CN113241067B - Voice interaction method and system and voice interaction equipment - Google Patents

Abstract

The present invention relates to a voice interaction method, system, and voice interaction device. The method includes: acquiring a first voice command input by a user, and obtaining a first voice text according to the first voice command; recognizing the first voice The dialogue state of the text; one of the first NLP engine and the second NLP engine is selected as the current NLP engine; then, the selected current NLP engine performs semantic recognition on the first voice text to obtain the first intent corresponding to the first voice text and the first corpus, and generate a control instruction corresponding to the first intention, send the control instruction to the execution unit to perform the corresponding task, and output the first corpus to the voice broadcast unit to broadcast the first corpus. The system corresponds to the method, and the voice interaction device includes the system or is capable of executing the method. The present invention can avoid the conflict and confusion of NLP engines in the voice interaction mode using multiple NLP engines.

Description

A voice interaction method, system, and voice interaction device

technical field

The invention relates to the technical field of vehicle voice interaction, in particular to a voice interaction method, a system thereof, and a voice interaction device.

Background technique

Intelligent voice interaction is one of the current mainstream modes of human-computer interaction. For in-vehicle voice interaction, the existing technology proposes an interactive mode that uses a local NLP engine and an online NLP engine based on subdivision scenarios to recognize voice. User input There are two ways of voice recognition in the vehicle. In the first method, the user enters the voice through the microphone of the vehicle terminal, and the vehicle terminal obtains the recognition result through the local NLP engine; The terminal obtains the recognition result through the online NLP engine. It is understandable that, after multiple NLP engines perform semantic understanding and return results, how to make optimal judgments, and in order to avoid conflicts, confusion and delays caused by the use of multiple NLP engines, a reasonable arbitration method is needed to coordinate the two NLPs. The working mechanism of the engine ensures the user experience.

In the process of realizing the present invention, the inventor found that the prior art has at least the following technical problems:

The existing NLP arbitration mechanism using the voice interaction mode of multiple NLP engines is not perfect enough to effectively improve the effectiveness of the response, and it is easy to cause conflict and confusion of the response results. The response is slow, which brings a bad user experience.

SUMMARY OF THE INVENTION

The present invention aims to provide a voice interaction method, a system, and a voice interaction device, so as to avoid the conflict and confusion of NLP engines in the voice interaction mode using multiple NLP engines, and improve user experience.

An embodiment of the present invention provides a voice interaction method, including:

Obtain the first voice command input by the user, and obtain the first voice text according to the first voice command;

Identifying the dialog state of the first voice text;

One of the first NLP engine and the second NLP engine is selected as the current NLP engine; wherein: if the dialogue state of the first voice text is a single-round dialogue, the text keywords of the first voice text are extracted, and the text keywords are combined with the pre-defined text keywords. The default keywords in the keyword table are matched to obtain the matching degree, and one of the first NLP engine and the second NLP engine is selected as the current NLP engine according to the comparison result of the matching degree and the preset threshold; if the first voice text If the dialogue state is multiple rounds of dialogue, the NLP engine used in the previous round of dialogue is determined as the current NLP engine;

The current NLP engine performs semantic recognition on the first voice text to obtain a first intent and a first corpus corresponding to the first voice text, and generates a control instruction corresponding to the first intent and sends the control instruction to the execution unit for execution For the corresponding task, output the first corpus to the voice broadcasting unit to broadcast the first corpus.

Preferably, selecting one of the first NLP engine and the second NLP engine as the current NLP engine specifically includes:

When the matching degree is less than the preset threshold, select the first NLP engine as the current NLP engine;

When the matching degree is greater than or equal to the preset threshold, the second NLP engine is selected as the current NLP engine.

Preferably, the method further includes:

When the second NLP engine is selected as the current NLP engine, if the second NLP engine fails to perform semantic recognition on the first voice text and cannot obtain the first intent and the first corpus corresponding to the first voice text, the output is used for Prompting the user to carry out the second corpus taught by the first voice instruction to the voice broadcasting unit to broadcast the second corpus;

Moreover, the second NLP engine determines the intent of the first voice text according to the teaching information input by the user, generates a control instruction corresponding to the intent of the first voice text, and sends the control instruction to the execution unit to execute the corresponding task.

Preferably, the intention of the second NLP engine to determine the first voice text according to the teaching information input by the user specifically includes:

The second NLP engine acquires the second voice text, performs semantic recognition on the second voice text to obtain the second intent, and generates a third corpus for prompting the user to confirm whether the intentions of the first voice text and the second voice text are consistent ; Wherein the second voice text is obtained according to the second voice command input by the user;

sending the third corpus to the voice playback unit to play the third corpus;

The second NLP engine obtains the confirmation information input by the user, and when the confirmation information is consistent with the confirmation, generates a control instruction corresponding to the second intention, and sends the control instruction to the execution unit to execute the corresponding task.

Preferably, the method further includes:

When the confirmation information is to confirm that the intentions of the first voice text and the second voice text are consistent, a mapping relationship is established between the first voice text and the second voice text, and the first voice text is added as a new corpus into the dynamic corpus.

Preferably, the first NLP engine is an online NLP engine, which includes an offline corpus; the second NLP engine is a local NLP engine, which includes a dynamic corpus; several corpora in the offline corpus are related to the dynamic corpus. Several corpora in the corpus have a mapping relationship;

The method also includes:

After the offline corpus of the first NLP engine is updated online, the preset keyword table is compared and updated according to the updated corpus of the offline corpus.

An embodiment of the present invention also provides a voice interaction system, including:

a voice text acquisition unit, configured to acquire the first voice command input by the user, and obtain the first voice text according to the first voice command;

a dialogue management unit, configured to identify the dialogue state of the first voice text;

The arbitration unit is used to select one of the first NLP engine and the second NLP engine as the current NLP engine; wherein: if the dialogue state of the first voice text is a single-round dialogue, then extract the text keywords of the first voice text, and the The text keywords are matched with the default keywords in the preset keyword table to obtain the matching degree, and one of the first NLP engine and the second NLP engine is selected as the current NLP engine according to the comparison result between the matching degree and the preset threshold; If the dialogue state of the first voice text is multiple rounds of dialogue, determine the NLP engine used in the previous round of dialogue as the current NLP engine;

The first NLP engine and the second NLP engine are used to perform semantic recognition on the first speech text when they are used as the current NLP engine to obtain the first intent and the first corpus corresponding to the first speech text, and generate the first intent and the first corpus corresponding to the first speech text. A control instruction corresponding to the intent is sent to the execution unit to execute the corresponding task, and the first corpus is output to the voice broadcast unit to broadcast the first corpus.

Preferably, the arbitration unit is specifically used for:

When the matching degree is less than the preset threshold, select the first NLP engine as the current NLP engine corpus;

When the matching degree is greater than or equal to the preset threshold, the second NLP engine is selected as the current NLP engine corpus.

Preferably, the second NLP engine is also used for:

When the second NLP engine is used as the current NLP engine, if the second NLP engine fails to perform semantic recognition on the first voice text and cannot obtain the first intent and the first corpus corresponding to the first voice text, the output is used for prompting The second corpus taught by the first voice instruction is sent to the voice broadcasting unit by the user to broadcast the second corpus;

Moreover, the second NLP engine determines the intent of the first voice text according to the teaching information input by the user, generates a control instruction corresponding to the intent of the first voice text, and sends the control instruction to the execution unit to execute the corresponding task.

Preferably, the second NLP engine is specifically used for:

Acquiring the second voice text, performing semantic recognition on the second voice text to obtain the second intent, generating a third corpus for prompting the user to confirm whether the intentions of the first voice text and the second voice text are consistent, and converting the third voice text The corpus is sent to the voice playback unit to play the third corpus, and the confirmation information input by the user is obtained, when the confirmation information is consistent with the confirmation, a control instruction corresponding to the second intention is generated, and the control instruction is sent to the execution unit for execution corresponding tasks;

The voice-text obtaining unit is further configured to obtain the second voice-text according to the second voice instruction input by the user and send it to the second NLP engine.

Preferably, the first NLP engine is an online NLP engine, which includes an offline corpus; the second NLP engine is a local NLP engine, which includes a dynamic corpus; several corpora in the offline corpus are related to the dynamic corpus. Several corpora in the corpus have a mapping relationship;

The second NLP engine is further configured to establish a mapping relationship between the first voice text and the second voice text when the confirmation information is to confirm that the intent of the first voice text and the second voice text is consistent, and map the first voice text to the second voice text. The first voice text is added to the dynamic corpus table as a new corpus;

The system further includes a keyword table updating unit, which is used to update the pre-predicted corpus according to the corpus of the updated offline corpus after the offline corpus of the first NLP engine is updated online. Set the keyword table to compare and update.

An embodiment of the present invention further provides a voice interaction device, including: the voice interaction system according to the foregoing embodiment; or, a memory and a processor, wherein the memory stores computer-readable instructions, and the computer-readable instructions are When executed by the processor, the processor is caused to execute the steps of the voice interaction method according to the foregoing embodiments.

The above technical solutions have at least the following advantages: by acquiring the first voice command input by the user, and obtaining the first voice text according to the first voice command; identifying the dialogue state of the first voice text, if the dialogue of the first voice text If the state is a single round of dialogue, extract the text keywords of the first voice text, match the text keywords with the default keywords in the preset keyword table to obtain a matching degree, and compare the matching degree with the preset threshold according to the matching degree. As a result, the first NLP engine or the second NLP engine is selected as the current NLP engine. If the dialogue state of the first speech text is multiple rounds of dialogue, the NLP engine used in the previous round of dialogue is determined as the current NLP engine; After the NLP engine, in the current NLP engine, semantically recognize the first voice text to obtain the first intent and the first corpus corresponding to the first voice text, and generate a control instruction corresponding to the first intent and control the instruction It is sent to the execution unit to perform the corresponding task, and the first corpus is output to the voice broadcast unit to broadcast the first corpus. Based on the above technical means, it can be quickly determined which NLP engine should be used for the first voice text by checking the dialogue state of the first voice text input by the user and the matching of the first voice text and the default keywords in the preset keyword table. The text is recognized and processed, which effectively avoids the conflict and confusion of the NLP engine in the voice interaction mode of multiple NLP engines, greatly reduces the arbitration time of the NPL engine, avoids the time delay and the slow response of the whole system, and greatly improves the voice interaction. user experience.

Other features and advantages of the invention will be set forth in the description which follows, and in part will be apparent from the description, or will be manifested by practice of the invention. The objectives and other advantages of the invention may be realized and attained by the description, claims and drawings.

Description of drawings

In order to explain the embodiments of the present invention or the technical solutions in the prior art more clearly, the following briefly introduces the accompanying drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only These are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained according to these drawings without creative efforts.

FIG. 1 is a flowchart of a voice interaction method in an embodiment of the present invention.

FIG. 2 is a frame diagram of a voice interaction system in another embodiment of the present invention.

Detailed ways

Various exemplary embodiments, features and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings. The same reference numbers in the figures denote elements that have the same or similar functions. While various aspects of the embodiments are shown in the drawings, the drawings are not necessarily drawn to scale unless otherwise indicated.

In addition, in order to better illustrate the present invention, numerous specific details are given in the following specific embodiments. It will be understood by those skilled in the art that the present invention may be practiced without certain specific details. In some instances, means well known to those skilled in the art have not been described in detail in order not to obscure the subject matter of the present invention.

An embodiment of the present invention proposes a voice interaction method, which adopts the pre-arbitration method of the NLP engine, that is, the arbitration is performed before entering the NLP engine for identification. FIG. 1 is a flowchart of the method described in this embodiment. The embodiment method includes the following steps S101-S104:

Step S101, obtaining a first voice command input by a user, and obtaining a first voice text according to the first voice command;

Exemplarily, the first voice command sent by the user's voice through the microphone can be converted into the first voice text by using an external automatic speech recognition (Automatic Speech Recognition, ASR) system to recognize and process the PCM signal of the first voice command sent by the user, Obtain the successfully converted first voice text through a preset interface.

Step S102, identifying the dialogue state of the first voice text;

Specifically, the dialogue state in voice interaction includes multi-turn dialogue and single-turn dialogue. Among them, the multiple rounds of dialogue are, for example, "very hot", "playing music", etc. The purpose of being very hot may be to open the sunroof, open the car window, turn on the air conditioner, etc. The purpose of "playing music" may be to play a certain song, In this case, the user's needs are more complex and there are many restrictions. It may be necessary to make statements in multiple rounds, and users can constantly modify or improve their own needs during the dialogue. The single-round dialogue is, for example, "open the sunroof" and "close the sunroof", which is an interactive dialogue of one question and one answer. In the step, if the dialogue state of the first voice text is recognized as multiple rounds of dialogue, then the first voice text is carried out for multiple rounds. Dialog marking, so that the current NLP engine can be locked according to the marking situation in subsequent steps.

Step S103, select one of the first NLP engine and the second NLP engine as the current NLP engine; wherein: if the dialogue state of the first voice text is a single-round dialogue, then extract the text keywords of the first voice text, and the text key The word is matched with the default keywords in the preset keyword table to obtain the matching degree, and one of the first NLP engine and the second NLP engine is selected as the current NLP engine according to the comparison result between the matching degree and the preset threshold; If the dialogue state of a voice text is multiple rounds of dialogue, the NLP engine used in the previous round of dialogue is determined as the current NLP engine;

Specifically, in step S103, the dialog state of the first voice text is first determined according to the marking status of step S102. If the dialog state of the first voice text is multiple rounds of dialog, no engine arbitration is performed, and the one used in the previous round of dialog is used. As the current NLP engine, the first voice text is sent to the current NLP engine through data transparent transmission.

If the dialogue state of the first voice text is a single-round dialogue, extract the text keywords of the first voice text. For example, if the first voice text is "turn on the air conditioner", the extracted text keyword is "air conditioner". It can be understood that, The text keyword is the word slot of the phonetic text; further, the extracted text keyword is matched with the default keyword in the preset keyword table to obtain the matching degree, and the preset keyword table stores a plurality of default keywords , and further, it is determined to select the first NLP engine or the second NLP engine as the current NLP engine according to the comparison result between the matching degree obtained by the matching and the preset threshold.

Step S104, the current NLP engine performs semantic recognition on the first voice text to obtain the first intent and the first corpus corresponding to the first voice text, and generates a control instruction corresponding to the first intent and sends the control instruction to the execution. The unit performs the corresponding task and outputs the first corpus to the voice broadcasting unit to broadcast the first corpus.

Specifically, after the arbitration is over, the current NLP engine will receive a start signal, and the current NLP engine performs semantic recognition on the first voice text according to the start signal to obtain the first intent and the first corpus corresponding to the first voice text, For example, the first voice text is "turn on the air conditioner", the first intention is to want to turn on the air conditioner, and the first corpus is "turn on the air conditioner", "turn on the air conditioner for you", "OK, the air conditioner is turning on", etc. The first corpus is preset, and the control instruction corresponding to the first intention is "start the air conditioner".

Wherein, the voice broadcasting unit is preferably a TTS (Text To Speech) system.

It can be seen from the description of the above embodiment that the application of the method of this embodiment can effectively improve the effectiveness of the reply, and avoid the conflict and confusion of the reply results, thereby greatly reducing the arbitration time of the NLP engine, avoiding the time delay and causing the whole system. The reply is slow, which greatly improves the user experience of voice interaction.

In a specific embodiment, in step S103, selecting one of the first NLP engine and the second NLP engine as the current NLP engine according to the comparison result between the matching degree and the preset threshold specifically includes:

When the matching degree is less than the preset threshold, select the first NLP engine as the current NLP engine;

When the matching degree is greater than or equal to the preset threshold, the second NLP engine is selected as the current NLP engine.

The matching degree refers to the same degree of the extracted keywords and the default keywords contained in the preset keyword table. It can be understood that the calculation of the matching degree can use any similarity between words. It is implemented by a degree calculation algorithm, so there is no specific limitation in this embodiment.

Exemplarily, the preset threshold may be 90%. When the matching degree obtained according to the matching is greater than or equal to 90%, the second NLP engine is selected as the current NLP engine; when the matching degree obtained according to the matching is less than 90%, the first NLP engine is selected. NLP engine as the current NLP engine.

In a specific embodiment, the method further includes steps S201 and S202;

Step S201, when selecting the second NLP engine as the current NLP engine, if the second NLP engine fails to perform semantic recognition on the first voice text and cannot obtain the first intent and the first corpus corresponding to the first voice text, then outputting the second corpus for prompting the user to teach the first voice instruction to the voice broadcast unit to broadcast the second corpus;

Specifically, the failure of the second NLP engine to perform semantic recognition on the first voice text means that the second NLP engine cannot recognize the semantics of the first voice text, does not know the intent of the first voice text, and there is no dynamic corpus. The corpus corresponding to the first voice text; the dynamic corpus is a corpus that can add corpus according to the user's personalized language habits, and the dynamic corpus may include user-defined corpus.

Exemplarily, the second corpus is "I don't understand this instruction yet, please teach me"; at this time, the voice playback unit plays the voice content of "I don't understand this instruction yet, please teach me".

Step S202, the second NLP engine determines the intent of the first voice text according to the teaching information input by the user, generates a control instruction corresponding to the intent of the first voice text, and sends the control instruction to the execution unit to execute the corresponding task.

Specifically, after getting the prompt of the second corpus, the user inputs a second voice instruction, that is, teaching information, and the intent corresponding to the second voice instruction can be obtained according to the second voice instruction input by the user, and the intent is the same as that of the first voice instruction. The intent of the speech text is consistent, and finally the second NLP engine wins the control instruction corresponding to the intent, and sends the control instruction to the execution unit to execute the corresponding task.

It should be noted that, in many cases, no valid reply can be found in the dynamic corpus of the second NLP engine, so the reply cannot be successful. If the second NLP engine replies "I don't understand this command yet, please teach me", it will lock the second NLP engine as the engine for recognizing speech and text within a certain period, and enter the multi-round dialogue process taught by the user. Interaction with the user completes incomprehensible questions until the system-defined period ends or the user refuses to cooperate with teaching.

In a specific embodiment, the second NLP engine in step S202 determines the intention of the first voice text according to the teaching information input by the user, which specifically includes step S301, step S302 and step S303;

Step S301, the second NLP engine obtains the second voice text, performs semantic recognition on the second voice text to obtain the second intent, and generates a message for prompting the user to confirm whether the intentions of the first voice text and the second voice text are consistent. The third corpus; wherein the second voice text is obtained according to the second voice command input by the user;

Specifically, after outputting the second corpus that prompts the user to teach the first voice instruction, the second voice instruction input by the user is acquired, and the second voice text is obtained according to the second voice instruction, and the second voice text is obtained. The method is the same as that of the first voice text; the second NLP engine performs semantic recognition on the second voice text to obtain the second intention, and generates a third corpus. If the voice text is to turn on the air conditioner, the third corpus information is, for example, "Does turning on the air conditioner mean turning on the air conditioner?".

Step S302, sending the third corpus to the voice playback unit to play the third corpus;

Step S303, the second NLP engine obtains the confirmation information input by the user, and when the confirmation information is consistent with the confirmation, generates a control instruction corresponding to the second intention, and sends the control instruction to the execution unit to execute the corresponding task.

Specifically, after being prompted by the third corpus, the user inputs confirmation information, such as "confirm the agreement", through a voice method or a physical input unit. After obtaining the confirmation information from the user, it is determined whether the confirmation information is consistent with the confirmation. If the confirmation is consistent, a control command corresponding to the second intention is generated. For example, the second voice text is to turn on the air conditioner, the execution unit is the air conditioner, and the control command to start the air conditioner. If it is confirmed that they are inconsistent, the control instruction corresponding to the second intention is not generated, and the user is further requested to perform teaching. Specifically, the process of steps S301 to S303 may be re-executed.

In a specific embodiment, the method further includes:

When the confirmation information is to confirm that the intentions of the first voice text and the second voice text are consistent, a mapping relationship is established between the first voice text and the second voice text, and the first voice text is added as a new corpus into the dynamic corpus.

Specifically, after establishing a mapping relationship between the first voice text and the second voice text, and adding the first voice text as a new corpus to the dynamic corpus table, the second NLP engine learns the first voice instruction, when the user sends the first voice instruction again, the second NLP engine can effectively identify and reply to the first voice text corresponding to the first voice instruction.

It should be noted that this embodiment proposes to perform incremental learning of the local corpus under the condition that the second NLP engine cannot recognize the first speech text, which can continuously expand the content of the dynamic corpus and the ability of front-end arbitration, so that the entire human The computer interaction system can continuously learn based on the scene and the user's personalized terminology, and realize the "thousands of people and thousands of faces" of human-computer interaction.

In a specific embodiment, the first NLP engine is an online NLP engine, which includes an offline corpus; the second NLP engine is a local NLP engine, which includes a dynamic corpus; several corpora in the offline corpus has a mapping relationship with several corpora in the dynamic corpus;

The method also includes:

After the offline corpus of the first NLP engine is updated online, the preset keyword table is compared and updated according to the updated corpus of the offline corpus to avoid repetition of keywords in the keyword table.

Among them, the construction of the preset keyword table in this embodiment is established by the offline corpus of the first NLP engine and the dynamic corpus of the second NLP engine through text mapping, and has the capability of dynamic update and continuous expansion.

It can be understood that the second NLP engine in this embodiment can be used as a supplement to online intelligent voice interaction, improve the hit rate of voice interaction problems in subdivided scenarios, enhance the accuracy and effectiveness of responses, and bring better results to users. Voice experience; increase the availability of intelligent voice interaction, whether online or offline, or difficult to understand voice problems, the system can effectively handle.

It should be noted that the sequence numbers of the steps in this document are only for distinguishing different steps, and do not constitute a limitation on the sequence of the steps. The actual execution sequence of the steps should be determined according to the possible solution of the overall technical solution.

Another embodiment of the present invention also proposes a voice interaction system. FIG. 2 is a frame diagram of the system according to the embodiment. Referring to FIG. 2 , the system according to this embodiment includes:

A voice text obtaining unit 1, configured to obtain a first voice command input by a user, and obtain a first voice text according to the first voice command;

Dialogue management unit 2, for identifying the dialogue state of the first voice text;

Arbitration unit 3, for selecting one of the first NLP engine and the second NLP engine as the current NLP engine; wherein: if the dialogue state of the first voice text is a single-round dialogue, then extract the text keywords of the first voice text, the The text keyword is matched with the default keyword in the preset keyword table to obtain a matching degree, and one of the first NLP engine and the second NLP engine is selected as the current NLP engine according to the comparison result between the matching degree and the preset threshold ; If the dialogue state of the first voice text is multiple rounds of dialogue, then determine the NLP engine adopted in the previous round of dialogue as the current NLP engine;

The first NLP engine 4 and the second NLP engine 5 are used to perform semantic recognition on the first speech text to obtain the first intent and the first corpus corresponding to the first speech text when they are used as the current NLP engine, and generate the same The control instruction corresponding to the first intent is sent to the execution unit to perform the corresponding task, and the first corpus is output to the voice broadcasting unit 20 to broadcast the first corpus.

In a specific embodiment, the arbitration unit 3 is specifically used for:

When the matching degree is less than the preset threshold, select the first NLP engine as the current NLP engine corpus;

When the matching degree is greater than or equal to the preset threshold, the second NLP engine is selected as the current NLP engine corpus.

In a specific embodiment, the second NLP engine 5 is also used for:

When the second NLP engine 5 is used as the current NLP engine, if the second NLP engine 5 fails to perform semantic recognition on the first voice text, and cannot obtain the first intent and the first corpus corresponding to the first voice text, the output will use in prompting the user to perform the second corpus taught by the first voice instruction to the voice broadcasting unit 20 to broadcast the second corpus;

Moreover, the second NLP engine 5 determines the intent of the first voice text according to the teaching information input by the user, generates a control instruction corresponding to the intent of the first voice text, and sends the control instruction to the execution unit to execute the corresponding task.

In a specific embodiment, the second NLP engine 5 is specifically used for:

Acquiring the second voice text, performing semantic recognition on the second voice text to obtain the second intent, generating a third corpus for prompting the user to confirm whether the intentions of the first voice text and the second voice text are consistent, and converting the third voice text The corpus is sent to the voice playback unit 20 to play the third corpus, and the confirmation information input by the user is obtained, when the confirmation information is consistent with the confirmation, a control command corresponding to the second intention is generated, and the control command is sent to the execution unit to perform corresponding tasks;

The voice-text obtaining unit 1 is further configured to obtain the second voice-text according to the second voice instruction input by the user and send it to the second NLP engine.

In a specific embodiment, the first NLP engine 4 is an online NLP engine, which includes an offline corpus; the second NLP engine 5 is a local NLP engine, which includes a dynamic corpus; Several corpora have a mapping relationship with several corpora in the dynamic corpus;

The second NLP engine 5 is also used to establish a mapping relationship between the first voice text and the second voice text when the confirmation information is to confirm that the intention of the first voice text and the second voice text is consistent, and to The first voice text is added to the dynamic corpus table as a new corpus;

The system also includes a keyword table updating unit 6, which is used to update all the keywords according to the updated offline corpus after the offline corpus of the first NLP engine is updated online. The preset keyword table is compared and updated.

The system embodiments described above are only illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in One place, or it can be distributed over multiple network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution in this embodiment.

It should be noted that the system described in the foregoing embodiment corresponds to the method described in the foregoing embodiment. Therefore, the undescribed part of the system described in the foregoing embodiment can be obtained by referring to the content of the method described in the foregoing embodiment, which will not be repeated here.

Moreover, if the voice interaction system described in the above embodiments is implemented in the form of software functional units and sold or used as an independent product, it can be stored in a computer-readable storage medium.

Another embodiment of the present invention further provides a voice interaction device, including: the voice interaction system according to the above embodiment; or, a memory and a processor, where computer-readable instructions are stored in the memory, and the computer-readable instructions When the instructions are executed by the processor, the processor is caused to execute the steps of the voice interaction method according to the above embodiment.

Of course, the voice control device may also have components such as a wired or wireless network interface, a keyboard, and an input/output interface for input and output, and the voice control device may also include other components for realizing device functions, which will not be repeated here. .

Exemplarily, the computer program may be divided into one or more units, and the one or more units are stored in the memory and executed by the processor to accomplish the present invention. The one or more units may be a series of computer program instruction segments capable of performing specific functions, and the instruction segments are used to describe the execution process of the computer program in the voice control device.

The processor may be a central processing unit (Central Processing Unit, CPU), other general-purpose processors, a digital signal processor (Digital Signal Processor, DSP), an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), an off-the-shelf processor Programmable Gate Array (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The processor is the control center of the voice control device, and uses various interfaces and lines to connect the entire voice control device. various parts.

The memory can be used to store the computer program and/or unit, and the processor implements the voice by running or executing the computer program and/or unit stored in the memory and calling the data stored in the memory. Control various functions of the device. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory such as hard disk, internal memory, plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card , a flash memory card (Flash Card), at least one magnetic disk storage device, flash memory device, or other volatile solid-state storage device.

Various embodiments of the present invention have been described above, and the foregoing descriptions are exemplary, not exhaustive, and not limiting of the disclosed embodiments. Numerous modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The terminology used herein was chosen to best explain the principles of the embodiments, the practical application or technical improvement in the marketplace, or to enable others of ordinary skill in the art to understand the embodiments disclosed herein.

Claims (12)

1. A method of voice interaction, comprising:

acquiring a first voice instruction input by a user, and acquiring a first voice text according to the first voice instruction;

recognizing a dialog state of the first speech text;

selecting one of the first NLP engine and the second NLP engine as a current NLP engine; wherein: if the dialogue state of the first voice text is single-turn dialogue, extracting text keywords of the first voice text, matching the text keywords with default keywords in a preset keyword table to obtain matching degree, and selecting one of the first NLP engine and the second NLP engine as a current NLP engine according to a comparison result of the matching degree and a preset threshold; if the conversation state of the first voice text is a multi-turn conversation, determining an NLP engine adopted in the previous turn of conversation as a current NLP engine;

the current NLP engine carries out semantic recognition on the first voice text to obtain a first intention and a first language material corresponding to the first voice text, generates a control instruction corresponding to the first intention, sends the control instruction to an execution unit to execute a corresponding task, and outputs the first language material to a voice broadcasting unit to broadcast the first language material.

2. The method of claim 1, wherein selecting one of the first NLP engine and the second NLP engine as the current NLP engine according to the comparison result between the matching degree and the preset threshold specifically comprises:

when the matching degree is smaller than a preset threshold value, selecting a first NLP engine as a current NLP engine;

and when the matching degree is greater than or equal to a preset threshold value, selecting the second NLP engine as the current NLP engine.

3. The voice interaction method of claim 1, further comprising:

when a second NLP engine is selected as a current NLP engine, if the second NLP engine fails to perform semantic recognition on the first voice text and cannot obtain a first intention and a first language material corresponding to the first voice text, outputting a second language material for prompting a user to perform first voice instruction teaching to a voice broadcasting unit so as to broadcast the second language material;

and the second NLP engine determines the intention of the first voice text according to the teaching information input by the user, generates a control instruction corresponding to the intention of the first voice text, and sends the control instruction to the execution unit to execute the corresponding task.

4. The speech interaction method of claim 3, wherein the determining, by the second NLP engine, the intention of the first speech text according to the teaching information input by the user specifically comprises:

the second NLP engine acquires a second voice text, performs semantic recognition on the second voice text to obtain a second intention, and generates a third corpus for prompting a user to confirm whether the intentions of the first voice text and the second voice text are consistent; the second voice text is obtained according to a second voice instruction input by a user;

sending the third corpus to a voice playing unit to play the third corpus;

and the second NLP engine acquires confirmation information input by the user, generates a control instruction corresponding to the second intention when the confirmation information is consistent, and sends the control instruction to the execution unit to execute the corresponding task.

5. The voice interaction method of claim 4, further comprising:

and when the confirmation information is that the intention of confirming the first voice text is consistent with that of confirming the second voice text, establishing a mapping relation between the first voice text and the second voice text, and adding the first voice text serving as a newly added corpus into a dynamic corpus table.

6. The method of any one of claims 1-5, wherein the first NLP engine is an online NLP engine comprising an offline corpus table; the second NLP engine is a local NLP engine and comprises a dynamic corpus table; the plurality of linguistic data in the off-line linguistic data table and the plurality of linguistic data in the dynamic linguistic data table have a mapping relation;

the method further comprises the following steps:

and after the off-line corpus table of the first NLP engine is updated on line, comparing and updating the preset keyword table according to the corpus of the updated off-line corpus table.

7. A voice interaction system, comprising:

the voice text acquisition unit is used for acquiring a first voice instruction input by a user and acquiring a first voice text according to the first voice instruction;

the dialogue management unit is used for recognizing the dialogue state of the first voice text;

an arbitration unit, configured to select one of the first NLP engine and the second NLP engine as a current NLP engine; wherein: if the dialogue state of the first voice text is single-turn dialogue, extracting text keywords of the first voice text, matching the text keywords with default keywords in a preset keyword table to obtain matching degree, and selecting one of the first NLP engine and the second NLP engine as a current NLP engine according to a comparison result of the matching degree and a preset threshold; if the conversation state of the first voice text is a multi-turn conversation, determining an NLP engine adopted in the previous turn of conversation as a current NLP engine;

the first NLP engine and the second NLP engine are used for performing semantic recognition on the first voice text to obtain a first intention and a first corpus corresponding to the first voice text when the first NLP engine and the second NLP engine are used as the current NLP engine, generating a control instruction corresponding to the first intention, sending the control instruction to the execution unit to execute a corresponding task, and outputting the first corpus to the voice broadcasting unit to broadcast the first corpus.

8. The voice interaction system of claim 7, wherein the arbitration unit is specifically configured to:

when the matching degree is smaller than a preset threshold value, selecting a first NLP engine as a current NLP engine corpus;

and when the matching degree is greater than or equal to a preset threshold value, selecting a second NLP engine as the current NLP engine corpus.

9. The voice interaction system of claim 7, wherein the second NLP engine is further configured to:

when the second NLP engine is used as the current NLP engine, if the semantic recognition of the first voice text by the second NLP engine fails and the first intention and the first language material corresponding to the first voice text cannot be obtained, outputting a second language material for prompting a user to teach the first voice instruction to a voice broadcasting unit to broadcast the second language material;

and the second NLP engine determines the intention of the first voice text according to the teaching information input by the user, generates a control instruction corresponding to the intention of the first voice text, and sends the control instruction to the execution unit to execute the corresponding task.

10. The voice interaction system of claim 9, wherein the second NLP engine is specifically configured to:

the method comprises the steps of obtaining a second voice text, carrying out semantic recognition on the second voice text to obtain a second intention, generating a third corpus for prompting a user to confirm whether the intentions of the first voice text and the second voice text are consistent or not, sending the third corpus to a voice playing unit to play the third corpus, obtaining confirmation information input by the user, generating a control instruction corresponding to the second intention when the confirmation information is consistent, and sending the control instruction to an execution unit to execute a corresponding task;

the voice text acquisition unit is also used for acquiring a second voice text according to a second voice instruction input by the user and sending the second voice text to the second NLP engine.

11. The system of claim 10, wherein the first NLP engine is an online NLP engine, which includes an offline corpus table; the second NLP engine is a local NLP engine and comprises a dynamic corpus table; the plurality of linguistic data in the off-line linguistic data table and the plurality of linguistic data in the dynamic linguistic data table have a mapping relation;

the second NLP engine is further used for establishing a mapping relation between the first voice text and the second voice text when the confirmation information is that the intentions of confirming the first voice text and the second voice text are consistent, and adding the first voice text serving as a new corpus into a dynamic corpus table;

the system further comprises a keyword list updating unit, wherein the keyword list updating unit is used for comparing and updating the preset keyword list according to the language material of the updated offline language material list after the offline language material list of the first NLP engine is updated on line.

12. A voice interaction device, comprising: the voice interaction system of any of claims 7-11; or a memory and a processor, the memory having stored therein computer readable instructions which, when executed by the processor, cause the processor to perform the steps of the voice interaction method according to any one of claims 1-6.

Images