Voice-interactive adaptive screen generation method and system

Abstract

This invention discloses a method and system for generating adaptive screens through voice interaction, belonging to the field of game voice interaction technology. It aims to solve problems such as low matching degree between voice interaction screens and player game intentions in game scenarios, incomplete data leading to screen deviations, and lack of dynamic optimization adapted to the game process. The method includes: real-time acquisition of player voice input signals in the game, conversion into voice input text fragments after multimodal audio preprocessing, and construction of a structured voice input sequence according to time sequence; parsing newly added structured units, extracting core game elements, and mapping and constructing a game screen requirement list; quantifying the richness of information in the list, determining whether to construct a dynamic semantic chain, calculating the time, frequency factor, and decision weight of sub-intention nodes, and pruning and optimizing; locating target game visual elements based on a game visual element mapping library, combining player preferences and device characteristics for correction, and constructing a screen interaction logic network to generate adaptive screens.

Description

Technical Field

[0001] This invention relates to the field of game voice interaction technology, and more specifically to a voice-interactive adaptive screen generation method and system. Background Technology

[0002] With the rapid development of smart terminals and voice interaction technology, voice has become one of the core methods of human-computer interaction. The need for generating adaptive screens based on voice commands is becoming increasingly urgent and is widely used in game interaction scenarios such as mobile games, console games, and PC games. As an important feedback carrier for game voice interaction, the quality of screen generation directly affects the player's interactive experience. It is necessary to accurately match the game screen function requirements, presentation logic, and game scene characteristics implied in the player's voice.

[0003] However, existing voice-interaction screen generation technologies still face numerous challenges in game applications: They are not comprehensive enough in extracting game-related information from voice commands, often focusing on a single core intent while neglecting key details such as game entity attributes and logical connections. Furthermore, they lack effective mechanisms for identifying and supplementing missing information, leading to incomplete screens with missing game functions or logical inconsistencies. Simultaneously, existing technologies often fail to prioritize game requirements, incorporating all game-related information into the screen presentation. This not only increases the rendering burden on the terminal but also reduces the efficiency for players in accessing core game functions. Moreover, they often ignore player preferences and the differences in hardware characteristics across various gaming devices, resulting in screens that are difficult to adapt to player habits and the display and operational capabilities of gaming devices, leading to poor visual and tactile experiences. In addition, the lack of quantitative assessment of the completeness and effectiveness of game-related information means that screen generation is still triggered even when core game information is missing, resulting in low matching between the screen and the player's actual operational needs, frequent deviations in game interaction feedback, and severely impacting the overall smoothness of game interaction. These problems make it difficult for existing technologies to meet players' demands for accuracy, adaptability, and efficiency in game voice-interaction screen generation. Therefore, in order to overcome these limitations, this invention proposes a voice-interactive adaptive screen generation method and system. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to provide a voice-interactive adaptive screen generation method and system. This method solves the problem of accurately capturing player screen generation needs and supplementing missing information in game voice interaction scenarios, while ensuring that the generated screen aligns with the player's core game intent, is free of redundancy, and conforms to the game scene and device characteristics, thereby improving the matching degree between screen and needs and the smoothness of game interaction. To achieve the above objectives, the present invention provides the following technical solution:

[0005] A voice-interactive adaptive screen generation method includes:

[0006] The system collects players' game voice input signals in real time, converts them into voice input text fragments, and constructs a structured voice input sequence according to the time sequence of the players' input.

[0007] The player's needs and intentions are analyzed for each newly added structured unit in the voice input sequence, the core game elements required for screen generation in a single round of voice interaction are extracted, and a list of game screen requirements is constructed.

[0008] Identify available cells in each column of the game screen requirement list, calculate the availability index of each column of the game screen requirement list, quantify the information richness score of the game screen requirement list, and use it to determine whether to trigger the screen generation operation. If not, start the supplementary query interaction, generate the missing column voice interaction signal to obtain the missing column information.

[0009] If a screen generation operation is triggered, a dynamic semantic chain containing the root node of the idea graph, sub-intention nodes, and entity nodes is constructed based on the current game screen requirement list. The target unit value of each node in the dynamic semantic chain is generated according to the available cells in the game screen requirement list. The time factor and frequency factor of each sub-intention node are calculated to generate the decision weight of the sub-intention node, which is used to prune and optimize the dynamic semantic chain.

[0010] Based on a pre-defined game visual element mapping library, the target game visual elements of the dynamic semantic chain are located, corrected, and a screen interaction logic network is constructed to generate adaptive screens.

[0011] Specifically, the steps for constructing a structured speech input sequence include:

[0012] The game voice input signal is divided into a continuous sequence of voice signal frames according to the preset voice frame length.

[0013] Text conversion is performed on each speech signal frame in the speech signal frame sequence to generate text segments of the speech signal frames, and adjacent frame text segments are concatenated into a speech input text segment;

[0014] The system performs word matching on each sentence of the voice input text segment, identifies and removes redundant segments, and performs contextual semantic consistency verification on the voice input text segment, identifies suspected erroneous words in non-redundant segments, and generates corrective candidate words for each suspected erroneous word.

[0015] Based on the semantic vector of the context of suspected erroneous words, the semantic similarity between the candidate words for correction of each suspected erroneous word and the corresponding speech input text segment is calculated, so as to select the target candidate words for correction to replace the suspected erroneous words.

[0016] Extract the start and end times of the game voice input signal from which each voice input text segment originates, and construct the input time interval of the voice input text segment;

[0017] Configure the segment identifier for each voice input text segment, and construct a structured unit for each voice input text segment by combining the start and end acquisition times of its input time interval;

[0018] Based on the start acquisition time of all structured units, all structured units are sorted in ascending order to generate a set of structured units; then, the structured units are integrated in ascending order according to their start acquisition time to generate a structured speech input sequence.

[0019] Specifically, the core elements of the game include an intent relationship table, an entity attribute association table, an entity logic relationship table, and a scene-based visual auxiliary information table;

[0020] The steps for extracting the core game elements required for screen generation in a single-turn voice interaction include:

[0021] When a new structured unit is detected, the speech input text segment within the new structured unit is extracted; a single-round text purification preprocessing is performed on the speech input text segment to generate the purified single-round requirement text corresponding to the new structured unit;

[0022] Keywords are extracted from the purified single-round demand text and compared with the preset intent keyword mapping library. The matching degree between the extracted keywords and the keywords corresponding to each idea map category in the intent keyword mapping library is calculated to determine the idea map.

[0023] The intent keyword mapping library consists of keywords extracted after preprocessing the high-frequency interactive corpus of the target scene, the mapping relationship between keywords and intent graph categories, and keyword weight values;

[0024] The intent association rule library is invoked to extract the sub-intent categories associated with the identified main intent. Text fragments matching the keywords corresponding to each sub-intent category are extracted from the purified single-round requirement text. Combining the semantic integrity and temporal context of the text fragments, single-round sub-intents are filtered out, and a single-round intent relationship table is constructed. The intent association rule library is used to clarify the sub-intent categories associated with each main intent graph, as well as the keywords corresponding to each sub-intent category.

[0025] Specifically, the steps for extracting the core game elements required for screen generation in a single-round voice interaction also include:

[0026] The contextualized entity dictionary adapted to the target scenario is invoked to match the nouns in the single-round requirement text with the typical words in the contextualized entity dictionary and map them to the corresponding screen entities; the semantic association between screen entities and attributes is mined to determine the attribute names and attribute values ​​corresponding to the screen entities and establish an entity attribute association table.

[0027] The contextual entity dictionary includes entity types and corresponding typical words related to image generation within the target scene;

[0028] The target scenario-based predicate library is invoked to extract key predicates and logical connectors from the purified single-round requirement text; combined with the entity attribute association table, the semantic association between predicates, logical connectors and screen entities is analyzed to determine the logical relationship of screen entities; and an entity logical relationship table is established.

[0029] The target-scenario-based predicate library includes predicates and logical connectors related to screen interaction in the target scenario.

[0030] Extract player preference parameters and device characteristic parameters for a single round of interaction, and construct a single-round auxiliary information table.

[0031] Specifically, the steps for constructing a list of game graphics requirements include:

[0032] A fixed column structure for the game screen requirements list is pre-constructed, including columns for basic identifiers, screen intents, screen entities, screen logic, and screen presentation. A blank new row is created at the end of the game screen requirements list, and the cell values ​​of all columns are uniformly initialized to a state to be filled.

[0033] Extract the structured unit identifiers from each core game element and perform consistency checks. If the structured unit identifiers are consistent, fill the corresponding cell in the screen intent category with the structured unit identifier and the corresponding start collection time of the structured unit; otherwise, trigger the exception logging mechanism.

[0034] After the structured unit identifier consistency check passes, the intent relationship table, entity attribute association table, entity logical relationship table, and single-round auxiliary information table are mapped to the corresponding cells of the screen intent class column, screen entity class column, screen logic class column, and screen presentation class column in sequence.

[0035] The system identifies multi-value cells containing multi-value data in the intent, entity, and logic columns of the screen, concatenates them uniformly according to preset delimiters, and performs format standardization processing on the filled cells of each column.

[0036] Specifically, the steps to determine whether to trigger the screen generation operation include:

[0037] For the basic identifier category, screen intent category, screen entity category, screen logic category, and screen presentation category in the game screen requirements list, establish a usability index for each category; the usability index value is the ratio of the number of usable cells in the corresponding category to the total number of columns in that category;

[0038] Availability checks are performed on cells in each column of the game screen requirements list in turn to identify and mark available cells in each column of the game screen requirements list;

[0039] The system counts the number of available cells in each column and calculates the availability index value for each column. Then, it integrates the availability index values ​​of each column using a weighted summation algorithm to generate an information richness score for the game screen requirement list.

[0040] A preset information richness threshold is set. If the information richness score is greater than the information richness threshold, the screen generation operation is triggered; otherwise, the columns in the unavailable state of each category are located as missing columns, and the missing columns are sorted according to the query priority of each category.

[0041] The system calls a contextualized query template library, sorts queries based on their priority for missing columns, generates voice interaction signals to obtain information about missing columns, and continues until the information richness score is greater than the information richness threshold.

[0042] Specifically, the steps for constructing a dynamic semantic chain that includes the root node of the idea graph, child idea nodes, and entity nodes include:

[0043] Locate the main intent column in the screen intent category of the game screen requirements list, identify the available cells in the idea image column, classify them according to the semantic consistency and data type of the available cell values, and generate a set of values ​​of the same type for idea images:

[0044] If the set of values ​​of the same kind in the main graph is unique, then use that set of values ​​of the same kind as the target set of values ​​of the same kind.

[0045] If the set of values ​​of the same type in the idea graph is not unique, the starting collection time of the structured unit where the main idea of ​​the set of values ​​of the same type is located is extracted. Based on the current interaction time, the time priority of each set is generated by normalizing the value mapping. The frequency of the main idea of ​​each set of values ​​of the same type is counted. The target set of values ​​of the same type is filtered by using the time priority as the main filtering condition and the frequency of the idea graph as the auxiliary filtering condition.

[0046] The values ​​of the same set of target values ​​are used as the target unit values ​​of the main graph node to construct the main intent root node of the dynamic semantic chain;

[0047] Locate the sub-intent column of the screen intent category in the game screen requirement list, identify the available cells in the sub-intent column, and generate a set of similar values ​​for the available cells in the sub-intent column to construct sub-intent nodes. Then, establish a connection between all sub-intent nodes and the root node of the main graph through intent hierarchy edges.

[0048] Construct a set of available cells for entities in each sub-intention, and classify them according to the semantic uniqueness of entity names. Generate a set of values ​​of the same type for entities to construct entity nodes for each sub-intention node and generate target cell values ​​for entity nodes. Establish associations between all entity nodes and corresponding sub-intention nodes through entity intention hierarchy edges.

[0049] Locate all available cells in the screen logic category column of the game screen requirements list, extract the target cell value of each available cell, match the associated entity pairs with the constructed entity nodes, construct entity logic edges, and connect the corresponding entity nodes.

[0050] Specifically, the steps for pruning and optimizing dynamic semantic chains include:

[0051] Extract the start collection time of the structured unit containing the set of values ​​of the same type for each sub-intention; use the current interaction time at the trigger moment of the dynamic semantic chain construction as the benchmark to calculate the time difference between each start collection time and the current interaction time; use a linear normalization algorithm to map the time difference to a preset numerical range to generate the time factor of the sub-intention node.

[0052] The frequency of the set of values ​​of the same type for each sub-intention is counted as the original frequency of the sub-intention. The original frequency is processed by the min-max normalization algorithm to generate the frequency factor of the sub-intention. The decision weight of each sub-intention node is generated by performing a fusion calculation through a weighted summation algorithm.

[0053] Configure the truncation threshold for decision weights, traverse all sub-intention nodes, truncate and remove non-core sub-intention nodes whose decision weights are lower than the truncation threshold, simultaneously delete edge connections associated with non-core sub-intention nodes, and identify and remove isolated nodes.

[0054] Specifically, the steps for generating adaptive images include:

[0055] The system calls a pre-defined game visual element mapping library and uses a semantic feature extraction algorithm to generate structured semantic feature vectors for the main intent root node, core sub-intent node, associated entity node, and entity logical edge of the dynamic semantic chain.

[0056] The semantic feature vectors of each node and logical edge are compared with the feature vectors of game visual elements in the mapping library using a vector cosine similarity matching algorithm. Candidate game visual elements with a cosine similarity greater than a preset similarity threshold are selected. The candidate game visual elements are then sorted to locate the target game visual element.

[0057] Extract player preference parameters and device characteristic parameters from the visual presentation category in the game visual requirements list, and modify the visual elements of the target game accordingly;

[0058] The visual elements of the target game are integrated according to the hierarchical structure of the main idea framework, sub-ideas functional areas, entity interaction controls and logical linkage rules to construct a network of screen interaction logic.

[0059] The visual logic verification mechanism is activated to verify the screen interaction logic network. The screen interaction logic network that passes the verification is converted into a standard screen data format that the device can directly render, and an adaptive screen is generated.

[0060] The voice-interactive adaptive image generation system includes a sequence construction module, an element extraction module, a requirement determination module, a semantic chain construction module, and an image generation module;

[0061] The sequence construction module collects the player's game voice input signals and constructs a structured voice input sequence; the element extraction module analyzes the player's needs and intentions from the voice input sequence, extracts the core game elements required for screen generation, and constructs a game screen requirement list; the requirement judgment module quantifies the information richness score of the game screen requirement list and determines whether to trigger the screen generation operation; the semantic chain construction module constructs a dynamic semantic chain and performs pruning and optimization on the dynamic semantic chain when the screen generation operation is triggered; the screen generation module locates the target game visual elements of the dynamic semantic chain based on a preset game visual element mapping library and generates adaptive screens.

[0062] The beneficial effects of this invention are:

[0063] This application effectively addresses issues such as requirement capture bias, information gaps, screen redundancy, and insufficient adaptation in game voice interaction scenarios, bringing significant application value: By constructing a structured voice input sequence and accurately parsing player intent, it effectively avoids deviations caused by voice signal interference and incomplete requirement parsing, ensuring a high degree of alignment between the core direction of the game screen and the player's actual needs; through information richness quantification assessment and missing information supplementation mechanisms, it proactively identifies and completes key requirement information, avoiding the generation of game screens with missing functions or chaotic logic due to insufficient information, ensuring the integrity and logical coherence of screen functions; through dynamic semantic chain pruning optimization, it can reasonably distinguish requirement priorities and eliminate non-core content, reducing the terminal rendering burden and helping players quickly focus on core game functions, improving interaction efficiency; by modifying game visual elements in combination with player preferences and device characteristics, it allows the generated screen to better adapt to player usage habits and game device hardware capabilities, optimizing the game visual experience and operation feel; at the same time, through requirement information quantification assessment and screen logic verification, it significantly improves the matching degree between the screen and the player's actual needs, reduces interaction feedback bias, significantly enhances the overall smoothness of game voice interaction, and comprehensively improves player satisfaction. Attached Figure Description

[0064] Figure 1 This is a flowchart of the voice-interactive adaptive screen generation method of the present invention;

[0065] Figure 2 A flowchart for constructing a game screen requirements list for this invention;

[0066] Figure 3 This is a flowchart illustrating the process of determining whether a screen generation operation is triggered according to the present invention.

[0067] Figure 4 This is a flowchart illustrating the pruning and optimization of the dynamic semantic chain in this invention. Detailed Implementation

[0068] Please see Figure 1 This embodiment introduces a voice-interactive adaptive screen generation method, including:

[0069] Step S1: Collect the player's game voice input signal in real time, and after multimodal audio preprocessing, convert the player's game voice input signal into voice input text segments. Construct a structured voice input sequence according to the time sequence of the player's input. Each voice input text segment in the voice input sequence is associated with a corresponding voice input timestamp. Multimodal audio preprocessing refers to a preprocessing technique that integrates multiple signal optimization algorithms and detection mechanisms for game voice input signals in voice interaction scenarios. This aims to improve signal purity, eliminate interference distortion, and retain effective voice components. Its core purpose is to provide high-quality, low-noise voice signal input for subsequent speech-to-text processing. Specifically, it includes: 1) Adaptive noise reduction processing, which dynamically adjusts filtering parameters by analyzing the spectral characteristics and energy changes of environmental noise and background sound effects specific to the game scene in real time. This accurately filters out steady-state noise, non-steady-state noise, and game background interference, preventing noise from masking the effective voice signal. 2) Echo cancellation processing, which extracts the feature templates of the device's own output audio and compares them with the player's game voice input signal to eliminate voice echo superposition caused by the coupling between the device's speaker and microphone, thus avoiding signal distortion. 3) Intelligent endpoint detection processing, which analyzes the energy threshold, spectral distribution, and zero-crossing rate characteristics of the voice signal to accurately identify the start and end positions of effective voice, eliminating invalid silence segments and redundant gaps in the game voice input signal, ensuring that only effective voice components containing the player's interactive intent are retained.

[0070] In this embodiment, step S1 effectively filters out steady-state and non-steady-state noise, game background interference, and voice echo distortion caused by device coupling by collecting the player's game voice input signal in real time and combining it with multimodal audio preprocessing such as adaptive noise reduction, echo cancellation, and intelligent endpoint detection. It also accurately removes invalid silent segments, significantly improving the purity and effective component ratio of the voice signal. At the same time, the optimized voice signal is converted into text segments, and a structured voice input sequence with associated timestamps is constructed according to the input time sequence. This ensures the temporal consistency between the voice text and the original interaction, and provides high-quality, structured basic data support for subsequent core processes such as game player demand analysis and game dynamic semantic chain construction, reducing the error risk of subsequent processing from the source.

[0071] Preferably, the specific steps for constructing a structured speech input sequence include:

[0072] The game voice input signal, after multimodal audio preprocessing, is retrieved and divided into a continuous voice signal frame sequence according to a preset voice frame length. The preset voice frame length is set based on the voice signal sampling rate to ensure the integrity of the voice features within the frame.

[0073] For each speech signal frame in the speech signal frame sequence, text conversion is performed using a collaborative speech-to-text architecture between the device and the cloud to generate text fragments of the speech signal frames. A semantic coherence detection algorithm then concatenates adjacent text fragments into a complete speech input text fragment, ensuring semantic consistency between the text fragment and the original speech signal. Specifically, the device-side speech-to-text architecture uses a lightweight deep neural network speech recognition model to perform preliminary translation on short speech signal frames (those shorter than a preset duration threshold) to generate candidate text results. The cloud-side architecture uses a large-scale pre-trained speech recognition model to extract Mel-frequency cepstral coefficients and spectrogram features from the speech signal frames. These features are then combined with the candidate text results from the device to perform language model decoding, generating candidate text results from the cloud. The translation confidence of the device-side and cloud-side candidate text results is compared, and the text result with higher confidence is selected as the text fragment for the corresponding speech signal frame. The semantic coherence detection algorithm performs semantic correlation analysis and logical consistency checks on the text fragments generated from adjacent speech signal frames. The specific execution flow of the verification algorithm is as follows: First, adjacent text segments are segmented and semantically encoded to generate contextual semantic vectors for each segment. The semantic correlation between vectors is calculated using the cosine similarity algorithm. Then, the consistency of adjacent segments in terms of grammatical collocation, semantic orientation, and logical relationship is verified. If the semantic correlation is lower than the preset correlation threshold or there is a logical contradiction, the original acoustic features of the corresponding adjacent speech signal frame are retrieved back, and speech-to-text processing is re-executed to correct the text segment until the adjacent segments meet the requirements of semantic correlation and logical consistency. This avoids the problem of semantic breaks and logical contradictions after the adjacent frame text segments are spliced ​​together, and provides semantically fluent basic text data for subsequent text validity verification and structured unit construction.

[0074] The system invokes a domain-specific dictionary adapted to the voice interaction scenario to perform word matching on each sentence of the generated voice input text fragment. Sentences containing only invalid modal particles or lacking actual interactive meaning are marked as redundant and removed. Contextual semantic consistency is then checked on the voice input text fragments to identify suspected errors in non-redundant fragments, including misspelled words and words with incorrect context matching. Correction candidate words are generated for each suspected error. For example, the domain-specific dictionary adapted to the voice interaction scenario is invoked to extract a scenario-specific error-prone word lookup table. This lookup table is constructed based on error statistics from a massive amount of voice interaction corpus in the target scenario, recording frequently occurring misspelled words in speech-to-text conversion, such as empty words, etc. The system first sets up a contextual mapping between incorrect words and their corresponding correct words, directly matching suspected incorrect words and extracting the corresponding correct words from the table as initial correction candidates to ensure coverage of known typical incorrect words in the scene. Then, it combines the voice input text fragment containing suspected incorrect words with adjacent text fragments to form a complete contextual information. This information is then input into a pre-trained language model, such as the RoBERTa model, fine-tuned with interactive corpus of the target scene. The model's masking language model function is enabled, and the positions of suspected incorrect words are set as masks. Based on the probability distribution of word predictions at the masked positions, the top 5 words with the highest probability are selected as extended correction candidates to cover atypical incorrect words not included in the scene-specific error-prone word comparison table.

[0075] Based on the context semantic vector of suspected erroneous words, the cosine similarity algorithm is used to calculate the overall semantic similarity between the candidate words for correction of each suspected erroneous word and the speech input text segment in which they are located. The candidate words for correction with the highest semantic similarity are selected as the target candidate words for correction and the replacement operation is performed.

[0076] The start and end times of the game voice input signal from which each voice input text segment originates are extracted from the game voice input signal acquisition stage. The extracted start and end times are combined into the input time interval of the voice input text segment, and a one-to-one mapping relationship is established with the corresponding voice input text segment.

[0077] Configure the segment identifier for each voice input text segment, and construct a structured unit for each voice input text segment by combining the start and end acquisition times of its input time interval;

[0078] Based on the start acquisition time of all structured units, all structured units are sorted in ascending order to generate a set of structured units with consistent timing. The set of structured units is then integrated in ascending order of their start acquisition time to generate a structured speech input sequence containing sequence header information. The sequence header information includes: total speech acquisition duration, total number of structured units, and sequence generation time.

[0079] Step S2: Monitor the update status of the structured speech input sequence in real time. When a new structured unit is detected, analyze the player's game intent for the set of new structured units in the structured speech input sequence. Using a multi-dimensional game core element extraction technology that combines intent layering recognition, entity attribute association extraction, logical relationship mining, and scenario-based detail supplementation, extract the core game elements required for game screen generation and construct a game screen requirement list. The core game elements refer to the key information dimensions that can fully support adaptive game screen generation and cover screen functions and presentation logic, including intent relationship tables, entity attribute association tables, entity logical relationship tables, scenario-based screen auxiliary information tables, and screen auxiliary information adapted to player game operation preferences and scene characteristics.

[0080] In this embodiment, step S2 uses a multi-dimensional game core element extraction technology, including intent layer recognition, entity attribute association extraction, logical relationship mining, and scenario-based detail supplementation, to accurately analyze the player's core game needs and intents in the structured voice input sequence. It comprehensively extracts the game intents, game entity attributes, game logical relationships, and scenario-based auxiliary information required for game screen generation, ensuring that the core game elements cover the key dimensions of screen function implementation and presentation logic. This provides accurate and structured support for subsequent adaptive game screen generation, effectively improving the matching degree between game screen and player needs and the smoothness of game interaction experience.

[0081] Preferably, the specific steps for extracting the core game elements required for image generation include:

[0082] When a new structured unit is detected, the voice input text segment within that new structured unit is extracted. A single-round text purification preprocessing is then performed on the voice input text segment: a Chinese word segmentation algorithm is used to segment the text segment sentence by sentence, removing punctuation marks and special characters from the text; a customized stop word list adapted to the target voice interaction scenario is called to filter out auxiliary words in the text that do not contribute semantically, generating the purified single-round requirement text corresponding to the new structured unit, laying the foundation for the extraction of core elements of the single-round game.

[0083] Keywords are extracted from the purified single-round requirement text and compared with a pre-defined intent keyword mapping library. A weighted matching algorithm based on keyword weights is used to calculate the matching degree between the extracted keywords and the keywords corresponding to each idea map category in the intent keyword mapping library. During the matching degree calculation, the keywords in the single-round requirement text are compared with the keywords of the corresponding idea map category in the intent keyword mapping library. If a matching keyword exists, the weight value corresponding to that keyword is accumulated. Finally, the accumulated weight value corresponding to each idea map category is used as the matching degree result. The idea map category with the highest matching degree is selected as the idea map corresponding to the current single-round requirement text. The intent keyword mapping library refers to a structured data set used to establish the correspondence between idea map categories and keywords in the target scenario. It consists of keywords extracted after preprocessing the high-frequency interaction corpus of the target scenario, the mapping relationship between keywords and idea map categories, and keyword weight values. The preprocessing of the high-frequency interaction corpus of the scenario includes word segmentation, purification, and invalid word filtering operations. Keywords are obtained through word frequency statistics and semantic distinguishability analysis. The keyword weight value is determined based on the frequency of occurrence and semantic contribution of the word in the corresponding idea map category corpus.

[0084] The intent association rule library is invoked to extract the sub-intent categories associated with the identified main intent. Text fragments matching the keywords corresponding to each sub-intent category are extracted from the purified single-round requirement text. Combining the semantic integrity and temporal context of the text fragments, single-round sub-intents are filtered out. A single-round intent relationship table is constructed according to the format of newly added structured unit identifier, main image name, and sub-intent name. The boundaries of the subdivided functional modules for screen generation are clearly defined. Specifically, the intent association rule library is constructed based on the functional module division logic corresponding to the main image in the target scenario, clarifying the sub-intent categories associated with each main image and the keywords corresponding to each sub-intent category. For example, in the main image air conditioning control interface display, the associated sub-intents are temperature adjustment area display, mode selection area display, and fan speed adjustment area display, with each sub-intent corresponding to specific keywords such as temperature, mode, and fan speed.

[0085] The system invokes a contextualized entity dictionary adapted to the target scenario. This dictionary includes entity types and corresponding typical words related to image generation within the target scenario. An entity linking algorithm is used to match nouns in the single-round requirement text with typical words in the contextualized entity dictionary, mapping them to corresponding image entities and marking the entity name and type. Dependency parsing is performed on the marked image entities to mine the semantic relationships between image entities and attributes, determining the attribute names and attribute values ​​corresponding to the image entities. An entity attribute association table is established according to the format of newly added structured unit identifier, entity name, entity type, attribute name, and attribute value.

[0086] The system invokes a target-scenario-based predicate library, which includes predicates and logical connectors related to screen interaction in the target scenario. A predicate recognition algorithm is used to extract key predicates and logical connectors from the purified single-round requirement text. Combined with an entity attribute association table, the semantic relationships between predicates, logical connectors, and screen entities are analyzed to determine the logical relationships between screen entities. These logical relationships include temporal, causal, and conditional relationships. For example, if the text contains temporal connectors or time attributes, the relationship between entities is determined to be temporal; if it contains causal connectors, it is determined to be causal; if it contains conditional connectors, it is determined to be conditional. An entity logical relationship table is then established according to the format of newly added structured unit identifiers, associated entity pairs, and logical relationship types.

[0087] The system retrieves player preference parameters from the player's historical interaction database, such as the most recent color style selection; it reads the real-time characteristic parameters of the current interactive device through the device hardware information interface to ensure that the parameters are consistent with the current device status; and it constructs a single-round auxiliary information table based on the scenario adaptation requirements of the single-round requirement text, according to the format of newly added structured unit identifiers, player preference parameters, and device characteristic parameters.

[0088] Please see Figure 2 Preferably, the specific steps for constructing the game graphics requirements list include:

[0089] Based on the comprehensive technical requirements for screen generation, a fixed column structure is pre-constructed for the game screen requirement list, including basic identifier columns, screen intent columns, screen entity columns, screen logic columns, and screen presentation columns. The basic identifier column includes newly added structured unit identifiers and requirement generation time, used to uniquely associate requirement information with a single voice interaction structured unit, clarify the temporal position of the requirement in the overall interaction process, and record the time benchmark for requirement generation, providing technical support for requirement traceability management, temporal retrieval, and anomaly backtracking. The screen intent column includes a main idea, main idea function description, sub-intentions, and sub-intention function descriptions, used to locate the core functional direction of screen generation, define the technical functional boundaries corresponding to the main idea, break down the sub-functional categories under the main idea, clarify the specific technical implementation form corresponding to the sub-intentions, and set the display weight of sub-functional modules, providing a technical basis for the design and priority ranking of screen function modules. The screen entity column includes entity code, entity name, entity type, entity attribute item, and entity attribute value, used to assign a unique technical identifier to the entity to be presented on the screen. The system identifies and clarifies the specific targets of entities, classifies the technical attribute categories of entities, defines the displayable technical dimensions of entities, provides specific values ​​for entity attributes, defines the spatial layout of entities in the screen, and defines the technical operation forms for player interaction with entities, supporting the visualization presentation and interactive function realization of entities; the screen logic category includes associated entity pairs, logical relationship types, and logical execution order, used to clarify logically related entity combinations, classify the technical categories of logical relationships between entities, set the technical prerequisites for the logical relationships to take effect, sort the execution sequence of multiple sets of logical relationships, and define the screen visualization feedback scheme after the logic is executed, providing technical specifications for the design and execution of dynamic interactive logic of the screen; the screen presentation category includes device adaptation parameters and player preference parameters, used to adapt to the hardware characteristics of the current interactive device, define the overall visual technical configuration of the screen, integrate the personalized configuration formed by the player's historical interactions, determine the spatial layout scheme of sub-intent modules, define the superimposed display order of screen elements, and clarify the technical update type of the screen for the current needs, providing technical standards for the visual design, device adaptation, and dynamic updates of the screen;

[0090] The system calls the fixed column structure definition of the game screen requirement list, creates a blank new row at the end of the game screen requirement list, and initializes the cell values ​​of all columns to a state to be filled. At the same time, it loads the data format constraint library, which stores the format requirements and value range of each column data to ensure that each piece of data filled in later meets the standards that the screen generation module can parse.

[0091] The process iterates through the intent relationship table, entity attribute association table, entity logical relationship table, and single-round auxiliary information table of the core elements of a single game round, extracting the structured unit identifiers used to identify a single interaction from each core element. A consistency check is performed. If the structured unit identifiers are consistent, the structured unit identifier and the corresponding start collection time of the structured unit are filled into the corresponding cell in the intent category column. Otherwise, it indicates an anomaly in the core element extraction process, where the multi-dimensional core elements corresponding to a single interaction have not formed a unified association mapping. In this case, the current row filling process is paused, and an anomaly logging mechanism is triggered to record in detail the inconsistent structured unit identifiers and extraction sources in each table. Simultaneously, the core element re-extraction trigger program is started, re-executing the single-round core element extraction process based on the original voice input text fragment. After ensuring the consistency of structured unit identifiers across multiple tables, subsequent filling operations continue.

[0092] After the structured unit identifier consistency check passes, according to the preset game core element table and class column mapping rules, the main intent, sub-intent, main image function description, and sub-intent function description information in the intent relationship table are directly mapped and filled into the corresponding cells of the screen intent class column; the entity unique code, entity name, entity type, entity core attribute item, and entity attribute value information in the entity attribute association table are mapped and filled into the corresponding cells of the screen entity class column; the associated entity code pairs, logical relationship types, logical triggering technical conditions, and logical execution timing order information in the entity logic relationship table are mapped and filled into the corresponding cells of the screen logic class column; and the device characteristic parameters and player preference parameters in the single-round auxiliary information table are mapped and filled into the corresponding cells of the screen presentation class column. During the mapping process, field name matching technology is used to ensure the accuracy of data filling.

[0093] The system identifies multi-valued cells containing multi-valued data in the screen intent, screen entity, and screen logic columns, such as multiple sub-intents, multiple entities, and multiple sets of related entity pairs. These multi-valued cells are then uniformly concatenated using a preset delimiter to form a structured string. Simultaneously, a data format constraint library is invoked to perform format standardization processing on the filler cells of each column. Non-standard format data, such as data with missing encoding type identifiers or insufficient time precision, are adjusted to the preset standard format using a format conversion tool. This ensures that all data meets the parsing requirements of the screen generation module and avoids parsing anomalies caused by format differences.

[0094] Step S3: Based on the constructed game screen requirement list, combined with the data existence determination logic and data format compliance determination logic, identify available cells, calculate the availability index of each column of the game screen requirement list to quantify the information richness score of the game screen requirement list, and use it to determine whether to trigger the screen generation operation. If not, start the supplementary query interaction, generate voice interaction signals to interact with the player according to the priority of the missing columns in the game screen requirement list, and obtain the missing column information.

[0095] In this embodiment, step S3 quantifies the availability indicators of various columns through dual judgment logic of data existence and format compliance, accurately assesses the information richness of the game screen requirement list, ensures sufficient data support when triggering game screen generation, and efficiently completes missing information through priority supplement queries, avoiding deviations in game screen generation caused by missing data or abnormal formats, effectively improving the matching accuracy between adaptive game screen and player needs and the smoothness of game voice interaction.

[0096] Please see Figure 3 Preferably, the specific steps for determining whether to trigger the screen generation operation include:

[0097] For the basic identifier category, screen intent category, screen entity category, screen logic category, and screen presentation category in the game screen requirements list, establish usability indicators for each category. The usability indicator value is the ratio of the number of usable cells in the category to the total number of columns in the category. Usable cells are defined as cells that are not empty after data existence judgment logic verification and that meet the format requirements of the corresponding column in the data format constraint library after data format compliance judgment logic verification.

[0098] The system employs a column-by-column traversal verification technique to sequentially perform availability checks on cells in each column. First, it checks whether the value of the logical verification unit is not empty by determining the existence of data. Then, it calls the format specifications of the corresponding column in the data format constraint library and checks whether the non-empty cells in the logical verification meet the format requirements by determining the compliance of data format. This is done to identify and mark available cells in each column of the game screen requirements list.

[0099] The number of usable cells in each column is counted, and the usability index value of each column is calculated. Then, the usability index values ​​of each column are integrated through a weighted summation algorithm to generate the information richness score of the game screen requirement list. The weighting coefficient is set based on the importance of each column to the technical support of screen generation.

[0100] A preset information richness threshold is set based on the minimum data support requirements for screen generation. The information richness score is compared with the information richness threshold. If the information richness score is greater than the information richness threshold, it is determined that the data requirements for screen generation are met, and the screen generation operation is triggered; otherwise, columns in the unavailable state among various columns are located as missing columns. The missing columns are sorted according to the query priority of screen entity category column, screen logic category column, screen intent category column, screen presentation category column, and basic identifier category column.

[0101] The system invokes a scenario-based query template library, prioritizes queries based on missing columns, generates voice interaction signals containing the type of missing column and data format requirements, and outputs them to the player through the interactive interface. Player supplementary voice input is collected in real time, and the multimodal audio preprocessing and speech-to-text process is repeated to extract the required supplementary data and fill it into the corresponding missing columns to update the unit values. After completion, the system re-executes the column-by-column traversal verification, usability index calculation, and comprehensive score quantification steps until the information richness score exceeds the information richness threshold.

[0102] Step S4: If a screen generation operation is triggered, a dynamic semantic chain is constructed based on the current game screen requirement list. According to the available cells in the game screen requirement list, target unit values ​​for each node of the dynamic semantic chain are generated, ensuring that the unit values ​​fully cover the core attributes of the node, such as the main idea name, sub-intention function description, entity attribute items, and value pairs. The nodes of the dynamic semantic chain include the main idea root node, sub-intention nodes, and entity nodes. Combining the frequency and sequence of the target unit values, frequency reflects the node's interaction heat, and sequence highlights the priority of recent interactions, ensuring that the weight calculation aligns with the player's actual game needs, each node of the dynamic semantic chain is generated. The decision weights of each point are determined, and a truncation threshold for the decision weights is set based on the current interaction depth. For example, the threshold adapts to complex needs in multi-round interactions and focuses on core demands in single-round interactions, improving the adaptability of the threshold to the interaction scenario. Non-core nodes with decision weights below the threshold are truncated and eliminated. The dynamic semantic chain is pruned and optimized to effectively simplify the semantic chain structure, eliminate redundant information, and avoid invalid nodes interfering with subsequent processing. This provides accurate semantic logic support for subsequent adaptive game screen generation, ensuring that the game screen generation can accurately match the player's game intentions, efficiently call game entity information, and improve the adaptability of the game screen to the requirements and the coherence of the presentation logic.

[0103] In this embodiment, step S4 transforms the available data of the game screen requirement list into target unit values ​​and constructs a dynamic semantic chain of three types of core nodes. Combining the frequency of occurrence of unit values ​​with the decision weight of time-series generation nodes, non-core information is truncated and eliminated through thresholding. This achieves both the structuring and logical clarity of semantic associations and focuses on core requirements, providing accurate semantic logic support for adaptive game screen generation and effectively improving the matching accuracy between game screens and player needs as well as the coherence of presentation logic.

[0104] Preferably, the specific steps for constructing the dynamic semantic chain of the current game screen requirements list include:

[0105] The main intent column of the screen intent category in the game screen requirement list is located as the concept image column, and the available cells in the concept image column are identified. A value clustering algorithm is used to process the available cells in the concept image column, classifying them according to the semantic consistency and data type of the available cell values, generating a set of values ​​of the same type for the concept image.

[0106] If the set of values ​​of the same type in the main graph is unique, then the set of values ​​of the same type is used as the target set of values ​​of the same type, and the values ​​of the target set of values ​​of the same type are used as the target unit values ​​of the main graph nodes to construct the main intent root node of the dynamic semantic chain.

[0107] If the set of values ​​of the same type in the idea graph is not unique, the starting collection time of the structured unit where the main intent of the set of values ​​of the same type is located is extracted. Based on the current interaction time, the time difference between each starting collection time and the current interaction time is calculated and normalized to the interval [0,1]. The time priority of each set is generated by mapping the normalized value. The frequency of the main intent of each set of values ​​of the same type is counted. The target set of values ​​of the same type is selected by using time priority as the main filtering condition and the frequency of the idea graph as the auxiliary filtering condition. The value of the target set of values ​​of the same type is used as the target unit value of the idea graph node to construct the root node of the main intent of the dynamic semantic chain. The value clustering algorithm is used to achieve effective classification of the values ​​of the idea graph and avoid confusion between different semantic values. Combining the filtering rules of time difference normalization and frequency statistics, the value of the idea graph that has been recently interacted and appears frequently is selected first, which improves the accuracy of the root node of the idea graph and the adaptability of the current interaction, and provides accurate core intent guidance for the construction of the entire semantic chain.

[0108] Assign a node identifier to the root node of the concept graph, and include attributes such as the main intent name and concept graph function description of the screen intent class.

[0109] Locate the column containing the sub-intents in the screen intent category of the game screen requirement list, and identify all available cells in that column. Then, use a value clustering algorithm to process the available cells in the sub-intent column to generate a set of values ​​of the same type for all sub-intents.

[0110] For each set of values ​​of the same type of sub-intention, construct a sub-intention node, assign a node identifier, and each node carries attributes such as the sub-intention name and sub-intention function description of the screen intention category; establish a connection between all sub-intention nodes and the root node of the main graph through the intention hierarchy edge.

[0111] The system retrieves available cells for entities in the entity category column under each sub-intention in the game screen requirement list, constructing a set of available entity cells for each sub-intention. It then categorizes entities according to the semantic uniqueness of their names, generating a set of values ​​for each entity and its target cell value to construct entity nodes for each sub-intention node. Each node is assigned a node identifier and carries complete entity attribute information, including entity name, type, attribute items, and value pairs. Finally, it establishes associations between all entity nodes and their corresponding sub-intention nodes through entity intention hierarchy edges, with edge attribute annotation functionality dependent.

[0112] Locate all available cells in the "Screen Logic Category" column of the game screen requirements list, extract the target cell value of each available cell, including related entity pairs with logical relationships and the logical relationship type; match the related entity pairs with the constructed entity nodes to determine the entity node pairs with logical relationships; for the successfully matched entity node pairs, construct entity logic edges to connect the corresponding entity nodes; label the logical relationship type of the edge attributes.

[0113] Please see Figure 4 The process involves extracting the initial acquisition time of the structured units corresponding to the same value set of each sub-intention, using the current interaction time triggered by the dynamic semantic chain as a benchmark, and calculating the time difference between each initial acquisition time and the current interaction time. A linear normalization algorithm is used to map the time difference to a preset numerical range, generating a sub-intention node time factor. The smaller the time difference, the higher the corresponding time factor value, intuitively quantifying the temporal priority of the sub-intention. The process also counts the occurrence frequency of the same value set of each sub-intention in multi-round structured speech input sequences, using this as the original frequency of the sub-intention. A min-max normalization algorithm is used to process the original frequency, eliminating the influence of frequency distribution differences under different interaction scenarios, generating a sub-intention frequency factor. The higher the original frequency, the higher the corresponding frequency factor value. Based on the preset temporal and frequency weight coefficients of the target scenario rule base, and adapting to the different interaction scenarios' emphasis on temporal correlation and interaction heat, the sub-intention node time factor and frequency factor are used as inputs, and a weighted summation algorithm is used to perform fusion calculations to generate the decision weight of each sub-intention node. The weight coefficients can be dynamically adjusted according to the functional requirements of the target interaction scenario to ensure that the weight calculation fits the scenario characteristics. The time factor accurately quantifies the temporal characteristics of sub-intents through linear normalization, ensuring that sub-intents with recent interactions receive higher weights, aligning with players' immediate needs. The frequency factor, after standardization, objectively reflects the interaction popularity of sub-intents, increasing the priority of high-frequency sub-intents. The combination of preset scenario-based weight coefficients and weighted summation algorithms avoids subjective judgment biases while adapting to different scenario requirements, ensuring the objectivity, rationality, and scenario adaptability of weight calculation.

[0114] Based on the current interaction depth and the core function adaptation standards of the target scenario, a truncation threshold for decision weights is configured. All sub-intention nodes are traversed, and non-core sub-intention nodes with decision weights below the truncation threshold are truncated and removed. Simultaneously, all edge connections associated with non-core sub-intention nodes are deleted. The connectivity of all nodes in the dynamic semantic chain is verified. Main graph nodes are associated with sub-intention nodes, sub-intention nodes are associated with entity nodes, and entity nodes with logical relationships are connected through entity logical edges. After identifying and removing isolated nodes, the semantic chain is converted into a standardized graph structure data format, and a unique semantic chain identifier is assigned. Node association description documents are generated simultaneously and output along with the standardized graph structure data.

[0115] Step S5: Based on the pruned and optimized dynamic semantic chain and standardized graph structure data, call the preset game visual element mapping library. This library stores the correspondence between main graph nodes and screen module frameworks, sub-intention nodes and functional display areas, entity nodes and interactive controls, and entity logic edges and visual linkage rules. Using a semantic feature matching algorithm, the main intention root node, core sub-intention nodes, associated entity nodes, and entity logic edges of the dynamic semantic chain are mapped to corresponding game visual elements, including module layout templates, functional control types, visual style parameters, and interactive linkage logic. Combining player preference parameters and device characteristic parameters from the screen presentation category in the game screen requirement list, the mapped game visual elements are then processed. Adaptive Adjustment: Optimize visual styles according to player preferences, and adjust control sizes and layout structures according to device characteristics to ensure that the visual presentation matches player habits and device hardware capabilities; adopt a game visual element integration algorithm to construct a complete screen structure according to the hierarchical order of main map framework, sub-intention areas, entity controls, and logical linkages, and simultaneously pass a visual logic verification mechanism to check the linkage rules corresponding to the layout of game visual elements and entity logic edges; finally, generate an adaptive screen that supports real-time output display, while reserving a semantic chain update interface. When subsequent interactions trigger dynamic semantic chain adjustments, it can quickly respond and update the screen to ensure that the screen always stays in sync with the player's needs, intentions, and interaction logic, providing players with a coherent and accurate game visual interaction experience.

[0116] In this embodiment, step S5 ensures that the generated adaptive game screen accurately matches the player's needs and interaction logic through precise mapping of semantic chains and game visual elements, adaptation and adjustment of player preferences and device characteristics, and visual logic verification, while also conforming to player habits and device hardware capabilities. The reserved semantic chain update interface enables rapid synchronization of the screen with semantic chain adjustments, effectively ensuring the continuity and accuracy of game visual interaction and improving the player's gaming experience.

[0117] Preferably, the specific steps for generating adaptive images include:

[0118] The system calls upon a pre-defined game visual element mapping library and employs a semantic feature extraction algorithm to generate structured semantic feature vectors for the main intent root node, core sub-intent nodes, associated entity nodes, and entity logical edges of the dynamic semantic chain. Using a vector cosine similarity matching algorithm, the semantic feature vectors of each node and logical edge are compared with the feature vectors of game visual elements in the mapping library, filtering candidate game visual elements with a cosine similarity greater than a pre-defined similarity threshold. Furthermore, the candidate game visual elements are sorted based on a pre-defined priority rule for the target scene to locate the target game visual element, including the screen frame template corresponding to the main intent, the functional area layout corresponding to the sub-intent, the interactive control type corresponding to the entity, and the visual linkage rule corresponding to the logical edge.

[0119] Extract player preference parameters and device characteristic parameters from the game's visual presentation category in the game's visual requirements list; adjust the style attributes of the target game's visual elements according to the player preference parameters, including attributes such as color, font, and control rounded corners; adjust the area attributes of the target game's visual elements according to the device characteristic parameters, including the proportion of functional areas, control spacing, and size.

[0120] A greedy layout integration algorithm is adopted to integrate the target game visual elements according to the hierarchy of main idea framework, sub-intention functional area, entity interactive control and logical linkage rules; based on the main idea framework template, the functional area space and display priority are allocated according to the decision weight of sub-intention nodes; entity controls are embedded into the corresponding sub-intention area; the logical relationship of entity logical edges is transformed into control interaction trigger conditions to construct the screen interaction logic network.

[0121] A visual logic verification mechanism is initiated to verify the on-screen interaction logic network. This includes: using a spatial conflict detection algorithm to check whether game visual elements overlap or occlude, ensuring layout regularity; comparing the semantic consistency between visual linkage rules and entity logic edges to ensure interaction logic matching; calling a device adaptation simulation model to simulate rendering effects and verify device compatibility; if problems are found, a correction mechanism is triggered: adjusting the area proportion and control position to resolve layout conflicts, remapping linkage rules to correct inconsistencies in logic, and optimizing control parameters to resolve compatibility anomalies, until verification passes.

[0122] The validated screen interaction logic network is converted into a standard screen data format that the device can directly render, including layout description files, style configuration files, and interaction logic scripts, to generate adaptive screens. Rendering optimization algorithms are used to compress and encode the screen data and perform cache preprocessing to improve real-time output efficiency. The screen data is output to the interactive interface through the device display interface to achieve real-time presentation of adaptive screens. A semantic chain update listening interface is reserved, and a semantic chain change detection mechanism is configured. When a node addition or deletion, attribute change, or logical relationship adjustment is detected in the dynamic semantic chain, an incremental screen update process is triggered. Only the game visual elements corresponding to the changed nodes are re-mapped and adapted, while the cached state of the unchanged parts is maintained to avoid interaction delays caused by full rendering.

[0123] This embodiment introduces a voice-interactive adaptive image generation system, including a sequence construction module, an element extraction module, a demand determination module, a semantic chain construction module, and an image generation module;

[0124] The sequence construction module is used to collect players' game voice input signals in real time, perform multimodal audio preprocessing, and improve the purity and effective component ratio of the voice signal. Through a speech-to-text architecture that is collaborative between the device and the cloud, the preprocessed voice signal is converted into text fragments. The complete text is then assembled using a semantic coherence detection algorithm, and text purification and error correction are performed using a domain-specific dictionary. The input time interval and fragment identifier of each voice input text fragment are extracted to construct structured units. These units are then integrated in ascending order of the start time to generate a structured voice input sequence containing sequence header information, providing high-quality, time-consistent basic data for subsequent requirement analysis.

[0125] The element extraction module monitors the update status of the structured speech input sequence in real time. When a new structured unit is detected, a single-round text purification preprocessing is performed on the speech input text segment within the new structured unit. The module extracts the main idea and sub-intentions through intent hierarchical recognition technology and constructs an intent relationship table. It extracts entity attribute information using entity linking algorithms and dependency parsing and constructs an entity attribute association table. It mines logical relationships between entities based on a target scenario-based predicate library and constructs an entity logical relationship table. It obtains auxiliary information by combining player historical interaction data and device hardware information and constructs a single-round auxiliary information table. According to a preset fixed column structure, the core elements of the game in the multi-table are mapped and filled into the corresponding cells, and after format standardization, a game screen requirement list is generated.

[0126] The requirement determination module, based on data existence and data format compliance logic, identifies available cells in the game screen requirement list and calculates availability indicators and information richness scores for various columns. It compares the overall score with a preset information richness threshold to determine whether to trigger the screen generation operation. If the triggering conditions are not met, it locates the missing column and generates a voice interaction signal containing the missing column type and data format requirements, prioritizing queries based on screen entity category, screen logic category, screen intent category, screen presentation category, and basic identifier category. It collects supplementary voice input from the player in real time, repeats the voice preprocessing and text-to-text process, extracts supplementary data, and updates the game screen requirement list until the information richness score meets the triggering requirements.

[0127] After triggering the screen generation operation, the semantic chain construction module constructs a dynamic semantic chain based on the available cells of the game screen requirement list: It locates available cells in the idea graph column, generates the idea graph root node through value clustering and time priority / frequency filtering; constructs all sub-intention nodes and associates them with the idea graph root node through intention-level edges; constructs entity nodes corresponding to each sub-intention and associates them with the corresponding sub-intention nodes through entity-intention-level edges; extracts screen logic category information and constructs entity logic edges to connect entity nodes with logical relationships; calculates the time factor and frequency factor of each sub-intention node and generates decision weights through a weighted summation algorithm; based on the current interaction depth and the target scene's core function adaptation standard configuration truncation threshold, it removes non-core sub-intention nodes and associated edges, verifies node connectivity and removes isolated nodes, and converts the optimized semantic chain into a standardized graph structure data format for output.

[0128] The screen generation module calls a pre-defined game visual element mapping library. Through semantic feature extraction and vector cosine similarity matching algorithms, it maps the core nodes and entity logic edges of the dynamic semantic chain to corresponding game visual elements. Combining player preference parameters and device characteristic parameters from the game screen requirement list, it adjusts the style and regional attributes of the game visual elements. A greedy layout integration algorithm is used to construct a screen interaction logic network hierarchically. After spatial conflict detection, logical consistency verification, and device compatibility verification, it is corrected and optimized. The verified screen interaction logic network is converted into a standard screen data format that can be rendered by the device. After rendering optimization, it is presented in real time through the device display interface. A semantic chain update listening interface is reserved. When the dynamic semantic chain is adjusted, an incremental update process is triggered, which only remaps the game visual elements corresponding to the changed nodes, ensuring real-time synchronization between the screen and the requirement intent and interaction logic.

[0129] Working principle and its effects:

[0130] This invention aims to accurately convert voice requests into adaptive visuals. Its overall working principle revolves around accurately capturing requests, supplementing key information, focusing on core intentions, and adapting to scene characteristics. It not only solves the problems of intention deviation, information loss, and visual redundancy in traditional voice interaction screen generation, but also provides players with a coherent and adaptive game interaction experience.

[0131] Specifically, in the voice input processing stage, this invention collects the player's game voice signal in real time, segments it according to a preset frame length, performs text conversion, and constructs a structured voice input sequence by combining redundant segment removal and error word correction. This principle effectively reduces the impact of environmental noise and invalid data on subsequent analysis by retaining the core effective content of the voice and filtering interference information, ensuring that the player's actual game screen requirements are initially and accurately captured. In the requirement analysis and list construction stage, based on the purified text, keyword matching is performed to create a subject graph. A contextual entity dictionary and predicate library are called to mine entity attributes and logical relationships. At the same time, missing information is located and supplementary queries are initiated through information richness quantification assessment. This principle comprehensively covers the requirements for game screen generation. Intent, entities, logic, and auxiliary information are used to avoid missing screen functions or logical confusion due to incomplete information, ensuring sufficient data support when triggering game screen generation. In the semantic chain optimization and screen generation stage, a dynamic semantic chain containing main graph, sub-intents, and entity nodes is constructed based on the game screen requirement list. The weight of sub-intents is calculated by time factor and frequency factor and non-core nodes are eliminated. Then, the target game visual element is located by calling the game visual element mapping library. After being modified by combining player preferences and device characteristics, the game screen interaction logic network is constructed. This principle reduces screen redundancy by distinguishing the priority of requirements and adapts to player and device characteristics, which not only improves the focus of the core screen functions, but also avoids problems such as poor game visual experience and operation feel.

[0132] In summary, this invention achieves a precise and efficient transformation from player voice requirements to adaptive game visuals through deep coupling of principles and effects in each stage: structured processing ensures accurate requirement capture, information completion mechanism guarantees no missing logic in the visuals, semantic chain pruning enhances the focus of core functions, and visual adaptation optimizes the player's interactive experience. This significantly improves the accuracy, adaptability, and player satisfaction of game visual generation in game voice interaction scenarios. The division of labor and cooperation among system modules further ensures the efficient and stable execution of the entire process.

[0133] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A voice-interactive adaptive image generation method, characterized in that, include: The system collects players' game voice input signals in real time, converts them into voice input text fragments, and constructs a structured voice input sequence according to the time sequence of the players' input. Player intent analysis is performed on each newly added structured unit of the voice input sequence to extract the core game elements required for screen generation in a single round of voice interaction and construct a list of game screen requirements. Identify available cells in each column of the game screen requirement list, calculate the availability index of each column of the game screen requirement list, quantify the information richness score of the game screen requirement list, and use it to determine whether to trigger the screen generation operation. If not, start the supplementary query interaction, generate the missing column voice interaction signal to obtain the missing column information. If a screen generation operation is triggered, a dynamic semantic chain containing the root node of the idea graph, child idea nodes, and entity nodes is constructed based on the current game screen requirement list. Based on the available cells in the game screen requirements list, generate the target unit values ​​for each node of the dynamic semantic chain; The time factor and frequency factor of each sub-intent node are calculated to generate the decision weight of the sub-intent node, which is used to prune and optimize the dynamic semantic chain. Based on a pre-defined game visual element mapping library, the target game visual elements of the dynamic semantic chain are located, corrected, and a screen interaction logic network is constructed to generate adaptive screens.

2. The voice-interactive adaptive image generation method as described in claim 1, characterized in that, The steps for constructing the structured speech input sequence include: The game voice input signal is divided into a continuous sequence of voice signal frames according to the preset voice frame length. Text conversion is performed on each speech signal frame in the speech signal frame sequence to generate text segments of the speech signal frames, and adjacent frame text segments are concatenated into a speech input text segment; The system performs word matching on each sentence of the voice input text segment, identifies and removes redundant segments, and performs contextual semantic consistency verification on the voice input text segment, identifies suspected erroneous words in non-redundant segments, and generates corrective candidate words for each suspected erroneous word. Based on the semantic vector of the context of suspected erroneous words, the semantic similarity between the candidate words for correction of each suspected erroneous word and the corresponding speech input text segment is calculated, so as to select the target candidate words for correction to replace the suspected erroneous words. Extract the start and end times of the game voice input signal from which each voice input text segment originates, and construct the input time interval of the voice input text segment; Configure the segment identifier for each voice input text segment, and construct a structured unit for each voice input text segment by combining the start and end acquisition times of its input time interval; Based on the start acquisition time of all structured units, all structured units are sorted in ascending order to generate a set of structured units; then, the structured units are integrated in ascending order according to their start acquisition time to generate a structured speech input sequence.

3. The voice-interactive adaptive image generation method as described in claim 1, characterized in that, The core elements of the game include an intent relationship table, an entity attribute association table, an entity logic relationship table, and a scene-based screen auxiliary information table. The steps for extracting the core game elements required for screen generation in a single-round voice interaction include: When a new structured unit is detected, extract the speech input text segment within that new structured unit; Perform a single-round text cleansing preprocessing on the voice input text segment to generate the cleaned single-round requirement text corresponding to the newly added structured unit; Keywords are extracted from the purified single-round demand text and compared with the preset intent keyword mapping library. The matching degree between the extracted keywords and the keywords corresponding to each idea map category in the intent keyword mapping library is calculated to determine the idea map. The intent keyword mapping library consists of keywords extracted after preprocessing the high-frequency interactive corpus of the target scene, the mapping relationship between keywords and intent graph categories, and keyword weight values. The intent association rule library is invoked to extract the sub-intent categories associated with the identified main intent. Text fragments matching the keywords corresponding to each sub-intent category are extracted from the purified single-round requirement text. Combining the semantic integrity and temporal context of the text fragments, single-round sub-intents are filtered out, and a single-round intent relationship table is constructed. The intent association rule library is used to clarify the sub-intent categories associated with each main intent graph, as well as the keywords corresponding to each sub-intent category.

4. The voice-interactive adaptive screen generation method as described in claim 3, characterized in that, The step of extracting the core game elements required for screen generation in a single-round voice interaction also includes: The contextualized entity dictionary adapted to the target scenario is invoked to match the nouns in the single-round requirement text with the typical words in the contextualized entity dictionary and map them to the corresponding screen entities; the semantic association between screen entities and attributes is mined to determine the attribute names and attribute values ​​corresponding to the screen entities and establish an entity attribute association table. The contextualized entity dictionary includes entity types and corresponding typical words related to the generation of images within the target scene. The target scenario-based predicate library is invoked to extract key predicates and logical connectors from the purified single-round requirement text; combined with the entity attribute association table, the semantic association between predicates, logical connectors and screen entities is analyzed to determine the logical relationship of screen entities; and an entity logical relationship table is established. The target-scenario-based predicate library includes predicates and logical connectors related to screen interaction in the target scenario. Extract player preference parameters and device characteristic parameters for this single-round interaction, and construct a single-round auxiliary information table.

5. The voice-interactive adaptive image generation method as described in claim 1, characterized in that, The steps for constructing the game screen requirements list include: A fixed column structure for the game screen requirements list is pre-constructed, including columns for basic identifiers, screen intents, screen entities, screen logic, and screen presentation. A blank new row is created at the end of the game screen requirements list, and the cell values ​​of all columns are uniformly initialized to a state to be filled. Extract the structured unit identifiers from each core game element and perform consistency checks. If the structured unit identifiers are consistent, fill the corresponding cell in the screen intent category with the structured unit identifier and the corresponding start collection time of the structured unit; otherwise, trigger the exception logging mechanism. After the structured unit identifier consistency check passes, the intent relationship table, entity attribute association table, entity logical relationship table, and single-round auxiliary information table are mapped to the corresponding cells of the screen intent class column, screen entity class column, screen logic class column, and screen presentation class column in sequence. The system identifies multi-value cells containing multi-value data in the intent, entity, and logic columns of the screen, concatenates them uniformly according to preset delimiters, and performs format standardization processing on the filled cells of each column.

6. The voice-interactive adaptive image generation method as described in claim 1, characterized in that, The steps for determining whether a screen generation operation has been triggered include: For the basic identifier category, screen intent category, screen entity category, screen logic category, and screen presentation category in the game screen requirements list, establish a usability index for each category; the usability index value is the ratio of the number of usable cells in the corresponding category to the total number of columns in that category; Availability checks are performed on cells in each column of the game screen requirements list in turn to identify and mark available cells in each column of the game screen requirements list; The system counts the number of available cells in each column and calculates the availability index value for each column. Then, it integrates the availability index values ​​of each column using a weighted summation algorithm to generate an information richness score for the game screen requirement list. A preset information richness threshold is set. If the information richness score is greater than the information richness threshold, the screen generation operation is triggered; otherwise, the columns in the unavailable state of each category are located as missing columns, and the missing columns are sorted according to the query priority of each category. The system calls a contextualized query template library, sorts queries based on their priority for missing columns, generates voice interaction signals to obtain information about missing columns, and continues until the information richness score is greater than the information richness threshold.

7. The voice-interactive adaptive image generation method as described in claim 1, characterized in that, The steps for constructing a dynamic semantic chain that includes the root node of the idea graph, child idea nodes, and entity nodes include: Locate the main intent column in the screen intent category of the game screen requirements list, identify the available cells in the idea image column, classify them according to the semantic consistency and data type of the available cell values, and generate a set of values ​​of the same type for idea images: If the set of values ​​of the same kind in the main graph is unique, then use that set of values ​​of the same kind as the target set of values ​​of the same kind. If the set of values ​​of the same type in the idea graph is not unique, the starting collection time of the structured unit where the main idea of ​​the set of values ​​of the same type is located is extracted. Based on the current interaction time, the time priority of each set is generated by normalizing the value mapping. The frequency of the main idea of ​​each set of values ​​of the same type is counted. The target set of values ​​of the same type is filtered by using the time priority as the main filtering condition and the frequency of the idea graph as the auxiliary filtering condition. The values ​​of the same set of target values ​​are used as the target unit values ​​of the main graph node to construct the main intent root node of the dynamic semantic chain; Locate the sub-intent column of the screen intent category in the game screen requirement list, identify the available cells in the sub-intent column, and generate a set of similar values ​​for the available cells in the sub-intent column to construct sub-intent nodes. Then, establish a connection between all sub-intent nodes and the root node of the main graph through intent hierarchy edges. Construct a set of available cells for entities in each sub-intention, and classify them according to the semantic uniqueness of entity names. Generate a set of values ​​of the same type for entities to construct entity nodes for each sub-intention node and generate target cell values ​​for entity nodes. Establish associations between all entity nodes and corresponding sub-intention nodes through entity intention hierarchy edges. Locate all available cells in the screen logic category column of the game screen requirements list, extract the target cell value of each available cell, match the associated entity pairs with the constructed entity nodes, construct entity logic edges, and connect the corresponding entity nodes.

8. The voice-interactive adaptive image generation method as described in claim 7, characterized in that, The steps for pruning and optimizing the dynamic semantic chain include: Extract the start collection time of the structured unit containing the set of values ​​of the same type for each sub-intention; use the current interaction time at the trigger moment of the dynamic semantic chain construction as the benchmark to calculate the time difference between each start collection time and the current interaction time; use a linear normalization algorithm to map the time difference to a preset numerical range to generate the time factor of the sub-intention node. The frequency of the set of values ​​of the same type for each sub-intention is counted as the original frequency of the sub-intention. The original frequency is processed by the min-max normalization algorithm to generate the frequency factor of the sub-intention. The decision weight of each sub-intention node is generated by performing a fusion calculation through a weighted summation algorithm. Configure the truncation threshold for decision weights, traverse all sub-intention nodes, truncate and remove non-core sub-intention nodes whose decision weights are lower than the truncation threshold, simultaneously delete edge connections associated with non-core sub-intention nodes, and identify and remove isolated nodes.

9. The voice-interactive adaptive image generation method as described in claim 1, characterized in that, The steps for generating adaptive images include: The system calls a pre-defined game visual element mapping library and uses a semantic feature extraction algorithm to generate structured semantic feature vectors for the main intent root node, core sub-intent node, associated entity node, and entity logical edge of the dynamic semantic chain. The semantic feature vectors of each node and logical edge are compared with the feature vectors of game visual elements in the mapping library using a vector cosine similarity matching algorithm. Candidate game visual elements with a cosine similarity greater than a preset similarity threshold are selected. The candidate game visual elements are then sorted to locate the target game visual element. Extract player preference parameters and device characteristic parameters from the visual presentation category in the game visual requirements list, and modify the visual elements of the target game accordingly; The visual elements of the target game are integrated according to the hierarchical structure of the main idea framework, sub-ideas functional areas, entity interaction controls and logical linkage rules to construct a network of screen interaction logic. The visual logic verification mechanism is activated to verify the screen interaction logic network. The screen interaction logic network that passes the verification is converted into a standard screen data format that the device can directly render, and an adaptive screen is generated.

10. A voice-interactive adaptive image generation system, used to implement the voice-interactive adaptive image generation method according to any one of claims 1-9, characterized in that, It includes a sequence construction module, an element extraction module, a requirement determination module, a semantic chain construction module, and a screen generation module; The sequence construction module is used to collect the player's game voice input signal and construct a structured voice input sequence; the element extraction module is used to analyze the player's needs and intentions from the voice input sequence, extract the core game elements required for screen generation, and construct a game screen requirement list; the requirement judgment module is used to quantify the information richness score of the game screen requirement list and determine whether to trigger the screen generation operation; the semantic chain construction module is used to construct a dynamic semantic chain and prune and optimize the dynamic semantic chain when the screen generation operation is triggered; the screen generation module is used to locate the target game visual elements of the dynamic semantic chain based on a preset game visual element mapping library and generate adaptive screens.

Images