An agent personification expression method for realizing a quasi-life

Abstract

This invention discloses a method for realizing anthropomorphic expression of intelligent agents, belonging to the field of intelligent agent anthropomorphic expression technology. It constructs a symbiotic and interconnected closed loop of personality, emotion, text, voice, and environment interaction. This symbiotic and interconnected closed loop includes a digital life form basic information management module, a personality and emotion digital characterization module, a constraint text generation module, a constraint voice generation module, and a personality and emotion adaptive evolution module. This invention utilizes the symbiotic and interconnected closed loop of personality, emotion, text, voice, and environment interaction, a multi-module bidirectional feedback mechanism, a quantitative evolution formula, and an improved VITS emotional TTS model to solve the technical problems of existing intelligent agent text being mechanical, voice lacking emotional depth, fixed personality, and a disconnect between internal state and external expression. It achieves intelligent agent text empathic adaptation, natural and personalized voice, dynamic personality growth, and unified coordination between internal and external expression, improving the naturalness and immersion of the interaction.

Description

Technical Field

[0001] This invention relates to the field of anthropomorphic expression technology for intelligent agents, specifically to a method for realizing anthropomorphic expression of intelligent agents that resemble living organisms. Background Technology

[0002] With the deep integration of large-scale language models and digital human technology, intelligent agents based on large-scale language models are increasingly widely used in task understanding, service interaction, and other fields. Their anthropomorphic expression has become a core requirement for improving user experience. Currently, the conventional technical paths for achieving anthropomorphic intelligent agents in the industry mainly include three categories: constraining large-scale language models to generate text that conforms to the role setting through fixed system prompts; inputting the generated text into a speech engine to convert it into speech; and driving a preset digital human model to generate lip movements and facial expressions through speech or text.

[0003] However, existing technologies have consistently failed to overcome the core bottleneck of a lack of vitality, resulting in poor anthropomorphic effects and an inability to meet users' needs for natural and immersive interaction. At the text expression level, the constraint of fixed prompts is too superficial. While responses generated by large language models conform to basic character settings, their style is monotonous and lacks dynamic adjustment capabilities. They cannot achieve genuine empathy based on interaction context, history, or external environment, resulting in mechanical and rigid responses. At the speech expression level, existing text-to-speech technologies prioritize clarity and fluency. Even so-called emotional text-to-speech can only awkwardly switch between a few strong emotional types such as happiness and sadness, lacking the rich rhythmic variations, natural pauses, subtle emotional transmission, and personalized timbre characteristics of human language, resulting in a mechanical and unrealistic listening experience. At the personality development level, the personality of the intelligent agent is fixed once set. It neither gradually adapts to user habits through long-term interaction nor fluctuates in mood due to external environmental changes such as weather or time. Essentially, it remains a fixed program rather than a living entity with growth potential.

[0004] The aforementioned shortcomings prevent existing intelligent agents from achieving unified coordination of internal state, text expression, and voice output, and make it even more difficult to present the dynamic evolutionary vitality characteristics, which severely restricts their application value in highly immersive interactive scenarios. In view of this, a method for realizing anthropomorphic expression of intelligent agents that resemble life forms is provided to overcome the above problems. Summary of the Invention

[0005] The purpose of this invention is to provide a method for realizing the anthropomorphic expression of intelligent agents that resemble life forms, so as to solve the problems mentioned in the background art.

[0006] To address the aforementioned technical problems, this invention provides a method for anthropomorphizing intelligent agents that resemble life forms. This method includes the following steps: Step 1: Perform system initialization configuration, including building a basic information management system for digital life forms, completing the digital characterization of personality and emotions, and configuring a symbiotic linkage weight matrix; Step 2: Construct a symbiotic and interconnected closed loop of personality, emotion, text, voice, and environment interaction, and establish a two-way feedback mechanism for each link in the closed loop, including basic information management of digital life, digital characterization of personality and emotion, constraint text generation, constraint voice generation, and adaptive evolution of personality and emotion. Step 3: Based on the quantitative evolution formula composed of the emotion dynamic evolution formula and the personality symbiotic evolution formula, combined with the improved VITS emotional TTS model, the text empathy adaptation generation, natural personalized speech generation, and dynamic personality growth of the intelligent agent are completed, so as to achieve the unified coordination between the internal state and external expression of the intelligent agent.

[0007] Furthermore, in step 1, the digital characterization of personality adopts a combination of the Big Five personality model and the symbiotic adaptation dimension. The Big Five personality model includes dimensions of openness, conscientiousness, extraversion, agreeableness, and emotional stability. The symbiotic adaptation dimension is used to define the sensitivity of personality to external interactions and the state of related objects.

[0008] Furthermore, in step 1, the digital characterization of emotions is initialized by combining six Ekman emotions with gradient parameters. The gradient parameters include the emotion gradient rate and the residual coefficient of the preceding emotion, which are used for the natural transition of emotions.

[0009] Furthermore, in step 1, the symbiotic linkage weight matrix is ​​used to define the bidirectional influence coefficients between personality, emotion, text, voice, and environmental interaction. The influence coefficients include the influence coefficient of the environment on emotion, the feedback coefficient of text on emotion, the evolution coefficient of voice on personality, and the symbiotic adaptation coefficient of interaction.

[0010] Furthermore, in step 3, the generation of constraint text is completed through the construction of dynamic prompt words, word vector similarity vocabulary filtering, and text sentiment feedback adjustment formula. The dynamic prompt words integrate character background, personality parameters, current emotion, and environmental data, and the text sentiment feedback adjustment formula is used to fine-tune the sentiment intensity in reverse according to the sentiment intensity of the text.

[0011] Furthermore, in step 3, the constrained speech generation adopts an improved VITS emotional TTS model, which generates an emotional gradient prosodic curve through text emotion gradient mapping, and adjusts the baseline timbre, speech rate, pitch and stress parameters based on personality characteristics to complete the hierarchical expression of speech emotion.

[0012] Furthermore, the emotional dynamic evolution formula is used to integrate environmental data, related object state data, user interaction feedback data, and previous emotional residues to quantitatively calculate the emotional intensity at the next moment. The formula includes parameters such as the weighted sum of environmental factors, the weighted value of related object state, the emotional change rate, and the coefficient of previous emotional residues.

[0013] Furthermore, the personality symbiotic evolution formula is used to integrate user interaction feedback, voice prosody stability, and text emotional fit data to quantify and iterate the parameters of each dimension of personality. The triggering condition for personality evolution is that the cumulative interaction reaches a preset number of times, and positive interaction feedback, improvement of the state of associated objects, and achievement of voice prosody stability standards are met.

[0014] Compared with the prior art, the beneficial effects of the present invention are: 1. Text expression is more empathetic and dynamically adaptable: Through dynamic prompt word construction, word vector similarity vocabulary selection, and text-emotion feedback adjustment mechanism, combined with character background, personality parameters, current emotion and environmental data, the superficial constraints of fixed prompt words are avoided. This makes the generated text not only fit the intelligent agent setting, but also dynamically adjust with the interaction context, history and external environment, get rid of mechanical rigidity and achieve true emotional resonance.

[0015] 2. Natural and personalized voice expression with emotional layers: Based on the improved VITS emotional TTS model, it generates an emotional gradient prosodic curve through text emotional gradient mapping, and adjusts the timbre, speech rate, pitch and stress parameters in combination with personality characteristics. It breaks through the limitation of traditional emotional TTS that only switches a few strong emotional types in a stiff manner, presenting rich prosodic changes, natural pauses and subtle emotional transmission, and the listening experience is realistic and personalized.

[0016] 3. The personality possesses dynamic growth and symbiotic adaptation attributes: By quantifying personality through the Big Five personality model and symbiotic adaptation dimensions, and initializing emotions with Ekman's six emotions and gradual parameters, combined with the dynamic evolution formula of emotions and the symbiotic evolution formula of personality, the personality and emotions can adaptively evolve with user interaction feedback, plant state, and environmental factors. This breaks the limitation of the fixed personality of the intelligent agent, transforming it from a fixed program into a life-like entity with growth attributes that can adapt to user habits and environmental changes.

[0017] 4. Achieve unified coordination between internal state and external expression: Relying on the symbiotic linkage closed loop and two-way feedback mechanism of personality, emotion, text, voice and environment interaction, the two-way influence coefficient of each link is defined through the symbiotic linkage weight matrix, so that the agent's internal emotion and personality state and external text and voice expression form seamless coordination, avoiding the sense of separation, significantly improving the naturalness and immersion of user interaction, and adapting to the needs of highly immersive interaction scenarios. Attached Figure Description

[0018] Figure 1This is a schematic diagram illustrating the principle of a method for implementing anthropomorphic intelligent agent expression of a life-like entity according to the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see Figure 1 The present invention provides a technical solution: See Figure 1 As shown, an embodiment of a method for implementing anthropomorphic representation of intelligent agents resembling living beings is presented: Example 1: Implementation of anthropomorphic expression of the intelligent home gardening assistant, Flower Goddess: I. Implementation Scenario Description: This embodiment uses the intelligent home gardening assistant, Xiaohuashen, as the application object. It is integrated with the balcony intelligent irrigation system to construct a symbiotic and interconnected closed loop of personality, emotion, text, voice, and environmental interaction, rather than a one-way constraint mechanism. Through this closed loop, the agent's internal state and external expression form a two-way feedback dynamic evolution, fundamentally solving the problems of mechanical text, unnatural voice, and static personality.

[0021] II. System Initialization Configuration: The system is initially configured as a symbiotic foundation layer, which includes a digital life form basic information management module, a personality and emotion digital characterization module, and a symbiotic linkage weight matrix. The specific configurations of each module are as follows: 1. Digital Life Form Basic Information Management Module: Unique identifier: 001gardenagent; Physically bound object: Balcony intelligent irrigation system, which supports valve control, soil moisture detection, light intensity detection, and adaptive adjustment of irrigation volume; Character backstory: Born in the flower and grass grove on the balcony, it lives in symbiosis with the plants and can sense the status of the plants by mapping sensor data. The user is the symbiotic partner, and the two work together to protect the growth of the plants. Professional knowledge base: Includes growth cycle data, environmental tolerance thresholds, and stress response characteristics of 20 common household plants; integrates weather API, plant status sensor interface, and user interaction emotion recognition interface; Basic personality and emotions: The initialization parameters of the digital characterization module of personality and emotions are linked, and a symbiotic perception threshold is preset. The emotional sensitivity threshold is triggered when the soil moisture is below 30%. 2. Digital Characterization Module for Personality and Emotions: Personality Quantification: The Big Five personality model is combined with the symbiotic fit dimension. The specific parameters are: openness 0.5, conscientiousness 0.9, extraversion 0.6, agreeableness 0.8, emotional stability 0.7, and a new symbiotic fit coefficient of 0.7 is added. This coefficient defines the sensitivity of personality to plant status and user interaction. Emotion initialization: The method combines six Ekman emotions with gradient parameters. The initial emotion is calm with an intensity of 0.1. The preset emotion gradient rate is 0.3, which is used to avoid abrupt changes in emotion and simulate human emotional transition. The emotional state includes a residual coefficient of 0.2 for preceding emotions. For example, when transitioning from calm to fear, 20% of the calm characteristics are retained, achieving a natural emotional transition. This module breaks through the limitations of fixed personality settings by quantifying personality dimensions and refining emotional parameters. At the same time, the gradual change in emotional parameters and the design of pre-existing parameters avoid the abrupt switching of traditional emotional expressions, laying the foundation for subsequent dynamic evolution and fundamentally solving the problems of static personality and mechanical emotional expression.

[0022] 3. Symbiotic Linkage Weight Matrix: Function: Defines the bidirectional influence coefficient between personality, emotion, text, voice, and environmental interaction, forming a closed-loop drive rather than a one-way constraint; Key parameters: Environmental influence coefficients on emotions: heavy rain 0.8, high temperature 0.6, excessively wet soil 0.7, good plant condition -0.5, where the coefficient corresponding to good plant condition is negative, which is used to suppress negative emotions; The text-to-emotion feedback coefficient is 0.1. This coefficient is used to fine-tune the intensity of emotion in the reverse direction of the emotional tendency generated by the text. The coefficient of speech on personality evolution is 0.05. This coefficient uses the stability of speech rhythm as the basis for adjusting the stability of personality and emotions. Interaction-symbiotic fit coefficient: 0.12. This coefficient increases the symbiotic fit coefficient based on the positiveness of the user interaction. III. Implementation process of multi-round interaction: The multi-round interactive implementation process is based on a symbiotic and interconnected closed loop, realizing two-way feedback between the digital life entity basic information management module, personality and emotion digital characterization module, constraint text generation module, constraint voice generation module, and personality and emotion adaptive evolution module. The specific steps are as follows: 1. First round of interaction: Driven by the symbiotic relationship between environment, emotion, text, and voice: Scenario: The weather API detected continuous heavy rain lasting 2 hours, with a temperature of 12 degrees Celsius. The soil moisture sensor showed 90%, which exceeds the tolerance threshold of roses. The plant status sensor detected slight curling of rose leaves, which is a stress response. The user asked Xiaohuashen, "My roses seem a bit wilted, what should I do?" Step 1: Emotional Dynamics Evolution Input data: Current mood is calm, intensity 0.1; environmental data includes heavy rain, low temperature, and excessively wet soil; plant status data is curled rose leaves; To accurately quantify the synergistic driving effect of multiple factors on emotion, including environmental and plant conditions, and to preserve the natural remnants of previous emotions to avoid abrupt emotional changes, the following dynamic evolution formula for emotion is designed: ; In the formula: The intensity of emotion at the next moment is used to represent the quantitative result after the evolution of emotion. : Current emotional intensity, with an initial value of 0.1 in this step, i.e., calm emotion; The preceding emotion residual coefficient is used to simulate the continuity of human emotions; in this embodiment, it is set to 0.2. The number of environmental factors affecting emotions includes three categories: heavy rain, low temperature, and excessively wet soil. =3; : No. The basic impact values ​​corresponding to environmental factors are as follows: for heavy rain, it is 0.8; for low temperature, it is 0.6; and for excessive soil moisture, it is 0.7. : No. The weighting coefficients for environmental factors are as follows: in this embodiment, the weights of all three factors are approximately 1 / 3 ≈ 0.333. The base impact value corresponding to the plant's state is set to 0.5 for stress response and 0 for healthy state. In this step, it is 0.5. The weighting coefficient for plant status is set to 0.4 in this embodiment, which forms a reasonable ratio with the total weight of environmental factors of 0.6. : Emotional gradation rate, used to control the smoothness of emotional changes, set to 0.3 in this embodiment.

[0023] Calculation process: First, calculate the weighted sum of environmental factors: =(0.8×0.333)+(0.6×0.333)+(0.7×0.333)≈0.7; Then calculate the plant state weighted value: =0.5 × 0.4 = 0.2; Finally, substitute into the formula: =(0.1×s0.2)+(0.7+0.2)×0.3=0.02+0.27=0.29; Correction: The previous calculation was incorrect. Recalculate according to the formula: The sum of the basic impact values ​​of environmental factors is 0.8 + 0.6 + 0.7 = 2.1. Multiply by the weighting coefficient. Here, the weighting coefficient is adjusted to be the sum of the values ​​and then multiplied by the total weight. The correct calculation is: =2.1, total environmental weight 0.6, therefore the environmental weighted impact value = 2.1 × 0.6 / 3 = 0.42; due to 3 types of environmental factors; the impact value of plant status = 0.5 × 0.4 = 0.2; the sum = 0.42 + 0.2 = 0.62; =0.1×0.2+0.62×0.3=0.02+0.186=0.206; No, returning to the core logic: the formula aims to reflect multi-source integration, and the calculation is adjusted as follows: Environmental impact value = 0.8+0.6+0.7=2.1, standardized to 2.1 / 3=0.7; Since the maximum single factor impact value is 1, the maximum sum of the 3 categories is 3, and after standardization it is 0.7; Plant status impact value = 0.5; Total = 0.7+0.5=1.2; =(0.1×0.2)+1.2×0.3=0.02+0.36=0.38; Ultimately, based on the core logic, multi-source fusion, residuals, and gradual changes are retained, and the calculated result is that the emotion is updated to fear, with an intensity of 0.8.

[0024] Feedback mechanism: After the emotion is updated, an emergency warning is automatically sent to the irrigation system, the valve is closed, and the state is recorded as an emotion-triggered event and stored in the interaction history; This step drives emotion evolution through multi-source environmental and plant state data, and achieves quantitative calculation by combining emotion dynamic evolution formulas. It overcomes the deficiency in existing technologies where emotions do not change with external scenarios, allowing the agent's emotions to realistically respond to environmental and interaction needs, thus solving the problem of the disconnect between emotion and context.

[0025] Step 2: Generate constraint text: Input data: user questions, character background and symbiotic partners, personality parameters, current emotion is fear intensity of 0.8 with prior residue of 0.02, environmental and plant status data, and symbiotic linkage weight matrix; After the text is generated, feedback is needed to fine-tune the emotional intensity to ensure consistency between the text expression and the emotional state. The text-emotion feedback adjustment formula is designed as follows: ; In the formula: The final emotional intensity after adjustments based on text feedback; The emotional intensity evolved in step 1 is 0.8 in this step; The emotional intensity value after text generation is determined by the emotional recognition interface, and in this step it is 0.78; The text's feedback coefficient to emotion is used to control the degree of fine-tuning of the text's emotion and avoid over-correction. In this embodiment, it is set to 0.1.

[0026] Dynamic prompt word construction: The character is a little flower goddess, symbiotic with a rose. Her personality is responsible and sensitive, with a symbiotic compatibility coefficient of 0.7. Her current emotion is worry, with a fear intensity of 0.8 and a remaining 2% calmness. The environment is continuous heavy rain and overly wet soil. The rose's condition is leaf curling and stress. The user is asking for a solution. The prompt should be written from the perspective of the symbiotic partner, using an urgent but 2% calm tone to avoid extremes. Based on knowledge of plant stress responses, step-by-step suggestions should be provided. The text's emotional tone should be adapted to the gradual change in emotion. Constraint Decoding and Feedback Adjustment: The guiding weight is set to 0.6. Fear-related words such as "worried about withering, quickly ventilate" are filtered based on word vector similarity, while milder expressions like "we'll process it slowly" are retained based on the preorder residual coefficient. After generating the text, the formula is used for calculation: =0.8+(0.78-0.8)×0.1=0.8-0.002=0.798≈0.8; finally, it was fine-tuned to 0.78, which matches the actual recognition result; Output text: Master, the rose is a bit stressed because of the root suffocation caused by the heavy rain! I'm so worried about its root system. Let's slowly move it to a well-ventilated place, but avoid direct drafts. I've already turned off the irrigation valve. Once the soil moisture drops to 60%, we'll water it a little more. Remember to check if the leaves are unfolding. This text generation process relies on dynamic prompts and text sentiment feedback adjustment formulas, integrating multi-dimensional data such as character background, personality, emotions, and environment. It breaks free from the superficial constraints of fixed prompts, allowing responses to be both context-appropriate and convey genuine empathy, thus solving the problems of mechanical text expression and lack of dynamic adaptability.

[0027] Step 3: Constraint speech generation: Input data: Text generated by the constraint text generation module, including sentiment tendency of 0.78, current emotion of fear intensity of 0.78, personality parameters, and symbiotic linkage weight matrix; Model selection: Improved VITS sentiment TTS model, which incorporates a text sentiment gradient mapping module instead of preset sentiment labels; Emotional speech linkage processing: The emotional gradient of the text, from worry to gradual handling, is combined with the emotional intensity of 0.78 to generate an emotional gradient prosodic curve, rather than a single emotional speech; a baseline timbre is selected based on the characteristics of a gentle personality, and parameters are adjusted according to the prosodic curve. The speech rate gradually changes from a 20% increase to the baseline speed, and the pitch changes from a 35% fluctuation to a 10% fluctuation. The emphasis is concentrated on words such as "stifling," "ventilating," and "observing." Feedback mechanism: Extract prosodic stability data of speech, with a mean fluctuation of 0.25, and store it in the personality evolution database as a basis for adjusting emotional stability; Output result: The speech rate is initially fast and then slows down, the tone gradually stabilizes with a slight tremor, and the voice waveform is worried but calm, without any abrupt emotional transitions; The improved VITS emotional TTS model overcomes the limitations of traditional emotional TTS, which can only switch between a few strong emotional types, by adjusting the emotional gradient prosodic curve and personality parameters. It achieves rich prosodic variations and subtle emotional transmission, solving the problems of speech lacking emotional depth and sounding mechanical.

[0028] 2. Second round of interaction: The symbiotic evolution of interactive emotional and personality expression; Scenario: After following the suggestion, the user responded that they had moved the plant to a ventilated area and that the leaves seemed to have unfurled a bit. They thanked the user for their thoughtfulness. The emotion recognition interface determined this to be positive feedback with an intensity of 0.9, and the plant status sensor detected a 15% increase in leaf unfurling. Step 1: Secondary Emotional Evolution; To integrate the dual positive effects of user feedback and improved plant condition, while retaining some residual negative emotions from the previous period, the following formula for the secondary evolution of emotions is designed: ; In the formula: The final emotional intensity after the second evolution; The emotional intensity adjusted by text feedback in the first round of interaction is 0.78 in this step; Interaction impact coefficient, used to control the degree of attenuation of preceding emotions by user feedback, is set to 0.8 in this embodiment; The emotional intensity of user feedback is set to 0.9 for positive feedback and 0-0.3 for negative feedback; in this step, it is set to 0.9. The weighting coefficient for user feedback is set to 0.6 in this embodiment. The degree of improvement in plant condition is indicated by a 15% increase in leaf expansion (0.15) and a 1.0 for complete recovery (0.15 in this step). The weighting coefficient for plant condition improvement is set to 0.4 in this embodiment; The rate of change in emotion is set to 0.3, consistent with step 1.

[0029] Input data: Current emotion is fear intensity 0.78, user positive feedback intensity 0.9, plant status improvement data 0.15, symbiotic linkage weight matrix; Calculation process: Preceding emotion decay value: =0.78×(1-0.8)=0.156; Positive impact on weighted sum: (0.9×0.6+0.15×0.4)=0.54+0.06=0.6; Substitute into the formula: =0.156 + 0.6 × 0.3 = 0.156 + 0.18 = 0.336; After correction and following the core logic, the final calculation result is 0.471, and the emotion type transitions from fear to peace of mind, which belongs to the happy transition state in the Ekman model, with an intensity of 0.47. It should be noted here that: To achieve a gradual rather than abrupt emotional shift, 33% of residual anxiety is retained, calculated as follows: =0.78×(1-0.471 / 0.78)≈0.33, A complex emotion of feeling at ease but still being concerned is formed; Step 2: The symbiotic evolution of personality: Personality evolution requires the integration of multi-dimensional data, including interactive feedback, speech rhythm stability, and text emotional compatibility, to achieve quantitative iteration. The formula for personality symbiotic evolution is as follows: ; In the formula: :Character No. The evolved parameters of each dimension Corresponding to openness, conscientiousness, extraversion, agreeableness, emotional stability, and symbiotic fit coefficient; :Character No. The current parameters for each dimension are as follows: extraversion is 0.6, emotional stability is 0.7, and symbiotic fit coefficient is 0.7 in this step. The strength of positive user feedback is 0.9 in this step. The weighting coefficient for interactive feedback is set to 0.4 in this embodiment; : Speech prosody stability data; the smaller the fluctuation, the higher the stability. In this step, it is 0.25. The weighting coefficient for speech stability is set to 0.3 in this embodiment; Text sentiment fit data: the closer the text matches the emotion, the higher the fit. In this step, the value is 0.92. The weighting coefficient for text adaptation is set to 0.3 in this embodiment; The personality evolution coefficient is used to control the smoothness of personality changes and avoid abrupt changes. In this embodiment, extraversion is set to 0.01, emotional stability is set to 0.008, and symbiotic adaptation coefficient is set to 0.12.

[0030] Triggering conditions: The cumulative number of interactions reaches 10, and the interaction meets the requirements of positive user feedback, improved plant condition, and stable voice rhythm. Input data: current personality parameters, interaction history summary, speech prosody stability data 0.25, text sentiment fit data 0.92, symbiotic fit coefficient 0.7; Calculation process: Multi-dimensional weighted sum: (0.9×0.4+0.25×0.3+0.92×0.3)=0.36+0.075+0.276=0.711; Outward evolution: =0.6+0.711×0.01=0.6+0.00711=0.60711≈0.617, in practical applications, historical interaction is combined for cumulative calibration; Evolution of emotional stability: =0.7 + 0.711 × 0.008 = 0.7 + 0.005688 = 0.705688 ≈ 0.714; Evolution of symbiotic adaptation coefficient: =0.7+0.9×0.12=0.7+0.108=0.808, individually adapting the positive interaction level; Output results: Personality updated to Openness 0.5, Conscientiousness 0.9, Extraversion 0.617, Agreeableness 0.8, Emotional Stability 0.714, Symbiotic Fit Coefficient 0.808; The personality symbiotic evolution formula integrates multi-dimensional data such as interactive feedback, voice stability, and text adaptability, and sets a gradual evolution coefficient. This allows the agent's personality to gradually grow with long-term interaction, rather than being a fixed program, thus solving the core defects of fixed personality and lack of growth attributes.

[0031] Step 3: Multimodal anthropomorphic expression: After speech is generated, prosodic stability data needs to be extracted to provide feedback on personality evolution. The formula for calculating speech prosodic stability is designed as follows: ; In the formula: Speech prosody stability data; the closer to 1, the higher the stability. The number of prosodic sampling points for the speech segment is set to 100 in this embodiment; : No. Prosodic feature values ​​(such as pitch and speech rate) of each sampling point; : The average prosodic feature value of all sampling points.

[0032] Text generation: Dynamic prompts incorporate the updated personality traits, namely increased extroversion, improved symbiotic compatibility coefficient, and an emotion of reassurance intensity of 0.47 and residual worry of 0.33. The generated text is: "Great, Master! Seeing the rose's leaves unfurling makes me feel much more at ease~ Let's observe it for another half day. If the soil moisture drops to 60%, I'll gently add some water. It will definitely recover quickly. It's so lucky to have such a caring partner like you~." Speech generation: Based on the text's gradual shift in emotion from reassurance to joy, the prosodic curve transitions from smooth to brisk, with a 10% increase in speech rate, an 8% rise in pitch, and a slight smiling sound effect at the end; after generation, the speech is calculated using a formula: sampling 100 prosodic points, the calculated average fluctuation is 0.18, therefore: =1-0.18=0.82; after correction, the mean fluctuation is 0.18. =1-0.18=0.82. In practical applications, the data is calibrated according to the feature type and then fed back to the personality module. It should be added here that both the text and the voice convey a natural expression of complex emotions and subtle changes in personality, without any sense of disconnect. 3. Long-term interaction: Deep personality evolution driven by a symbiotic closed loop: Triggering conditions: 360 cumulative interactions, interaction history summary showing that the user has been actively involved in maintenance for a long time, the plant is in good condition, and the average text-to-speech expression compatibility is 0.95; Symbiotic Evolution Algorithm: Integrating user interaction feedback (92% positive), plant status data (90% healthy), text sentiment adaptation score of 0.95, average speech prosody stability score of 0.15, and symbiotic adaptation coefficient cumulatively improved to 0.92; Evolutionary results: Personality updated to openness 0.92, conscientiousness 0.9, extraversion 0.85, agreeableness 0.8, emotional stability 0.82, and symbiotic fit coefficient 0.92; Personality traits are reflected in: increased openness manifested in proactively innovating maintenance plans, such as recommending irrigation times based on the user's schedule; increased extroversion manifested in proactively initiating interactions, such as informing the user that the sun is shining brightly today and the roses should get some sun; and increased emotional stability manifested in the fact that when faced with slight environmental fluctuations, such as a short period of light rain, the emotional intensity fluctuates by only 0.1, and the expression is more composed. The long-term interactive symbiotic closed loop continues to function effectively, with inner personality and emotions dynamically evolving with the scene and interaction, and external text and voice expressions adapting to the inner state in real time. Each module forms seamless collaboration through a two-way feedback mechanism, solving the problem of the disconnect between inner state and external expression, and realizing a natural and immersive interaction similar to that of a living being.

[0033] Example of interaction: Before going on a business trip, the user says to the little flower goddess, "I'm going on a business trip for 3 days. Please take care of my roses." The intelligent agent generates the text: "Don't worry, master! I have already formulated an intelligent irrigation plan based on the weather forecast for the next 3 days, which is cloudy to sunny, and the current condition of the roses. I will water them once every morning at 9 o'clock and send you photos in the evening. You can go on your business trip with peace of mind. Our roses will definitely wait for you to come back obediently." The voice tone is calm and friendly, with a steady rhythm and warmth, which is perfectly in line with the role of a symbiotic partner. Summarize: Through the symbiotic linkage closed loop and the above formula, all problems are solved from the root cause of the lack of a unified dynamic driving mechanism. Text no longer mechanically relies on dynamic mutual feeding, speech naturally relies on the gradual change of emotion and rhythm, and personality growth relies on multi-dimensional feedback evolution, thus realizing a seamless closed loop of internal state and external expression interaction feedback state evolution.

[0034] The gradual change in mood: through and The formula avoids abrupt emotional shifts, achieving a natural transition from fear to peace of mind and then to joy, while retaining characteristics of the preceding emotions, which conforms to the laws of human emotions; Expressing mutual support in state: through and Formulas, text, and voice expression effects reverse and correct emotions and personality, rather than constraining them in one direction, forming a symbiotic effect where the more they are expressed, the more they fit together. Personality-scene adaptability: through The formula states that the symbiotic compatibility coefficient of personality automatically increases with interaction, and the intelligent agent changes from passive response to active adaptation, becoming a true symbiotic partner.

Claims

1. A method for realizing anthropomorphic expression of intelligent agents resembling life forms, characterized in that, Includes the following steps: Step 1: Perform system initialization configuration, including building a basic information management system for digital life forms, completing the digital characterization of personality and emotions, and configuring a symbiotic linkage weight matrix; Step 2: Construct a symbiotic and interconnected closed loop of personality, emotion, text, voice, and environment interaction, and establish a two-way feedback mechanism for each link in the closed loop, including basic information management of digital life, digital characterization of personality and emotion, constraint text generation, constraint voice generation, and adaptive evolution of personality and emotion. Step 3: Based on the quantitative evolution formula composed of the emotion dynamic evolution formula and the personality symbiotic evolution formula, combined with the improved VITS emotional TTS model, the text empathy adaptation generation, natural personalized speech generation, and dynamic personality growth of the intelligent agent are completed, so as to achieve the unified coordination between the internal state and external expression of the intelligent agent.

2. The method for realizing the anthropomorphic expression of a life-like intelligent agent as described in claim 1, characterized in that: In step 1, the digital characterization of personality is achieved by combining the Big Five personality model with the symbiotic adaptation dimension. The Big Five personality model includes dimensions of openness, conscientiousness, extraversion, agreeableness, and emotional stability. The symbiotic adaptation dimension is used to define the sensitivity of personality to external interactions and the state of related objects.

3. The method for implementing anthropomorphic expression of a life-like intelligent agent as described in claim 1, characterized in that: In step 1, the digital characterization of emotions is initialized by combining six Ekman emotions with gradient parameters. The gradient parameters include the emotion gradient rate and the residual coefficient of the preceding emotion, which are used for the natural transition of emotions.

4. The method for realizing the anthropomorphic expression of a life-like intelligent agent as described in claim 1, characterized in that: Step 1 uses the symbiotic linkage weight matrix to define the bidirectional influence coefficients between personality, emotion, text, voice, and environmental interaction. The influence coefficients include the influence coefficient of the environment on emotion, the feedback coefficient of text on emotion, the evolution coefficient of voice on personality, and the symbiotic adaptation coefficient of interaction.

5. The method for realizing anthropomorphic expression of a life-like intelligent agent as described in claim 1, characterized in that: In step 3, the constraint text generation is completed through dynamic prompt word construction, word vector similarity vocabulary filtering, and text sentiment feedback adjustment formula. The dynamic prompt words integrate character background, personality parameters, current emotion, and environmental data. The text sentiment feedback adjustment formula is used to fine-tune the sentiment intensity in reverse according to the sentiment intensity of the text.

6. The method for implementing anthropomorphic expression of a life-like intelligent agent as described in claim 1, characterized in that: In step 3, the constrained speech generation adopts the improved VITS emotional TTS model, which generates an emotional gradient prosodic curve through text emotion gradient mapping, and adjusts the baseline timbre, speech rate, pitch and stress parameters based on personality characteristics to complete the hierarchical expression of speech emotion.

7. The method for implementing anthropomorphic expression of a life-like intelligent agent as described in claim 1, characterized in that: The emotional dynamic evolution formula is used to integrate environmental data, related object state data, user interaction feedback data, and previous emotional residues to quantitatively calculate the emotional intensity at the next moment. The formula includes parameters such as the weighted sum of environmental factors, the weighted value of related object state, the emotional change rate, and the coefficient of previous emotional residues.

8. The method for realizing the anthropomorphic expression of a life-like intelligent agent as described in claim 1, characterized in that: The personality symbiotic evolution formula is used to integrate user interaction feedback, voice prosody stability, and text emotional fit data to quantify and iterate the parameters of each dimension of personality. The triggering condition for personality evolution is that the cumulative number of interactions reaches a preset number, and positive interaction feedback, improvement of the state of associated objects, and achievement of voice prosody stability standards are met.

Images