METHOD FOR OPERATING A SPEECH DIALOGUE SYSTEM AND SPEECH DIALOGUE SYSTEM

Description

[0001] The invention relates to a method for operating a speech dialogue system in which voice commands from a voice input are detected and assigned a priority for further processing. The invention also relates to a speech dialogue system configured to implement such a method.

[0002] Voice dialog systems or voice assistants in vehicles are well known from the state of the art. Problems arise when identical or phonetically similar commands can be used to control different systems. For example, a voice dialog system for operating vehicle functions is known from US 2008 / 0059195 A1. In this system, each vehicle function is assigned to a group of commands. At the same time, priorities are assigned to the commands and the respective vehicle function. Further processing is then based on these priorities in order to avoid conflicts with functions that can be activated with the same or similar commands.

[0003] In addition to such adaptation of phonetically similar voice commands for different functions, in practice it is often the case that voice inputs contain several individual voice commands. These are then often not recognized, not recognized in full, or even recognized incorrectly. In this case, an incorrect command or only part of the command(s), especially a recognized voice command, is often processed and subsequently executed. If several voice commands in the voice input are recognized correctly at all, they are typically processed in the uttered or spoken order, which is often not practical. Since the person speaking typically does not give any thought to serial processing of their voice commands, this order often does not correspond to their actual intention.

[0004] US 2020 / 0302924 A1 discloses a trained model for determining the execution order of instructions issued by a user via voice input. The trained model reflects a preference of a majority of users, with conditional instructions being taken into account during training.

[0005] EP 3 115 886 A1 discloses assigning system responses to execution devices for which priorities are defined. The output time data is generated depending on the priorities of the respectively assigned execution devices. The output data can thus advantageously be output in a chronological sequence that is adapted to the respective execution devices. The priorities defined for the execution devices can include further prioritization for different system responses of the same execution devices.

[0006] The object of the present invention is to provide a method which enables a sensible order of execution when several commands are spoken one after the other.

[0007] According to the invention, this object is achieved by a method having the features of claim 1, and in particular in the characterizing part of claim 1. Advantageous embodiments and further developments emerge from the dependent claims. Furthermore, a speech dialogue system according to claim 9 achieves this object.

[0008] The method according to the invention, which can preferably be used in a vehicle, provides that, when several consecutive voice commands are given in a single voice input, a content group of commands is assigned to each of the recognized voice commands. These content groups can, for example, be the control of vehicle functions as one of the groups or the control of a navigation system as another group. Other groups could, for example, be the control of a media system, the control of the interior lighting, the operation of a telephony system, or the like. The voice command itself is then linked to a first priority value specified for the respective group. This priority value can, in principle, be fixed or it can be flexibly adapted, for example, programmed by the vehicle manufacturer, but also individually set by the user based on user specifications.Automatic learning of user preferences over time is also conceivable here.

[0009] After the corresponding voice command has been associated with a group and linked to the corresponding priority value, the voice command itself is assigned a second priority value within the group according to a predefined priority list of the commands assigned to the group. The individual commands within a group are thus in turn prioritized in their order via a priority list. This priority list can also be flexibly adapted accordingly, for example, by programming, learning, or even setting it by the user.

[0010] In a subsequent step, a third priority value is then formed from the first and second priority values, for example, by summing these two values. The crucial aspect of the method is that the voice commands, regardless of their spoken order, are sorted into a sequence according to their final priority values—here, the third priority value, which is preferably the sum of the first two priority values—and are processed in this sequence.

[0011] With this type of double prioritization of the values, first according to the group to which the voice command is assigned and then within the group according to a priority list, it is already possible to achieve a corresponding prioritization for determining a meaningful order of the several consecutively spoken voice commands of a voice input.

[0012] A further very advantageous embodiment of the method according to the invention can provide that the second priority value is smaller than the difference between two adjacent first priority values. This is a decisive advantage, particularly in the variant in which the first and second priority values ​​are added together. By configuring the first priority value, for example, in increments of 10, and the second priority value, for example, in increments of 1, it is achieved that if a recognized voice command is prioritized according to its group assignment, it cannot leave the assigned group again using the second priority value and thus change the prioritization based on the group. Due to this special configuration of the size of the two priority values ​​in relation to one another, prioritization based on the group always takes precedence over classification according to the group's priority list.

[0013] A further advantageous embodiment of the method according to the invention further provides that, if the final priority values ​​are the same, the sorting sequence is based on the spoken order of the voice commands. Therefore, if two voice commands have the same final priority value, the system cannot independently determine a meaningful order of the commands based on the prioritization. In this case, the order the voice commands had within the voice input is used; the first voice command spoken is sorted before the next voice command with the same final priority and processed accordingly.

[0014] The method according to the invention then further evaluates the speech input with regard to temporal adverbs relating to the speech commands. Such temporal adverbs could, for example, include "before" or "after" and are used in practice in such a way that, for example, the speech input states that a first command should be executed before another command, that the latter should be executed after it, or that an action initiated by the command should be executed first or last. In the case of such detected temporal adverbs, the method according to the invention provides that for each of the recognized directions—i.e., whether the command should be executed before or after—a correction value is generated, which changes in the same sense with each additional temporal adverb in the same direction.In the same sense, this means that an upgrade, i.e., an increase in the priority value, based on a previous value, leads to a further increase in the priority value in the case of a further previous value, and vice versa. In principle, it doesn't matter how the order is sorted according to the priority values—i.e., whether the highest or lowest value is sorted first—but the change chosen here must, of course, fit this basic principle of changing the priority value.

[0015] The respective third priority value of each voice command is finally offset against the correction value generated using the temporal adverbs, if present.

[0016] Here, too, a simple summation of the correction value and the third priority value, which in turn preferably consists of the sum of the first and the second priority value, can preferably be carried out.

[0017] A particularly advantageous embodiment of this variant of the method according to the invention also provides that the correction value includes a positive or negative correction factor depending on the direction of the temporal adverb. This would allow the recognized direction in particular to be mapped via the sign of the correction value, while still maintaining a simple calculation, for example by summing. Depending on whether the highest or lowest priority value is sorted at position 1, the preferential treatment (the lower the priority value), for example, would require the before direction to be assigned a factor of -1 and the after direction to be assigned a factor of +1 in order to obtain the desired additional prioritization through the temporal adverbs.

[0018] In addition, a counter can be incremented in each direction with each additional temporal adverb, thus adjusting the priority value accordingly by multiplying the counter by the first specified correction value. The value of the correction factor can preferably be set higher than the largest first priority value of the group with the highest priority value, so that when temporal adverbs are recognized, prioritization by the temporal adverbs with priority takes precedence over prioritization by the groups. This helps avoid overlaps and achieves equalization of the sorted voice commands within the sequence.

[0019] Furthermore, according to a very advantageous embodiment of the method according to the invention, it can also be provided, alternatively or additionally, that the voice input is evaluated with regard to specific information regarding the sequence of the voice commands, for example, if a voice input includes a first voice command and then, in conjunction with it, for example, a "and after" or "but first" with another voice command. A sequence thus recognized, which obviously corresponds to the will of the person initiating the voice input, is then used with priority for the final priority values, i.e., the third priority value or the third priority value offset against the correction value.

[0020] The specific information regarding the sequence can lead to potential gaps within the sequence of voice commands, for example, if a "third" is spoken during voice input even though there was only one previous command, or something similar. However, such gaps in the sequence of prioritized and / or ranked voice commands are undesirable, as they lead to unnecessary delays in the process. The method can therefore provide for filling such gaps accordingly and, if a gap arises in the sequence, moving the subsequent voice commands forward one position each to close the gap accordingly.

[0021] According to an advantageous embodiment, the voice commands are further checked for similarity. In the case of voice commands that exceed a predetermined degree of similarity, voice commands from the same group of voice commands are combined. For example, voice commands such as "Activate the seat heating on the driver's seat" and "Activate the seat heating on the passenger seat" can be combined into a single command in which the command "Activate the seat heating in the front" is executed. This saves time due to serial processing of the commands, and voice commands further down the sequence are executed sooner. The check for potential similarities can be performed using a semantic check.Alternatively, or in particular in addition, summaries of voice commands can also be provided using tables, logic functions and / or the like.

[0022] In particular, the individual voice commands can be checked for similarity in pairs, since combining very similar commands spoken directly one after the other enables particular efficiency, while commands spoken further apart often no longer reveal such a possibility or necessity.

[0023] Furthermore, according to an exceptionally advantageous development of the method according to the invention, the voice commands can be checked for defined rules in order to eliminate logically non-sensical voice commands or to replace them with further voice commands from the same group of voice commands. For example, if two consecutive voice commands refer to the playback of media content, a rule can be defined that only the later voice command is executed and the previously spoken voice command is ignored, in order to avoid rapid switching, for example, between different playlists or stations in the media device.

[0024] A further very advantageous embodiment of the method according to the invention can additionally provide that error messages that occur during the processing of the voice commands according to their sorted order are output only once for multiple voice commands of the same group. For example, if an error message is generated for a specific group of functions, it can only be output once to avoid being repeated with each command assigned to the similar group. For example, if the voice command "Turn on the massage for the driver and turn on the massage for the rear left" is output, the two recognized commands would result in the same error message twice in a conventional system, for example, "The vehicle does not have massage seats."In this case, this error message can be summarized so that it is only displayed once, thus giving the person using the speech dialogue system the feeling of a comparatively "natural" conversation in which the same content is not displayed identically several times in succession.

[0025] The speech dialogue system according to the invention is constructed very similarly to a conventional speech dialogue system and has at least one microphone and a device for speech recognition and natural language understanding. The speech dialogue system according to the invention comprises at least one computing unit configured to execute a method as described above.

[0026] In particular, voice outputs can also be provided in the speech dialogue system, for example, a voice output of error messages or other feedback of any kind. For this purpose, a unit for natural language generation can be provided according to an extremely advantageous further development of the speech dialogue system.

[0027] The speech dialogue system and the associated procedure can, in principle, be used wherever speech dialogue systems are used. This can be particularly useful in vehicles. A speech dialogue system is particularly advantageous in vehicles, as it allows a person driving the vehicle to operate functions within the vehicle without having to turn away from the traffic, which benefits both comfort and road safety.

[0028] Further advantageous embodiments of the method according to the invention and of the speech dialogue system according to the invention also emerge from the exemplary embodiment which is described in more detail below with reference to the figures.

[0029] Showing: Fig. 1 shows a vehicle with a voice dialogue system according to the invention; Fig. 2 shows a schematic process for preparing the sequence of voice commands; and Fig. 3 shows a schematic representation of the process during the processing of the voice commands.

[0030] In the presentation of the Figure 1a speech dialogue system 1 is shown in a schematically indicated vehicle 2. In the embodiment shown here, which is generally usual, the speech dialogue system 1 has four individual microphones 3 for each of the passengers of the vehicle 2 and an output unit in the form of a loudspeaker 4 assigned to the respective microphone 3. The speech dialogue system 1 itself can be designed as a speech dialogue system 1 operating purely in the vehicle 2, which is typically referred to as an embedded system. However, it can also be designed as a pure cloud system and outsource its computing operations to a server external to the vehicle, which is indicated here as cloud 5. In particular, the speech dialogue system 1 should be designed as a hybrid system, i.e. with an embedded component and a cloud component.

[0031] The speech inputs recorded by the microphones are passed on to a head unit 6, which contains the actual speech processing module 7, e.g. as an integrated computing unit. This speech processing module 7 comprises signal pre-processing 8, a component 9 for speech recognition and a component for semantic analysis, which is designated here by 10 and is often also referred to by the English term Natural Language Understanding (NLU). The NLU component 10 is able to process several voice commands spoken one after the other by a person in the vehicle 2 into a single utterance or speech input and to split them up into several individual voice commands, so-called intents. In addition, the speech processing module 7 has a text-to-speach component 11 and a component for natural speech output, which is often also referred to as Natural Language Generation or NLG.It is designated 12 here. Furthermore, a dialogue manager 13 is provided within the speech processing module 7. The head unit 6 itself is connected to a display 14, which can also be used for corresponding displays and the like within the context of the speech dialogue.

[0032] If the individual voice commands or intents were processed in the recognized or spoken order, the problem arises that the time information given by the person using the system is not taken into account and that the spoken order may not necessarily be the actually desired and, in particular, not the most efficient order for processing the voice commands.

[0033] For example, the command "Call Max Mustermann and start route guidance to XY" would result in route guidance to XY only being started after the phone call has ended if the command was processed sequentially. However, with appropriate prioritization of the recognized voice commands, an optimized sequence can now be implemented. For example, route guidance can be started first and the call made only afterward. As a result, route guidance is available to the person using vehicle 2 even during the phone call, which is often an advantage, but at least never a disadvantage.

[0034] Therefore, a possibility will be described below to prioritize recognized voice commands, to take into account spoken information on the temporal processing of the voice commands and to substitute two or more consecutive commands with a high degree of similarity by a single meaningful command and, if necessary, to anticipate errors in the processing.

[0035] To achieve this, all voice commands or intents recognized in the voice input or utterance are first extracted from the data provided by NLU component 10, including the information associated with the respective voice commands. Each of the voice commands is initially assigned an initial priority value, for example, a value of 0. Additionally, each of the voice commands is assigned a value that corresponds to the spoken position in the voice input or utterance.

[0036] In the following, several optional steps are described to influence the priority value and thus the order in which the voice commands are processed.

[0037] First, it checks whether intents have been identified for which clear information about the desired order has also been spoken. These could be lists such as "first," "second," "start with this," "last," and "finally." These intents are excluded from the subsequent steps and are only inserted at the recognized or spoken position in a sequence of voice commands before processing begins. If the same position is spoken multiple times, the first time information can be ignored, for example, or the intents with the same time information can be arranged in direct succession.

[0038] Several optional steps follow for the remaining voice commands or intents: First, a group- or domain-specific weighting is performed to ensure that intents based on the group domain to which they belong are always processed before intents from other groups or domains. For example, the initially specified priority value can be increased by setting an initial priority value depending on the group to which the voice command is assigned.

[0039] For example, if the voice dialogue system has the ability to operate the groups telephony, vehicle functions, media, messages and navigation, a weighting could look like the one shown below, where a lower value corresponds to an earlier execution - and thus a higher priority. Vehicle functions: 10 Navigation: 20 Media: 30 Messages: 40 Telephony: 50

[0040] Ideally, these values ​​are not hard-coded, but can be read from a configuration file, allowing the behavior to be easily optimized or the weighting to be easily deactivated. Depending on the associated group, the initial priority value is increased by the first priority value thus determined. This can preferably be done by a simple summation.

[0041] In the next step, the voice commands or intents are prioritized within their group. For example, a second priority value can be added to the existing first priority value. It can be advantageous if the third priority values, which are then the sum of the first and second priority values, are not changed in such a way that the voice commands fall into the value range of a different group. According to the example described above, the second priority values ​​would therefore be between 1 and 9.

[0042] Ideally, these values ​​aren't hard-coded, but can be read from a configuration file to easily optimize behavior. If the intra-domain-specific second priority value is set to 0 for each intent, no additional weighting is applied within the domain.

[0043] In order to additionally respond to temporal adverbials such as "before" or "after," a further correction value is added to the priority values ​​in a next step. The intents are processed according to the spoken order. Furthermore, each intent is checked to determine whether a statement regarding previous or subsequent processing has been made. For example, if it is detected that a "before" was spoken for an intent, the correction value -100*i*a is added to this intent and all intents spoken after it, with i increasing by 1 for each additional "before." In this example, the variable a is used to deactivate the weighting by setting the value to either 1 (on) or 0 (off). If an "after" or similar is detected, a slightly modified value of +100*j*b is added, where b also serves analogously to deactivate the weighting.The variable j is initially 1 and is incremented by 1 with each subsequent "after".

[0044] Finally, the intents are sorted according to their final priority values, i.e., the sum of the first and second priority values ​​and the correction value. If the final priority values ​​are the same, the affected intents are additionally sorted according to the spoken order.

[0045] In order to shorten dialogues if necessary, the intents are checked for similarity, preferably in pairs, and if necessary replaced by a single equivalent intent according to a set of rules.

[0046] The intents for which a specific time was spoken are then sorted according to the spoken information. Any gaps that arise are eliminated. For example, with an input such as "Turn on the seat heating and turn on the lights third," the voice commands are sorted according to the described process. However, after the second voice command is sorted in third position, a gap arises in second position, which is eliminated here by moving the intent sorted in third position forward.

[0047] Finally, the intents can optionally be post-processed according to a defined set of rules to avoid senseless behavior or perceived errors, or to combine multiple intents into a single meaningful intent. For example, if two consecutive intents refer to playing media content, a rule could be defined that only the later intent is executed. For example, if the voice input "Play Michael Jackson on Spotify and play the station SWR3," only the voice command "Play SWR3" would be executed.

[0048] Figure 2This is illustrated by the fact that NLU component 10 is symbolized again in the top left box. In the following box, labeled 201, the corresponding data is then extracted in order to form the basis for the sequence. If numbering has already been specified, box 202 branches directly to box 203, which groups the voice commands together in their sequence and submits them to final processing by the rule set in box 204 before the final sequence is determined in box 205. If such numbering is not present, which will be the more typical case, prioritization is carried out in the area of ​​box 206 according to the group in which the voice command is assigned, i.e. the first priority value is determined. In box 207, the second priority value is then determined, i.e. the voice command is prioritized again within its group.Box 208 then processes the temporal adverbials and the spoken temporal sequences in the manner described above, thus ultimately generating the correction values, which are then calculated accordingly in box 209 and fed into the merge of the voice commands with their priority values ​​or the numbers in their sequence according to the priority values. The processes in boxes 204 and 205 then follow in a similar manner.

[0049] A concrete example might look like this: The voice command should be: "Set Sindelfingen as the intermediate destination, drive to Stuttgart, activate the seat heating for the driver, also activate the seat heating for the passenger, set the temperature to 21°C before that, and activate the massage first!"

[0050] Intents in spoken order: Add navigation.intermediate destination <sindelfingen>; Add navigation.destination <stuttgart>; function.switch on.seat heating <fahrer>; function.switch on.seat heating <beifahrer>; function.set.temp <21> <davor>; function.switch on.massage<als erstes>

[0051] Intents after checking for spoken order: With specification: function.switch on.massage<als erstes> Without specification: Navi.add.intermediate destination <sindelfingen>; Add navigation.destination <stuttgart>; function.switch on.seat heating <fahrer>; function.switch on.seat heating <beifahrer>; function.set.Temp <21 > <davor>

[0052] Intents without expected order according to domain-specific prioritization: function.switch on.seat heating <fahrer>; function.switch on.seat heating <beifahrer>; function.set.Temp <21> <davor>; Add navigation.intermediate destination <sindelfingen>; Add navigation.destination <stuttgart>;

[0053] Intents without expected order according to intra-domain-specific prioritization: function.switch on.seat heating <fahrer>; function.switch on.seat heating <beifahrer>; function.set.Temp <21> <davor>; Add navigation.destination <stuttgart>; Add navigation.intermediate destination <sindelfingen>;

[0054] Intents without an expected order after prioritization according to the occurring temporal adverbs or temporal conjunctions: function.set.temp <21> <davor>; function.switch on.seat heating <fahrer>; function.switch on.seat heating <beifahrer>; Add navigation.destination <stuttgart>; Add navigation.intermediate destination <sindelfingen>;

[0055] Intents after merging: function.switch on.massage<als erstes> ; function.set.Temp <21> <davor>; function.switch on.seat heating <fahrer>; function.switch on.seat heating <beifahrer>; Add navigation.destination <stuttgart>; Add navigation.intermediate destination <sindelfingen>;

[0056] Intents after rule-based post-processing: function.switch on.massage<als erstes> ; function.set.Temp <21> <davor>; function.switch on.seat heating <vorne>; Add navigation.destination <stuttgart>; Add navigation.intermediate destination <sindelfingen>; After sorting in this order, the system begins to process the voice commands individually and one after the other in the determined order.

[0057] If errors occur during the processing of individual intents that would occur to the same extent for intents from the same group that are to be processed later, the system is able to eliminate these later intents from the processing queue based on a fixed set of rules. For example, the vehicle should not have massage seats, but the user says "Turn on the massage for the driver and turn on the massage for the rear left." If both detected intents were simply processed, the same error prompt "Unfortunately, this vehicle does not have massage seats" would be displayed twice. However, by summarizing the errors, the error prompt is displayed only once. This is sufficient to inform the user and conveys a more natural dialog behavior.

[0058] If a user speaks additional voice commands while executing previously recognized voice commands, the system can optionally ignore the newly spoken voice commands or add them to the queue of commands still to be processed.

[0059] In the presentation of the Figure 3 This is taken up. Box 301 starts the processing of the previously created, already sorted or prioritized voice commands of an input chain, i.e. a command chain with consecutive voice commands. Box 302 then symbolizes the reading of the input chain, query 303 checks whether an input chain exists; if not, the determination is ended in box 304 and the process aborted. The voice dialog is then ended. The same applies if the dialog is aborted by the user, as indicated in box 305. If the input chain is not empty, the first voice command in the prioritized input chain is executed in box 306. If a check in box 307 results in an error, the command processed in 306 is removed from the input chain in box 308 and the commands remaining in the input chain are checked against a set of rules.If a command is expected to cause the same or a similar error, these commands are deleted from the command queue in box 309. An example of an error scenario is "Turn on the XY massage for the driver and turn on the YZ massage for the rear right." The commands cannot be combined and would therefore be executed sequentially. When executing the first command, an error message occurs: "Unfortunately, there are no massage seats in this vehicle." The command is then deleted from the command queue because it has already been processed. Next, the remaining commands are checked, and if so, the second spoken command is also discarded, and the dialog is terminated.

[0060] Box 312 then describes the normal case, namely that the voice command was processed successfully and without interruption. Afterward, it is deleted from the input chain according to box 313. Afterward, the program returns to box 302 to restart the process and execute the next command in the input chain. This process is repeated until all voice commands in the input chain have been processed accordingly and the process ends in box 304.

[0061] Box 310 handles the case where a new voice input is received while the commands in the input chain have not yet been fully processed. In this case, the currently processed command is executed first in box 311. This command is then removed from the command chain, and the newly recognized command is inserted at the top of the command chain. The input chain, supplemented with the new voice input, is then returned to box 302 for processing.

[0062] To further optimize functionality, the speech dialog system 1 for outputting prompts, such as error or confirmation messages, has a classic text-to-speech component 11 and, optionally, an NLG component 12, which allows the output prompts to be more natural. If a user speaks several voice commands that can be executed directly without further questions, this results in the respective confirmation prompts being output sequentially without the need for an NLG component 12 or complex rule sets. For example, "I turn on the seat heating. I turn on the reading light." At this point, the NLG component 12 enables prompts to be linked or formulated more appropriately. With NLG component 12, the feedback could then be, for example, "I turn on the seat heating and I turn on the reading light" or, even more naturally / better, "I turn on the seat heating and the reading light."

[0063] In addition to or as an alternative to the voice output, error or confirmation messages may also be shown on the display 14.< / sindelfingen> < / stuttgart> < / vorne> < / davor> < / sindelfingen> < / stuttgart> < / beifahrer> < / fahrer> < / davor> < / sindelfingen> < / stuttgart> < / beifahrer> < / fahrer> < / davor> < / sindelfingen> < / stuttgart> < / davor> < / beifahrer> < / fahrer> < / stuttgart> < / sindelfingen> < / davor> < / beifahrer> < / fahrer> < / davor> < / beifahrer> < / fahrer> < / stuttgart> < / sindelfingen> < / davor> < / beifahrer> < / fahrer> < / stuttgart> < / sindelfingen>

Claims

1. Method for operating a speech dialog system (1), in which voice commands of a voice input are recorded and assigned a priority for further processing, characterized in that out of a plurality of successive voice commands, each recognized voice command is allocated to a content-related group of commands, the voice commands being linked to a first priority value which is predefined for the particular group, after which a second priority value is allocated to each of the voice commands within the group according to a predefined priority list of the commands allocated to the group, characterized in that a third priority value is formed from the first priority value and the second priority value, and after which the voice commands are sorted into a sequence according to their final third priority value and are fed for processing in this order, the voice input being additionally evaluated with respect to temporal adverbs for the voice commands, a correction value being generated for each of the two recognized directions of the temporal adverbs, which correction value changes with each further temporal adverb in the same direction, the particular third priority value of each voice command being offset against the correction value, and the recognized directions indicating whether a command should be executed before or after.

2. Method according to claim 1, characterized in that the second priority value is smaller than the difference between two adjacent first priority values.

3. Method according to either claim 1 or claim 2, characterized in that if the final priority values are the same, the sorting is done according to the spoken sequence of the voice commands.

4. Method according to any of claims 1 to 3, characterized in that the correction value includes a positive or negative correction factor depending on the direction of the temporal adverb.

5. Method according to any of claims 1 to 4, characterized in that the voice input is evaluated with respect to concrete information on the sequence of the voice commands, a recognized sequence being primarily used for the final priority values for sorting into the sequence.

6. Method according to any of claims 1 to 5, characterized in that the voice commands are checked for similarities determined by semantic testing, voice commands that exceed a predefined degree of similarity being grouped together from the same group of voice commands.

7. Method according to any of claims 1 to 6, characterized in that the voice commands are checked for defined rules in order to eliminate voice commands that do not make logical sense, the rules defining that in the case of two successive voice commands relating to the playback of media content, only the later voice command is to be executed.

8. Method according to any of claims 1 to 7, characterized in that during the processing of voice commands according to their sorted sequence, error messages that occur for a plurality of voice commands of the same group are in each case output only once.

9. Speech dialog system (1) having at least one microphone (3) and a device for speech recognition (9) and for natural speech understanding (10), having at least one computing unit (7) which is designed to carry out a method according to any of claims 1 to 8.