word and generates the specific action before the end of the pronunciation of the command word.

According to one embodiment, the continuous magnitude of the voice signal according to which the action is driven is located in time after the detection phoneme detected by the discrimination machine.

According to one implementation, as soon as the presence of a voice signal is detected, even before the recognition of the command word, a generic action is generated.

According to one implementation, the specific action may be different from the generic action.

According to one embodiment, the method implementing at least two discrimination machines, several specific actions are engaged successively during the successive detection of phonemes or groups of characteristic phonemes in the control word by each of the machines.

According to one embodiment, two discriminating machines being used, a first specific action is generated after the detection of the first detection phoneme of the control word by a first discrimination machine able to discriminate N different phonemes, so that this a first action can be generated among N first possible actions, and - a second specific action is generated after the detection of the second characteristic phoneme of the control word by a second discrimination machine able to discriminate M distinct phonemes, so that this second action can be generated among N x M second possible actions.

According to one embodiment: - as soon as one detects a phoneme or a group of characteristic phonemes, or - as soon as one of the continuous parameters of the voice signal exceeds a certain threshold, for example as soon as the voice signal disappears, - the specific action ends. According to one implementation, 1 and 7 discrimination machines are implemented in order to be able to differentiate between 1 to 15 control words.

According to one embodiment, the discrimination machine uses the hidden Markov models.

According to one embodiment, it is decided to generate actions and to drive synchronously with the video frames of a game to be piloted.

According to one embodiment, the method according to the invention is implemented in combination with a control method by another control interface, such as for example a game pad, a keyboard, an eye tracking device, or a device. motion tracking.

According to one embodiment, to withstand noisy environments, a contact microphone is used.

The invention further relates to a device for controlling an application, for example a video game, by a voice signal characterized in that: - a control vocabulary, that is to say a set of words of command, preferably onomatopoeia, being defined, each control word having a characteristic phoneme or several characteristic phonemes making it possible to distinguish each word from the other words of the command vocabulary, it comprises: means for detecting a command word contained in the voice signal, these means comprising at least one discrimination machine making it possible to detect and discriminate the pronunciation of different phonemes while possibly excluding a set of other phonemes, this discrimination machine ensuring for example the distinction between at least two phonemes characteristics or the distinction between at least two groups of characteristic phonemes, - means for generating an act specific ion corresponding to the detected control word, and - means for controlling the specific action generated as a function of at least one continuous magnitude of the voice signal, such as duration, intensity, pitch or timbre of the signal .

The invention will be better understood on reading the description which follows and on examining the figures which accompany it. These figures are given for illustrative but not limiting of the invention. They show :

Figure 1: a schematic representation of the various modules of the voice control system according to the invention for controlling an action in a video game;

2: tables showing examples of detection phonemes and control phonemes used in the method according to the invention for discriminating control words and controlling the corresponding actions;

Figures 3a, 3b, 3c: Tables showing the different discriminating machines that can be used in the method according to the invention for detecting the phonemes characteristic of several pairs of control words, in the case of a detection optimized by characteristic phoneme ;

Figure 3d: a table comparing conventional speech recognition and phoneme-optimized detection on an example;

FIG. 4: a schematic representation of a system according to the invention enabling the control of an action in a video game from the words bazooka or rifle, in the case of an optimized detection with the characteristic phonemes [y] and [ u].

Identical elements retain the same reference from one figure to another.

Figure 1 shows a schematic representation of a voice control system 1 according to the invention for controlling an action in a video game 3 from a voice signal 4.

For this purpose, the system 1 comprises a microphone 6 connected to the input of an analysis module 9 via an analog / digital converter 7. The microphone 6 may be a conventional overhead microphone or any other voice pickup device, particularly a microphone designed to be unresponsive to ambient noise such as a contact microphone.

The module 9 able to extract parameters from the voice signal 4 is connected at the input of a module 10 called a mapping module in relation to the inputs of the video game 3 to be controlled, the mapping module 10 ensuring the correspondence between the parameters of the voice signal 4 extracts and the steering of the action in the game 3 video.

More specifically, the analysis module 9 comprises a voice presence detection module 13 capable of detecting the presence or absence of the voice signal 4, regardless of its content.

The analysis module 9 also comprises a classification module 20 able to detect a control word contained in the voice signal. This control word, for example an onomatopoeia of the type pif, paf, has a characteristic phoneme making it possible to distinguish it from the other predefined command words. The characteristic phoneme-optimized detection uses discrimination machines to detect these characteristic phonemes (see Figures 3a-3d).

For this purpose, the module 15 detects the detection part of the control word. In the case of detection optimized by characteristic phoneme, the module 15 comprises at least one discrimination machine or a combination of several discrimination machines each for discriminating at least two phonemes. The combination of several discrimination machines makes it possible in particular to use a concatenation of characteristic phonemes, rather than a single characteristic phoneme.

Thus, the classification carried out by the module 15 can operate on the following elements: elements such as phonemes (for example A, P,...); Combinations of elements, such as sets of phonemes; not temporally ordered (eg [P + A] or [C + L + A]), io - combinations of elements such as temporally ordered phonemes (eg [gau, che, bazoo, ka), - any combination simple and complex elements.

In the case of detection by conventional speech recognition, the phoneme recognition machines are known to those skilled in the art and in particular used in the field of speech recognition. These machines can be made from algorithms such as, for example, statistical and probabilistic models of the HMM (Hidden Markov Model) type, distance models, etc. Such a voice recognition system corresponds to an example of a discrimination machine. .

In the case of the characteristic phoneme-optimized detection, in one example, the discrimination machines are chosen from a vowel discrimination machine A / I making it possible to differentiate the control words at a from the control words at i, a machine of 25 vowels A / 1 / 0U for differentiating the command words in a, control words in i, and control words in or, a discrimination machine P / K consonants to differentiate the control words in p control words in k, a machine of consonants F, CH, SSS / vowels to differentiate the control words in f, ch, sss from any vowel, a machine P, T, K / B , D, G for differentiating the control words in p, t or k with respect to the control words in b, d or g. However any other type of discriminating machine is possible, insofar as it respects the principle of non-detection of the previous phonemes of the command words.

In a simple method of word recognition, only one discrimination machine is used to detect control words varying only by the only phonemes detectable by the discrimination machine chosen. Thus, using a NI vowel discrimination machine, it will be possible to distinguish, for example, the Flip and Flap control words, the machine preferably respecting the principle of non-detection of the previous phonemes of the control words, that is, that is, it does not produce false detection at the pronunciation of F and L. By using a discrimination machine P / K, it will be possible to distinguish the command words Pan and Clac, etc.

For more complex control word detection, several discrimination machines can be used successively, so that a specific action can be taken when detecting each of the different detection phonemes (loop 2). For example, if we combine two machines A / I and P / K, we can recognize 4 different words Paf / Pif / Clac / Clic.

As a variant, one of the classes of the one or more discriminating machines corresponds to all that is not the one or one of the phonemes to be detected, with the exception of the previous phonemes of the control words. Thus, it is possible to associate an action with a word or a beginning of a word and another action with any pronounced unrecognized word (any other group of phonemes).

The module 9 also comprises a control module 17 capable of extracting the continuous parameters of the voice signal such as duration, intensity, pitch, timbre. This extraction module 17 can be made as that described in the patent application FR-2905510.

The mapping module 10 comprises a generic action module 20, a specific action generation module 21, and a specific action modulation module 22 associated respectively with the analysis modules 13, 15 and 17 which generate actions in the game to be controlled according to the voice parameters detected by the module 9. Examples of mapping are also described in the patent application FR-2905510.

Thus, in a use phase, when the user transmits a voice signal 4, this voice signal is converted by the converter 7 into a digital voice signal 4.

As soon as the module 13 detects the presence of the voice signal 4, a generic action is generated by the module 20 in the set 3. Thus, the response time of the method according to the invention is almost instantaneous.

Then, when the classification module detects the control word present in the voice signal by means of the discrimination machine or machines, the specific action module 21 generates a specific action corresponding to the detected onomatopoeia.

The modulation module 22 of the action then makes it possible to control the action as a function of the continuous parameters of the voice signal extracted by the extraction module 17. Each of the continuous values can constitute a steering axis of the action. For example, if the specific action generated by the module 21 following the detection is the start of the race of a character, the height of the voice signal may control the direction of movement, while the intensity may control the direction of movement. speed.

In some implementations, several specific actions can be engaged successively during the successive detection of the phonemes in the command word. Thus, for example, if a first discrimination machine P / K is used in combination with a discrimination machine NI to detect the 4 onomatopoeia Paf / Pif / Clac / Clic, when the first consonant is detected, a first action can be controlled first two possible actions corresponding to the two groups Paf + Pif and Clac + Clic. Then, when the vowel is detected, a second action can be commanded among the 4 groups Paf, Pif, Clac and Clic. Thus, an action can be generated as soon as the first characteristic phoneme of the control word is detected by the classification module. This triggering of early action further increases the impression of instantaneous control of the action. It is then possible to drive the specific action on the control phoneme a or i of the control word continuously.

In a variant, the voice presence and corresponding action detection modules 13 can be omitted, so that an action is generated only when a detection phoneme has been detected by a machine. of discrimination.

FIG. 2 shows tables showing, for a pair of control words, the detection phonemes used for the detection, and the control phonemes used to control the action taken following the detection.

It is possible to control the action initiated from a phoneme different from that used to detect the control word. Thus the words bazooka and rifle are written respectively in phonetic language [b] [a] [z] [u] [k] [a] and [f] [y] [z] [i]. The detection phonemes [b] [a] [z] [u] [k] and [f] [y] [z] are used for the detection of one of the two words, while the driving phoneme [a ] ensures the control of the action taken following the detection of the word bazooka and the phoneme of control [i] ensures the control of the action taken following the detection of the word gun.

It is also possible to control the action detected from one of the phonemes used to carry out the classification of the word. Thus the words left and right are respectively written in phonetic language [g] [o] [f] and [d] [R] [u] [a] [t]. The phonemes [o] and [a] are used both as detection phonemes of the control words (module 15), and as control phonemes (module 17) of the action corresponding to the control words. In this case, the detection / driving phoneme triggers the action and sound-holding of the detection / control phoneme to control this action.

Moreover, the action can be stopped, either by a detection phoneme indicating the end of the action, or by detecting the end of the voice signal. Thus for the left / right pair it will be the detection of the phoneme ending the word ([f] for left or [t] for right) which will end the action while for the pair bazooka / rifle the action is stopped during the end of the pronunciation of the word (Figure 2). It is also possible to end the action when a predefined threshold is reached on any of the prosodic parameters (for example duration: at the end of a period, the action, intensity or height is stopped: when an intensity or height threshold is reached, the action stops).

In the remainder of the document and in the tables of FIGS. 3a to 3d showing machines that can be used to discriminate between different control words, the expression [x] + [y] corresponds to the cumulative recognition of the two phonemes [x ] and [y], the word being detected after the phonemes [x] and [y] have been detected by the discrimination machines regardless of the order in which they have been identified the expression [x] - [y ] corresponds to a cumulative successive recognition of the phonemes [x] and [y], the word being detected after the phoneme [x] and then the phoneme [y] have been detected by the discrimination machines 20 the expression ([x] ) indicates the unique existence of an optional phoneme [x] while the expression {[x] + [y]} indicates a concatenation of phonemes, if [x] = [y] this corresponds to a [x] of long duration.

Figure 3a shows the recognition of two words (flip and flap) using a discrimination machine between two phonemes I and A.

Figure 3b shows the recognition of two words (left and right) using a detection on the phonemes G and D, or O and A. In order to obtain a more robust detection, it is possible to combine two discriminating machines to perform cumulative recognition of two or more phonemes of type G + O and D + A. Alternatively, type O and OR or A recognition is performed; in this case, we build a specific machine in which the OU and the A are considered as one and the same phoneme. In the case of combined discrimination machines, the specific action can be engaged from the first detection phoneme and then continued if the second detection phoneme confirms that the onomatopoeia detected during the first test, or stopped otherwise. However, there are cases in which it is impossible to recognize words by combinations of simple elements without the introduction of a notion of time. An example is given in Figure 3c. Thus the word Babafi can not be differentiated from the word fibaba by considering only the simple phonemes F, I, B, A. It is therefore necessary to take into account the sequence of recognized phonemes to differentiate the word babafi from the word fibaba. Thus, the detection of the phoneme B followed by the phoneme F makes it possible to distinguish the word babafi from fibaba, which would be characterized by the phoneme F followed by the phoneme B.

On the other hand, it will be noted that the granularity of the classification may vary from word to phoneme.

This makes it possible, for example, to recognize two almost identical words, one having a steering part and the other not, but also to differentiate two words whose steering part uses the same phoneme. For example, in Figure 3d, two examples of implementations are shown which make it possible to recognize the words plus, pluuuuuuus, minus and minuuuuuus.

In a classical word learning as it appears in column 28, the system learns two classes: the short word without piloting (plus and minus) and the long word with piloting (pluuuuuus and 25 minuuuuus). Piloting is made possible before the end of the word because the recognition system sends its recognition hypotheses during the pronunciation of the word. The duration of the long word can be variable.

With an optimized learning by characteristic phonemes as it appears in column 29, the system makes a phoneme 30 recognition by phoneme: [p], [I], [y], [s], [m], [i], [ n] (corresponding to the letters P, L, U, S, M, I, N). The detection of the word is done on the recognition of the concatenation of the first characteristic phonemes, then the piloting phase is characterized by a multiple and continuous recognition of the driving phoneme [y] (letter U). In our example, the detection of the phoneme [s] at the end of the word is optional. However, it can be useful for specifying an end of word recognition action. The use of word with control can then be combined with the word recognition without piloting, and it is possible to assign different actions depending on whether the recognized word has a control phase during its recognition.

FIG. 4 gives an example of implementation of the method according to the invention in a shooting game in which two actions can be controlled respectively by means of two distinct words. Thus the word Rifle will command a rifle action, while the Bazooka 15 will command a bazooka shooting action, the prosodic parameters of the end of these words controlling for example the direction of the shot, the intensity, or any other characteristic of this shot.

For this purpose, a system 35 comprises a classification module 37 comprising a vowel discrimination machine 38 [u] / [y], a specific action module 41 making the correspondence between the word detected by the module 37 and the action to be triggered, as well as two control machines 44 and 45 (one per action that can be triggered) for continuously or discretely controlling the triggered action. A microphone 47 associated with an analog / digital converter 48 is related to the classification and control modules 37, 45, 45 for routing the digitized voice signal to these different modules.

Thus, as shown, when a user 50 pronounces in the microphone 47 the word bazooka, the machine 38 detects the presence of the phoneme [u] referenced 51, and the module 41 then triggers the bazooka firing action corresponding to the phoneme detected.

Then the control module 45 extracts parameters of the voice signal, in particular the phoneme [a] referenced 53, such as the duration, the pitch, the intensity, the timbre or combinations of these parameters, and if necessary, performs statistical calculations from these parameters to generate a set of continuous or discrete parameters.

These parameters 55 are then used by the module 45 to modulate the action triggered by the bazooka shot. In the context of a discrete control, emitting a high or low sound with respect to a threshold makes it possible to direct the shot to the left or to the right. In the context of continuous monitoring, the more or less pronounced pronunciation of the bazooka [a] will make it possible, for example, to continuously vary the intensity of the shot or its orientation.

In a variant, the system also includes modules 13 and 20, so that it is possible for the game character to perform an action, such as arming his weapon, as soon as a sound is detected, and even before know what kind of weapon (bazooka or rifle) will be ordered.

The invention is generalizable to multi-syllabic words. For example one can have a vocabulary consisting of boat and cut; from the boat b, we know the difference between cutting. Then we can hold the boat a to generate a continuous control, then move to the next state at the pronunciation of t, and finally pilot the action as long as we pronounce 20 o. This concept of looping is symbolized by the arrow 2 of Figure 1.

The control method according to the invention can be used as the sole mode of control but also in the context of a multimodal use, for example a use in combination with a game pad, a keyboard, or a motion recognition system, such as that implemented within the Wii console (registered trademark) or Eyetoy games (registered trademark), or an eye tracking system.

In the control method according to the invention, one or more words of the control vocabulary can be replaced by a sound such as a clapping of hands. An example of a control vocabulary can be constituted by {Up, Down, Clapping Hands}.

For an efficient implementation in the context of a video game system, we choose to work with the mapping outputs of the modules 20 to 22 synchronized with the video frames, that is to say that we synchronize with the refresh rate of the screen which is conventionally 60 frames per second. Indeed, the video game programmer can influence the game actions at each refresh of the screen, ie 1 time every 60th of a second. It is therefore useful to be able to provide synchronized control parameters for this period or multiples of this period.

It will be noted that the processing of the voice signal is carried out at a frequency determined by the acoustics of the phenomena to be analyzed, the frequency decorrelated at 60 frames per second, a synchronization system making it possible to retrieve the latest available voice analysis information from synchronized with the video refresh rate.

However, it is also possible to synchronize at faster frequencies.

Claims (14)

REVENDICATIONS1. Method for controlling an application, for example a video game, with a voice signal (4) characterized in that it comprises the following steps: defining a control vocabulary, that is to say a set of command words, preferably onomatopoeia, each having a characteristic phoneme or a plurality of characteristic phonemes making it possible to distinguish each word from the other words of the control vocabulary; - detecting (15) a control word contained in the voice signal (4) by means of at least one discriminating machine (38) for detecting and discriminating the pronunciation of different phonemes while possibly excluding a set of other phonemes, such discrimination machine (38) ensuring for example the distinction between at least two characteristic phonemes or the distinction between at least two groups of characteristic phonemes, - generating (21, 41) a specific action corresponding to the control word de tected, then 20 - driving (17, 44, 45) the specific action generated as a function of at least one continuous magnitude (55) of the voice signal, such as duration, intensity, pitch or timbre of the signal. Method according to claim 1, characterized in that the discriminating machine detects the phoneme or group of phonemes characteristic of the control word and generates the specific action before the end of the pronunciation of the control word. 3. Method according to claim 1 or 2, characterized in that the continuous magnitude of the voice signal (4) according to which the action is driven is located in time after the detection phoneme detected by the machine (38). ) of discrimination. 4. Method according to one of claims 1 to 3, characterized in that as soon as the presence of a voice signal (4) is detected (13), even before the recognition of the command word, a generic action is generated . 5. Method according to claim 4, characterized in that the specific action may be different from the generic action. 6. Method according to one of claims 1 to 5, characterized in that the method employing at least two discrimination machines, several specific actions are engaged successively io during the successive detection of phonemes or groups of phonemes features in the command word by each of the machines. 7. Method according to claim 6, characterized in that, two discriminating machines being used, a first specific action is generated after the detection of the first detection phoneme of the control word by a first discriminating machine capable of discriminating N different phonemes, so that a first action can be generated among N first possible actions, and 20 - a second specific action is generated after the detection of the second characteristic phoneme of the control word by a second discriminating machine able to discriminate M phonemes distinct, so that this second action can be generated among N x M second possible actions. 25 8. Method according to one of claims 1 to 7, characterized in that: - as soon as one detects a phoneme or a group of characteristic phonemes, or - as soon as one of the continuous parameters of the voice signal exceeds one certain threshold, for example as soon as the voice signal disappears, - the specific action ends. 9. Method according to one of claims 1 to 8, characterized in that 1 and 7 discrimination machines are implemented in order to be able to differentiate between 1 to 15 control words. 10. Method according to one of claims 1 to 9, characterized in that the discrimination machine uses hidden Markov models. 11. Method according to one of claims 1 to 10 characterized in that one chooses to generate actions and to drive in a manner synchronized with the video frames of a game to be driven. 12. Method according to one of claims 1 to 11, characterized in that it is implemented in combination with a control method by another control interface, such as for example a game pad, a keyboard, a device eye tracking, or motion tracking device. 13. Method according to one of claims 1 to 12, characterized in that to resist noisy environments, using a contact microphone. 20 14. Device for controlling an application, for example a video game, with a voice signal (4) characterized in that - a control vocabulary, that is to say a set of command words, preferably Onomatopoeia, being defined, each control word having a characteristic phoneme or several characteristic phonemes making it possible to distinguish each word from the other words of the control vocabulary, it comprises: means (15) for detecting a control word contained in the voice signal (4), said means (15) comprising at least one discrimination machine (38) for detecting and discriminating the pronunciation of different phonemes while possibly excluding a set of other phonemes, said machine (38) of discrimination ensuring, for example, the distinction between at least two characteristic phonemes or the distinction between at least two groups of characteristic phonemes; means (21, 41) for generating a specific action corresponding to the detected control word, and - means (17, 44, 45) for controlling the specific action generated as a function of at least one continuous magnitude (55) of the voice signal, such as the duration, the intensity, pitch or timbre of the signal.