Multi-modal fusion dialogue management method and device, electronic equipment and medium

Abstract

The invention discloses a multi-modal fusion dialogue management method and device, electronic equipment and a medium. The method comprises the steps of collecting multi-modal data in a preset target scene; wherein the multi-modal data is composed of audio data, video data and touch data; identifying the multi-modal data to obtain target data; wherein the target data is data used for a dialogue management process; and if the target data meets the preset dialogue condition, generating an answer according to the target data. According to the technical scheme, the method has the precise ambiguity resolution capability, contexts can be guaranteed to be coherent and smooth, and the interaction process has the self-adaptive characteristic, safety and reliability.

Description

Technical Field

[0001] This invention relates to the field of dialogue management technology, and in particular to a multimodal dialogue management method, apparatus, electronic device, and medium. Background Technology

[0002] With the development of human-computer interaction technology, dialogue systems are widely used in real-world scenarios such as smart homes and in-vehicle terminals, and the fluency and accuracy of their interaction directly affect the user experience.

[0003] However, existing dialogue systems still face many pressing technical challenges in practical applications: First, intent recognition is ambiguous. Users often use referential language during interaction, and relying solely on voice information cannot accurately determine the target of the operation, leading to the system's inability to respond correctly to commands. Second, context is fragmented. Traditional dialogue state tracking methods only process text sequences, ignoring non-verbal cues such as visual and behavioral cues, making it difficult to fully understand the user's interaction intent. Third, interaction strategies are rigid. The system's interaction actions (such as command confirmation and task execution) are usually based on static rules, making it impossible to dynamically adjust the interaction logic according to the user's patience level or the complexity of the task. Fourth, invalid requests are rampant. The system lacks an effective filtering mechanism for non-human voice input, access from unregistered users, and input in non-target languages, which easily leads to false triggering problems and wastes computing resources.

[0004] The aforementioned problems severely restrict the interactive performance and practical value of dialogue systems. Therefore, it is urgent to propose a multi-turn dialogue interaction optimization scheme that can solve these problems. Summary of the Invention

[0005] This invention provides a multimodal fusion dialogue management method, device, electronic device and medium, which has accurate ambiguity resolution capability, can ensure contextual coherence and fluency, and the interaction process has both adaptive characteristics and security and reliability.

[0006] According to one aspect of the present invention, a multimodal fusion dialogue management method is provided, the method comprising:

[0007] Multimodal data is collected in a preset target scenario; wherein the multimodal data consists of audio data, video data, and touch data;

[0008] The multimodal data is identified to obtain target data; wherein, the target data is data used for the dialogue management process;

[0009] If the target data meets the preset dialogue conditions, then an answer is generated based on the target data.

[0010] According to another aspect of the present invention, a multimodal fusion dialogue management device is provided, the device comprising:

[0011] A multimodal data acquisition module is used to acquire multimodal data in a preset target scene; wherein the multimodal data consists of audio data, video data, and touch data;

[0012] The target data acquisition module is used to identify the multimodal data to obtain target data; wherein, the target data is data used for the dialogue management process;

[0013] The answer generation module is used to generate an answer based on the target data if the target data meets preset dialogue conditions.

[0014] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0015] At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform a multimodal fusion dialogue management method according to any embodiment of the present invention.

[0016] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement a multimodal fusion dialogue management method as described in any embodiment of the present invention.

[0017] The technical solution of this invention involves collecting multimodal data in a preset target scenario, then identifying the multimodal data to obtain target data. If the target data meets preset dialogue conditions, an answer is generated based on the target data. This technical solution possesses accurate ambiguity resolution capabilities, ensures smooth and coherent context, and the interaction process combines adaptive characteristics with security and reliability.

[0018] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1This is a flowchart of a multimodal fusion dialogue management method provided according to Embodiment 1 of the present invention;

[0021] Figure 2 This is a schematic diagram of a multimodal fusion dialogue management process provided in Embodiment 2 of the present invention;

[0022] Figure 3 A schematic diagram of the dialogue management system provided in Embodiment 2 of this application;

[0023] Figure 4 A schematic diagram of the joint intent recognition unit provided in Embodiment 2 of this application;

[0024] Figure 5 This is a schematic diagram of a multi-turn dialogue interaction provided in Embodiment 2 of this application;

[0025] Figure 6 This is a flowchart of the rejection mechanism provided in Embodiment 2 of this application;

[0026] Figure 7 This is a schematic diagram of the structure of a multimodal fusion dialogue management device provided in Embodiment 3 of the present invention;

[0027] Figure 8 This is a schematic diagram of the structure of an electronic device that implements a multimodal fusion dialogue management method according to an embodiment of the present invention. Detailed Implementation

[0028] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0029] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0030] Example 1

[0031] Figure 1 This is a flowchart of a multimodal fusion dialogue management method according to Embodiment 1 of the present invention. This embodiment is applicable to dialogue management scenarios. The method can be executed by a multimodal fusion dialogue management device, which can be implemented in hardware and / or software and can be configured in a device. For example, the device can be a backend server or other device with communication and computing capabilities. Figure 1 As shown, the method includes:

[0032] S110. Collect multimodal data in a preset target scene; wherein the multimodal data consists of audio data, video data and touch data.

[0033] In this embodiment, the target scenario is a family scenario, which refers to an environmental scenario that takes a family living space (such as a living room, bedroom, kitchen, study, etc.) as a carrier and covers the daily living, entertainment, learning, housework and other activities of family members.

[0034] Audio data is digital data that uses sound wave signals as a carrier to record the physical characteristics of sound, such as frequency, amplitude, and duration.

[0035] In this embodiment, the video data consists of a sequence of consecutive frame images, which is structured / semi-structured data combining visual imagery and temporal information.

[0036] In this solution, touch data is time-series data that records parameters such as position, pressure, duration, and trajectory of touch actions during human-computer interaction.

[0037] For example, in a home setting, family members sit around watching a smart TV, switching channels via voice commands and adjusting volume using a touch remote. The corresponding multimodal data includes three categories: first, audio data, including the sound of TV programs playing, voice commands from family members such as "turn on the children's channel," and conversations between them; second, video data, consisting of family members' postures, facial expressions, and gestures such as pointing at the screen, captured by a camera; and third, touch data, specifically the pressing signals of the volume and channel switching buttons on the remote, as well as click operations such as manually selecting programs on the smart TV touchscreen.

[0038] Specifically, multimodal data is collected using pre-configured smart terminals in the target scenario.

[0039] Optionally, multimodal data can be collected in a preset target scenario, including:

[0040] Multimodal data is collected in a preset target scenario based on a pre-installed smart terminal; wherein the smart terminal consists of a microphone array and a camera.

[0041] In this solution, the smart terminal can be a home service robot, a smart speaker with a camera, etc.

[0042] In this embodiment, the hardware configuration of the smart terminal is limited to a microphone array and a camera. The microphone array is responsible for collecting audio data (such as ambient sound and voice signals), and the camera is responsible for collecting video data (such as images and video streams).

[0043] Furthermore, based on user touch operation signals (such as clicks and swipes on the terminal screen) received by the smart terminal or associated devices, touch data of user-initiated interaction can be obtained.

[0044] By collaboratively collecting multimodal data from preset target scenarios, more comprehensive and accurate information support can be provided for subsequent perception, analysis, and decision-making.

[0045] S120. The multimodal data is identified to obtain target data; wherein, the target data is data used for the dialogue management process.

[0046] In this embodiment, the target data is the data of the user dialogue management process, which includes text information and visual information.

[0047] In this solution, a joint intent recognition model can be used to identify multimodal data and obtain target data. The joint intent recognition model is a deep learning model that integrates multimodal features and can simultaneously perform intent classification and semantic element extraction. Its core function is to perform unified semantic understanding and intent determination on the input multimodal data, ultimately outputting target data that accurately represents the user's or device's target needs.

[0048] Furthermore, the joint intent recognition model designs dedicated feature extraction branches for different types of input data.

[0049] The audio data feature extraction branch extracts acoustic and linguistic features from audio data.

[0050] Furthermore, the audio data feature extraction branch's processing flow includes: Preprocessing: Noise reduction, framing, and windowing of the audio data to transform continuous sound wave signals into discrete frame sequences, addressing the non-stationarity of the audio data. Acoustic Feature Extraction: Extracting Mel-frequency cepstral coefficients (MFCC), linear predictive cepstral coefficients (LPCC), or Mel spectrograms. These features simulate the human ear's perception of sound and effectively characterize information such as pitch, loudness, and formants of speech. For noisy scenarios, wavelet transform or spectral subtraction can be introduced to further enhance feature robustness. Language Feature Extraction: Integrating lightweight speech recognition models (such as CNN-CTC, Transformer-ASR) to transform acoustic features into text token sequences, and then extracting semantic features through pre-trained language models (such as BERT-Small) to directly associate them with the user's spoken intent (such as "turn on the lights" or "adjust the volume"). Feature Fusion and Dimensionality Reduction: Weighted fusion of acoustic and language features using an attention mechanism, and then mapping to text information of a unified dimension through a fully connected layer.

[0051] In this embodiment, the video data feature extraction branch focuses on static features (such as gestures and facial expressions) and dynamic features (such as action sequences) in the visual modality to capture the user's non-verbal intentions (such as waving to wake up or nodding to confirm).

[0052] The video data feature extraction branch's processing flow includes: Preprocessing: Cropping, normalizing, and de-jittering the video data to extract keyframes (e.g., using inter-frame difference or optical flow methods). Static visual feature extraction: Using lightweight convolutional neural networks (e.g., MobileNet, ShuffleNet) to extract features from keyframes, capturing static features such as gestures and facial muscle changes. Dynamic visual feature extraction: Using recurrent neural networks (RNN), long short-term memory networks (LSTM), or temporal convolutional networks (TCN), temporal modeling of the static features of consecutive keyframes is performed to capture the evolution of actions (e.g., the intention corresponding to a continuous gesture sequence of "waving → clenching a fist → pointing"). Spatiotemporal feature fusion: Using a spatiotemporal attention mechanism, static and dynamic features are weighted and fused to obtain visual information strongly correlated with the intention (e.g., core joints of gestures and key muscle groups of facial expressions).

[0053] In this solution, the touch data feature extraction branch extracts features related to the operation intent from the input data of the touch device (such as the click, swipe, pressure, and trajectory coordinate sequence of the touch screen). For example, "clicking the icon" corresponds to the intent to launch the application, and "two-finger zoom" corresponds to the intent to adjust the size of the interface.

[0054] Furthermore, the touch data feature extraction branch processing flow includes: Data structuring: Parsing touch data into multi-dimensional temporal data, including spatiotemporal features (click coordinates (x, y), start / end / path of the swipe trajectory, operation timestamp) and physical features (press pressure value, contact area). Manual feature extraction: Calculating the geometric features of the trajectory (such as swipe distance, direction angle, curvature), temporal features (such as operation duration, click frequency, swipe speed), and interaction features (such as single-finger and multi-finger operations, number of consecutive clicks). These features can directly characterize the behavioral patterns of touch operations. Deep feature extraction: Using a one-dimensional convolutional neural network (1D-CNN) or LSTM to model the structured temporal touch data, automatically mining hidden deep features (such as the intent corresponding to the compound operation of "rapid swipe → pause → click"). Feature filtering: Filtering out textual information highly related to the intent label through Pearson correlation coefficient or mutual information calculation.

[0055] S130. If the target data meets the preset dialogue conditions, then an answer is generated based on the target data.

[0056] Among them, the preset dialogue conditions are the constraints imposed on the target data, such as the voice identifier in the target data being a human voice identifier, or the voiceprint similarity being greater than or equal to a preset threshold.

[0057] In this approach, it is determined whether the target data meets the preset dialogue conditions. If it does, an answer is generated based on the target data.

[0058] Specifically, the system performs prediction processing on the target data based on a preset model and outputs the corresponding answer. The preset model can be a deep learning model, including but not limited to network architectures such as convolutional neural networks (CNN), recurrent neural networks (RNN), and Transformer. The model's network layer count, number of neurons, activation function, and other parameters can be adjusted and optimized according to the accuracy and efficiency requirements of the actual application scenario.

[0059] Optionally, generating an answer based on the target data includes:

[0060] The text information in the target data is encoded to obtain a first encoding; and the visual information in the target data is encoded to obtain a second encoding.

[0061] The first code and the second code are fused to generate the intent distribution probability;

[0062] The intent distribution probability is input into a pre-configured dialogue state tracking model, and the corresponding intent is output through the dialogue state tracking model.

[0063] Generate an answer that matches the stated intent.

[0064] In this scheme, the text information in the target data is encoded using BERT to obtain the first encoding. The video information in the target data is encoded to obtain the second encoding. The dimension of the vector is d.

[0065] Furthermore, a gating fusion method is adopted to fuse the first and second codes to generate an intent distribution probability. For example, the probability of turning on the air conditioner is 0.85, the probability of adjusting the temperature is 0.12, and the probability of closing the curtains is 0.03.

[0066] Specifically, , ;in, As weight, The Sigmoid activation function outputs values ​​between (0,1), which are used as the gating coefficients. and Here are the weight matrix and bias terms of the gated network. Let be the probability distribution of the intention.

[0067] Among them, the core function of the Dialogue State Tracking (DST) model is to accurately determine the user's core intent by combining the context and the current input.

[0068] In this embodiment, the intent distribution probability is input into a pre-configured dialogue state tracking model, and the corresponding intent is output through the dialogue state tracking model, that is, the intent distribution probability is converted into an executable intent. For example, the intent is to turn on the air conditioner.

[0069] Furthermore, the answer generation template corresponding to the intent is invoked, and key information related to the intent is filled into the answer generation template to generate the answer.

[0070] By integrating textual and visual encoding of target data and combining it with a dialogue state tracking model to accurately identify intent, it is possible to generate answers that highly match the intent, thereby improving the accuracy and relevance of answer generation in multimodal interaction scenarios.

[0071] The technical solution of this invention collects multimodal data in a preset target scenario, then identifies the multimodal data to obtain target data; if the target data meets preset dialogue conditions, an answer is generated based on the target data. By implementing this technical solution, it has accurate ambiguity resolution capabilities, ensures smooth and coherent context, and the interaction process combines adaptive characteristics with security and reliability.

[0072] Example 2

[0073] Figure 2This is a schematic diagram of a multimodal fusion dialogue management process provided in Embodiment 2 of the present invention. The relationship between this embodiment and the above embodiments is a detailed description of the dialogue management process. Figure 2 As shown, the method includes:

[0074] S210. Collect multimodal data in a preset target scene; wherein the multimodal data consists of audio data, video data and touch data.

[0075] S220. The multimodal data is identified to obtain target data; wherein, the target data is data used for the dialogue management process.

[0076] S230. If the target data does not meet the preset dialogue conditions, a refusal to answer prompt is generated.

[0077] In this embodiment, a "no-answer" message refers to a standardized statement used to clearly inform the user that an answer cannot be provided. For example, a "no-answer" message could be: "I'm sorry, I cannot provide the answer to the operation you described."

[0078] Among them, the preset dialogue conditions are the constraints imposed on the target data, such as the voice identifier in the target data being a human voice identifier, or the voiceprint similarity being greater than or equal to a preset threshold.

[0079] In this solution, it is determined whether the target data meets the preset dialogue conditions. If not, a rejection message is generated.

[0080] Optionally, if the target data does not meet the preset dialogue conditions, a rejection message is generated, including:

[0081] If the voice identifier in the target data is a preset voice identifier, a rejection message will be generated.

[0082] The preset sound identifier is configured to 0, which represents non-human voice attributes.

[0083] Specifically, when the sound identifier in the target data When this happens, a rejection message is generated. For example, if the sound in the target data is detected as television sound, its corresponding sound identifier is... At this point, a message indicating that the answer is refused will be generated.

[0084] By pre-matching the sound identifiers of the target data and triggering a rejection mechanism, query requests corresponding to unexpected sound data can be effectively filtered out, improving the security and accuracy of interactive responses.

[0085] Optionally, if the target data does not meet the preset dialogue conditions, a refusal to answer prompt is generated, which further includes:

[0086] If the voiceprint similarity in the target data is less than a preset first threshold, a refusal to answer prompt is generated.

[0087] The preset first threshold can be set according to the needs of dialogue management. Preferably, the first threshold is set to 0.75.

[0088] In this scheme, the text information in the target data is matched with the text template constructed based on the voice features of pre-registered users on the smart terminal through a text-voiceprint association mapping mechanism, and the similarity is calculated and the voiceprint similarity corresponding to the voiceprint features is output.

[0089] Furthermore, it is determined whether the voiceprint similarity is greater than or equal to a preset first threshold. If the voiceprint similarity is less than the preset first threshold, a refusal to answer prompt is generated. For example, if a visitor is detected speaking, and the voiceprint similarity is less than the first threshold, a refusal to answer prompt is generated.

[0090] By using a voiceprint similarity threshold determination mechanism to block unauthorized voiceprint input, interaction requests that do not meet the required permissions can be directly rejected, effectively ensuring the access security of the voice interaction system.

[0091] Optionally, if the target data does not meet the preset dialogue conditions, a refusal to answer prompt is generated, which further includes:

[0092] If the language information in the target data is a preset language and the confidence level in the target data is less than a preset second threshold, then a rejection message is generated.

[0093] The preset language information needs to be configured according to the dialogue management requirements. For example, the preset language can be set to a language other than Chinese.

[0094] In this embodiment, the preset second threshold can be set according to the needs of dialogue management. Preferably, the second threshold is set to 0.6.

[0095] Among them, confidence level refers to the matching confidence level when the language information to be identified matches the preset language information.

[0096] In this solution, it is determined whether the language information in the target data is Chinese, and whether the confidence level in the target data is greater than or equal to a preset second threshold. If the language information in the target data is the preset language and the confidence level in the target data is less than the preset second threshold, a refusal to answer prompt is generated. For example, if the system detects that the user is speaking English and the confidence level is less than the second threshold, a refusal to answer prompt is generated.

[0097] In this embodiment, no voice or interface feedback is generated after a refusal to answer, making it suitable for sensitive scenarios such as nighttime and meetings.

[0098] When the target data is in a preset language and the confidence level is lower than a preset second threshold, a rejection message is generated. This can accurately block unexpected language input with low confidence, ensuring the rigor and accuracy of the interaction logic.

[0099] The technical solution of this invention collects multimodal data in a preset target scenario, then identifies the multimodal data to obtain target data; if the target data does not meet the preset dialogue conditions, a rejection message is generated. By implementing this technical solution, it possesses secure and reliable characteristics, and its rejection mechanism ensures that responses are only made to legitimate, valid, and meaningful user input.

[0100] In this plan, Figure 3 This is a schematic diagram of the dialogue management system provided in Embodiment 2 of this application, as shown below. Figure 3 As shown, the core process of the system is as follows: the multimodal acquisition unit simultaneously acquires audio, video, and touch signals and transmits them to the feature extraction engine; this engine has a built-in ASR submodule (based on Whisper-large-v3 to output text and language tags), a visual encoder (based on YOLOv8 and MediaPipe to output object ID, coordinates, and gestures), a behavior analyzer (based on OpenFace to output facial expression valence), and a VAD detection submodule (based on Silero VAD to output human voice identification). The extracted multi-dimensional features are simultaneously sent to the joint intent recognition unit and the multi-dimensional rejection decision unit; after the joint intent recognition unit completes the cross-modal fusion of speech and visual semantics, it transmits the results to the DST unit, which uses GNN to verify consistency to maintain the structured state slots and outputs them to the dynamic policy decision-maker; the dynamic policy decision-maker selects the optimal action based on PPO reinforcement learning, drives the response generation unit to generate TTS speech, screen prompts, and device commands, and executes external devices; at the same time, the user feedback analysis unit calculates relevant feedback parameters. The multi-dimensional rejection decision unit completes the four-dimensional rejection judgment, and the results of both are fed back to the dynamic strategy decision-maker. If the multi-dimensional rejection decision unit determines that a rejection is required, it directly blocks the process.

[0101] Furthermore, Figure 4 This is a schematic diagram of the joint intent recognition unit provided in Embodiment 2 of this application, as shown below. Figure 4 As shown, a gating fusion mechanism is used to fuse text encoding and visual encoding to generate the intent distribution probability.

[0102] In this embodiment, Figure 5 This is a schematic diagram of a multi-turn dialogue interaction provided in Embodiment 2 of this application, such as... Figure 5As shown, taking "lowering the air conditioner temperature" as an example, the complete collaborative process of multimodal interaction is demonstrated: the user initiates a control request through "voice command + gesture pointing at the air conditioner", the smart speaker accurately locates the controlled object by combining visual recognition and voice input, and then clarifies the specific temperature requirements to the user through voice and screen display. After the user confirms the target temperature by "voice + nod", the speaker executes the command to adjust the air conditioner and simultaneously provides feedback on the execution result through voice and screen. The whole process integrates multiple interaction methods such as voice, vision, and gesture, demonstrating the collaboration and efficiency of multimodal technology in smart home scenarios.

[0103] Furthermore, Figure 6 Here is a flowchart of the rejection mechanism provided in Embodiment 2 of this application, as follows: Figure 6 As shown, two independent judgment paths are used to determine whether to trigger silent ignoring: In the high-risk path, when the detected voice identifier is a preset voice identifier and the voiceprint similarity is less than the preset first threshold, silent ignoring will be triggered; in the low-risk path, when the language sensitive words are identified to appear ≥2 times and the semantic negative tendency score is ≥0.7, silent ignoring will also be triggered, and finally no output action will be performed, thereby realizing the automatic interception and silent rejection of high-risk or low-risk illegal content.

[0104] Example 3

[0105] Figure 7 This is a schematic diagram of the structure of a multimodal fusion dialogue management device provided in Embodiment 3 of the present invention. Figure 7 As shown, the device includes:

[0106] A multimodal data acquisition module 710 is used to acquire multimodal data in a preset target scene; wherein the multimodal data consists of audio data, video data, and touch data;

[0107] The target data acquisition module 720 is used to identify the multimodal data to obtain target data; wherein, the target data is data used for the dialogue management process;

[0108] The answer generation module 730 is used to generate an answer based on the target data if the target data meets the preset dialogue conditions.

[0109] Optionally, the device further includes:

[0110] The module for generating a refusal-to-answer prompt is used to generate a refusal-to-answer prompt if the target data does not meet preset dialogue conditions.

[0111] Optional, a rejection message generation module, specifically used for:

[0112] If the voice identifier in the target data is a preset voice identifier, a rejection message will be generated.

[0113] Optionally, the module for generating a rejection message may also be used for:

[0114] If the voiceprint similarity in the target data is less than a preset first threshold, a refusal to answer prompt is generated.

[0115] Optionally, the module for generating a rejection message may also be used for:

[0116] If the language information in the target data is a preset language and the confidence level in the target data is less than a preset second threshold, then a rejection message is generated.

[0117] Optional, the answer generation module 730 is specifically used for:

[0118] The text information in the target data is encoded to obtain a first encoding; and the visual information in the target data is encoded to obtain a second encoding.

[0119] The first code and the second code are fused to generate the intent distribution probability;

[0120] The intent distribution probability is input into a pre-configured dialogue state tracking model, and the corresponding intent is output through the dialogue state tracking model.

[0121] Generate an answer that matches the stated intent.

[0122] Optional, the multimodal data acquisition module 710 is specifically used for:

[0123] Multimodal data is collected in a preset target scenario based on a pre-installed smart terminal; wherein the smart terminal consists of a microphone array and a camera.

[0124] The multimodal fusion dialogue management device provided in this embodiment of the invention can execute a multimodal fusion dialogue management method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0125] Example 4

[0126] Figure 8A schematic diagram of an electronic device 10, which can be used to implement embodiments of the present invention, is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0127] like Figure 8 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0128] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0129] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as a multimodal fusion dialogue management method.

[0130] In some embodiments, a multimodal fusion dialogue management method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the multimodal fusion dialogue management method described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to perform a multimodal fusion dialogue management method by any other suitable means (e.g., by means of firmware).

[0131] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transferring data and instructions to the storage system, the at least one input device, and the at least one output device.

[0132] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0133] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0134] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0135] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0136] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0137] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication unit 19, or installed from storage unit 18, or installed from ROM 12. When the computer program is executed by processor 11, it performs the functions defined in the methods of the embodiments of the present invention.

[0138] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0139] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A multimodal fusion dialogue management method, characterized in that, include: Multimodal data is collected in a preset target scenario; wherein the multimodal data consists of audio data, video data, and touch data; The multimodal data is identified to obtain target data; wherein, the target data is data used for the dialogue management process; If the target data meets the preset dialogue conditions, then an answer is generated based on the target data.

2. The method according to claim 1, characterized in that, After identifying the multimodal data to obtain the target data, the method further includes: If the target data does not meet the preset dialogue conditions, a refusal to answer prompt will be generated.

3. The method according to claim 2, characterized in that, If the target data does not meet the preset dialogue conditions, a rejection message is generated, including: If the voice identifier in the target data is a preset voice identifier, a rejection message will be generated.

4. The method according to claim 2, characterized in that, If the target data does not meet the preset dialogue conditions, a refusal to answer prompt is generated, which also includes: If the voiceprint similarity in the target data is less than a preset first threshold, a refusal to answer prompt is generated.

5. The method according to claim 2, characterized in that, If the target data does not meet the preset dialogue conditions, a refusal to answer prompt is generated, which also includes: If the language information in the target data is a preset language and the confidence level in the target data is less than a preset second threshold, then a rejection message is generated.

6. The method according to claim 1, characterized in that, Generating an answer based on the target data includes: The text information in the target data is encoded to obtain a first encoding; and the visual information in the target data is encoded to obtain a second encoding. The first code and the second code are fused to generate the intent distribution probability; The intent distribution probability is input into a pre-configured dialogue state tracking model, and the corresponding intent is output through the dialogue state tracking model. Generate an answer that matches the stated intent.

7. The method according to claim 1, characterized in that, Collect multimodal data in a preset target scenario, including: Multimodal data is collected in a preset target scenario based on a pre-installed smart terminal; wherein the smart terminal consists of a microphone array and a camera.

8. A multimodal fusion dialogue management device, characterized in that, include: A multimodal data acquisition module is used to acquire multimodal data in a preset target scene; wherein the multimodal data consists of audio data, video data, and touch data; The target data acquisition module is used to identify the multimodal data to obtain target data; wherein, the target data is data used for the dialogue management process; The answer generation module is used to generate an answer based on the target data if the target data meets preset dialogue conditions.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform a multimodal fusion dialogue management method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute and implement the multimodal fusion dialogue management method according to any one of claims 1-7.

Images