A pet companion voice generation method and device based on voiceprint modeling, equipment and storage medium

Abstract

The present application relates to a kind of pet companion voice generation method, device, equipment and storage medium based on voiceprint modeling.The method comprises: obtaining the voiceprint feature of target user and constructing personalized timbre model based on quality score;Pet behavior data collected by pet companion terminal is obtained, the behavior category of pet is identified, and the behavior intensity representing the intensity of pet behavior;According to the behavior category, determine the basic tone template, and dynamically adjust the voice parameter in the basic tone template according to the preset mapping relationship based on the behavior intensity, obtain target tone parameter;Based on the target tone parameter, the personalized timbre model and the target text matched with the behavior category, synthesis target companion voice and send to the pet companion terminal for playing.The present application realizes emotional, adaptive pet automation companion, and through recording quality score mechanism and sensitive content filtering mechanism, ensures modeling accuracy and use safety.

Description

Technical Field

[0001] This invention belongs to the field of intelligent voice interaction and intelligent pet devices, and particularly relates to a method, device, equipment and storage medium for generating pet companion voice based on voiceprint modeling. Background Technology

[0002] Existing pet companion voice devices have the following technical shortcomings: Monotonous voice: Existing devices mostly use preset or standard synthesized voices, which lack the owner's voice characteristics, making it difficult for pets to form emotional connections and resulting in poor companionship.

[0003] Interactive passive: Usually requires users to manually record in real time or operate remotely, and cannot automatically adjust the tone of voice according to the pet's real-time status.

[0004] Security risks: Voice cloning technology carries the risk of misuse and lacks effective mechanisms for filtering sensitive content and tracing copyright. Summary of the Invention

[0005] The purpose of this invention is to provide a method, apparatus, device, and storage medium for generating pet companion voice based on voiceprint modeling, so as to achieve emotional and adaptive automated pet companionship, and to ensure modeling accuracy and security of use through a recording quality scoring mechanism and a sensitive content filtering mechanism.

[0006] This invention provides a method for generating pet companion voice based on voiceprint modeling, comprising the following steps: Step 1: Obtain the voiceprint features of the target user and build a personalized voice model based on the quality score; Step 2: Obtain pet behavior data collected by the pet companion terminal, identify the pet's behavior categories and the intensity of the behavior indicating the intensity of the pet's behavior; Step 3: Determine the basic tone template based on the behavior category, and dynamically adjust the speech parameters in the basic tone template according to the preset mapping relationship based on the behavior intensity to obtain the target tone parameters; Step 4: Based on the target tone parameters, the personalized timbre model, and the target text matching the behavior category, synthesize the target companion voice and send it to the pet companion terminal for playback.

[0007] Furthermore, the step 1 of constructing a personalized timbre model based on quality scores includes: 1) Perform multi-index quantitative scoring on the target voice samples recorded by the user to obtain the quality score Q; 2) Divide the target speech sample into multiple speech slices according to semantics, and score each slice independently; 3) Only prompt the user to re-record for voice segments whose scores are below a preset threshold; 4) Input the verified standard speech segment into the speaker coding network to extract the speaker vector. and scored by quality Aggregate the weights to obtain the global timbre representation vector. .

[0008] Furthermore, the calculation logic for the behavior intensity includes: Audio dimension: determined based on the sound pressure level (dB) or fundamental frequency change rate of the pet's vocalizations; Image dimension: determined based on the displacement vector of the pet's movement target, the proportion of the movement amplitude, or the movement frequency of the skeletal joints.

[0009] Furthermore, the mapping relationship includes a linear proportional mapping relationship or a nonlinear function mapping relationship; the target text is generated in real time by a pre-trained natural language generation model based on the behavior category and environmental context information.

[0010] Furthermore, step 3 also includes: Identify the duration of the pet's behavior; If the duration exceeds a preset threshold, the current basic tone template will be upgraded to a target template with a higher intervention level, and a non-voice intervention strategy will be superimposed.

[0011] Furthermore, the method also includes: After playing the target companion voice, the pet's feedback behavior data is continuously collected; If the feedback behavior data indicates that the pet's behavior has not improved, the adjustment intensity of the target tone parameter will be automatically increased or the tone template will be switched to form an interactive closed loop.

[0012] Furthermore, the method also includes security defense steps: Before synthesizing the target accompanying voice, the target text is semantically classified and detected. If sensitive semantics involving financial payments, identity information, or verification codes are identified, the voice generation is blocked. An imperceptible digital watermark containing device identification and timestamp information is embedded in the synthesized target companion voice.

[0013] The present invention also provides a pet companion voice generation device based on voiceprint modeling, for performing the method, including: The timbre modeling module is used to build personalized timbre models for target users; The behavior recognition module is used to identify the pet's behavior categories and the intensity of the behavior, indicating its severity. The tone mapping module is used to determine the basic tone template based on the behavior category and dynamically adjust the speech parameters based on the behavior intensity to obtain the target tone parameters; The speech synthesis module is used to generate and output the target companion speech with the owner's timbre and the target's tone characteristics.

[0014] The present invention also provides an electronic device, including a memory and a processor having a computer program stored thereon, wherein the processor executes the program to implement the method described thereon.

[0015] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described thereon.

[0016] The above-described solution, through a method, apparatus, device, and storage medium for generating pet companion voice based on voiceprint modeling, achieves the following technical effects: 1) Highly realistic companionship: Achieves high-precision reproduction of the owner's voice, enhancing the pet's trust and emotional stability.

[0017] 2) Low latency response: Through the end-to-cloud collaborative architecture and voice caching mechanism, both generation quality and real-time interaction are taken into account.

[0018] 3) High security: Effectively identifies and blocks the generation of high-risk content in non-accompaniment scenarios, and complies with the protection standards for biometric data.

[0019] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Attached Figure Description

[0020] Figure 1 This is a flowchart of the pet companion voice generation method based on voiceprint modeling according to the present invention; Figure 2 This is a schematic diagram of the overall system architecture of the present invention; Figure 3 This is a flowchart of the recording quality scoring and re-recording mechanism in one embodiment of the present invention; Figure 4 This is a schematic diagram of the master voice timbre modeling process in one embodiment of the present invention; Figure 5 This is a schematic diagram of the pet behavior recognition module structure in one embodiment of the present invention; Figure 6 This is a flowchart illustrating the linkage between behavior recognition and tone mapping in one embodiment of the present invention; Figure 7 This is a schematic diagram of the tabular logic of the tone mapping strategy in one embodiment of the present invention; Figure 8 This is a schematic diagram of the speech generation module structure in one embodiment of the present invention; Figure 9is a schematic diagram of the end-cloud collaborative processing architecture according to an embodiment of the present invention; Figure 10 is a flowchart of end-cloud collaborative communication and abnormal handover according to an embodiment of the present invention; Figure 11 is a schematic diagram of the overall framework of the security mechanism according to an embodiment of the present invention Figure 12 is a flowchart of encrypted storage and access control of a timbre model according to an embodiment of the present invention; Figure 13 is a flowchart of digital watermark embedding, tracing and verification according to an embodiment of the present invention; Figure 14 is a flowchart of closed-loop control for pet stress detection and protection strategies according to an embodiment of the present invention; Figure 15 is a comparison diagram of the success rate of pet behavior correction according to an embodiment of the present invention; Figure 16 is a comparison diagram of the duration of pet emotional stability after interaction according to an embodiment of the present invention. DETAILED DESCRIPTION OF EMBODIMENTS

[0021] The specific embodiments of the present invention are described in further detail below with reference to the accompanying drawings and embodiments. The following embodiments are intended to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0022] With reference Figures 1 to 14 to the illustration, the present embodiment provides a pet companionship speech generation method based on voiceprint modeling, comprising the following steps: Step S1: acquiring the voiceprint feature of the target user and constructing a personalized timbre model based on a quality score; Step S2: acquiring pet behavior data collected by a pet companionship terminal, and identifying a behavior category of the pet and a behavior intensity representing the severity of the pet's behavior; Step S3: determining a basic intonation template according to the behavior category, and dynamically adjusting the speech parameters (such as volume, speech rate, and pitch) in the basic intonation template based on the behavior intensity according to a preset mapping relationship to obtain target intonation parameters; Step S4: synthesizing target companionship speech based on the target intonation parameters, the personalized timbre model, and a target text matching the behavior category, and transmitting the target companionship speech to the pet companionship terminal for playback.

[0023] The present invention achieves emotional and adaptive automated pet companionship, and ensures modeling accuracy and use safety through a recording quality scoring mechanism and a sensitive content filtering mechanism.

[0024] Further, the constructing a personalized timbre model based on a quality score in step S1 specifically comprises: 1) Perform multi-index quantitative scoring on the target voice samples recorded by the user to obtain the quality score Q; 2) Divide the target speech sample into multiple speech slices according to semantics, and score each slice independently; 3) Only prompt the user to re-record for voice segments whose scores are below a preset threshold; 4) Input the verified standard speech segment into the speaker coding network to extract the speaker vector. and scored by quality Aggregate the weights to obtain the global timbre representation vector. .

[0025] Furthermore, the calculation logic for the behavior intensity includes: Audio dimension: determined based on the sound pressure level (dB) or fundamental frequency change rate of the pet's vocalizations; Image dimension: determined based on the displacement vector of the pet's movement target, the proportion of the movement amplitude, or the movement frequency of the skeletal joints.

[0026] Furthermore, the mapping relationship includes a linear proportional mapping relationship or a nonlinear function mapping relationship; the target text is generated in real time by a pre-trained natural language generation model based on the behavior category and environmental context information.

[0027] Furthermore, step S3 also includes: Identify the duration of the pet's behavior; If the duration exceeds a preset threshold, the current basic tone template will be upgraded to a target template with a higher intervention level, and a non-voice intervention strategy will be superimposed.

[0028] Furthermore, the method also includes: After playing the target companion voice, the pet's feedback behavior data is continuously collected; If the feedback behavior data indicates that the pet's behavior has not improved, the adjustment intensity of the target tone parameter will be automatically increased or the tone template will be switched to form an interactive closed loop.

[0029] Furthermore, the method also includes security defense steps: Before synthesizing the target accompanying voice, the target text is semantically classified and detected. If sensitive semantics involving financial payments, identity information, or verification codes are identified, the voice generation is blocked. An imperceptible digital watermark containing device identification and timestamp information is embedded in the synthesized target companion voice.

[0030] The present invention also provides a pet companion voice generation device based on voiceprint modeling, for performing the method, including: The timbre modeling module is used to build personalized timbre models for target users; The behavior recognition module is used to identify the pet's behavior categories and the intensity of the behavior, indicating its severity. The tone mapping module is used to determine the basic tone template based on the behavior category and dynamically adjust the speech parameters based on the behavior intensity to obtain the target tone parameters; The speech synthesis module is used to generate and output the target companion speech with the owner's timbre and the target's tone characteristics.

[0031] The present invention will be further described in detail below through specific embodiments.

[0032] Example 1: Personalized Timbre Modeling Based on Quality Scoring This embodiment details how to ensure the accuracy of the timbre model through recording quality control.

[0033] S101 Guided Data Collection: The system displays commonly used pet interaction texts such as "Don't worry, my darling, I'll be right back" to the user, guiding them to record a 3-8 second clip on their mobile device.

[0034] S102 Quality Quantification Scoring: Scores are calculated using a weighted linear model. It covers indicators such as signal-to-noise ratio, reverberation, and clipping distortion.

[0035] S103 Fault Cause Location and Re-recording: If Q < 70, the system will accurately prompt "excessive noise" or "too many pauses" and trigger segmented re-recording. Only low-segment segments need to be re-recorded, reducing the user's burden.

[0036] S104 Voiceprint Vector Generation: Input the verified standard speech segment into the voiceprint coding network to extract the voiceprint vector. and scored by quality Aggregate the weights to obtain the global timbre representation vector. .

[0037] Example 2: Interaction Triggering and Text Generation Based on Behavior Recognition This embodiment describes how the system automatically identifies pet needs and generates semantically aligned text.

[0038] S201 Multimodal Perception: The companion terminal collects image sequences through a camera, collects pet sounds through a microphone, and records environmental context information E (such as day / night, whether there are strangers).

[0039] S202 Multidimensional Quantification of Behavior: The behavior recognition model outputs behavior category B, behavior intensity S (0~1 continuous value) and duration T.

[0040] S2021 Intensity Parameter Extraction: When the behavior category B is identified as "barking", the system extracts the root mean square energy (RMS Energy) of the audio signal as a sound pressure level indicator.

[0041] S2022 Action Feature Association: When "destructive" behavior is detected, motion vectors in the pet image sequence are extracted using an optical flow algorithm, and the proportion of the action amplitude to the screen is normalized to a continuous value S of 0 to 1.

[0042] S203 Event Trigger Decision: When it is detected that "the pet lingers at the door for more than 10 seconds" or "scratches on the sofa for more than 5 seconds", the corresponding interaction event is automatically triggered.

[0043] S204 Semantic Generation: Based on the event type, generate target text such as "Don't bite this, stop!" through a pre-trained natural language generation model or a preset speech library.

[0044] Example 3: Behavior Intensity-Driven Tone Mapping Strategy This embodiment implements emotion adaptation in voice feedback.

[0045] Option A (Expert Rule): Preset a value for k, such as the speech rate adjustment coefficient. Setting it to 0.5 makes the speech rate increase linearly with intensity.

[0046] Option B (Reinforcement Learning Optimization): Based on the "effective interaction" feedback records in Example 9, the system uses a reinforcement learning algorithm to automatically iterate the k value, making the tone adjustment more in line with the psychological expectations of a specific pet.

[0047] S301 Tone Template Selection: Match templates based on behavior category B, such as matching "warning template" for destructive behavior and "encouragement template" for eating behavior.

[0048] S302 Parameter Dynamic Adjustment Formula: Adjust tone parameters in real time based on behavioral intensity $S$. , ; The more intense the behavior, the more intimidating or commanding the owner's voice becomes.

[0049] S303 Duration-Driven Template Upgrade: If destructive behavior continues for more than 30 seconds, the tone automatically upgrades from "mild reprimand" to "strong warning," and a feeding incentive strategy is added to divert the pet's attention.

[0050] Example 4: Streaming Speech Synthesis and Digital Watermarking Embedding This example illustrates how to balance real-time interactivity with data traceability.

[0051] S401 Hybrid synthesis: The TTS model loads the user's timbre vector and the tone parameter set P to generate the Wave_out audio signal.

[0052] S402 Streaming segmented output: Adopt a streaming generation mechanism, play while generating, to minimize interaction latency.

[0053] S403 Traceable watermark embedding: Embed an imperceptible digital watermark containing the device ID and timestamp into the audio, to prevent the owner's voiceprint data from being misused through secondary recording.

[0054] Embodiment 5: Cache acceleration mechanism based on index structure This embodiment reduces the edge-cloud communication cost through pre-generation technology.

[0055] S501 Cache library construction: Pre-synthesize high-frequency sentences such as "Don't be afraid, good boy" under different tone templates, and store them in the local CacheDB.

[0056] S502 Fast matching logic: The system performs index query based on Text_hash + tone template T + timbre ID, and achieves zero-latency playback if a match is hit.

[0057] Embodiment 6: Dynamic selection of edge-cloud collaborative architecture This embodiment improves the system's adaptability to complex environments.

[0058] S601 Local offline disaster recovery: When the network is unavailable, the system automatically degrades, uses the edge-side lightweight recognition model and basic speech library to ensure basic companion functions.

[0059] S602 Cloud high-fidelity mode: When the network is stable, highly realistic timbre synthesis and deep semantic understanding are realized through the cloud large model.

[0060] S603 Intelligent scheduling decision: The system automatically switches the generation mode according to the formula Mode = Select(Net, CPU, Latency, S).

[0061] Embodiment 7: Sensitive semantic filtering and blacklist control This embodiment ensures the security of voice cloning technology at the social level.

[0062] S701 Sensitive semantic blocking: Scan the text before synthesis. If the content involves non-pet companion scenarios such as "transfer", "verification code" or "identity certification", the system will forcibly refuse to generate.

[0063] S702 Abnormal Frequency Limit: When abnormal behavior such as generating long continuous segments of speech at high frequency within a short period of time is detected, access to the corresponding timbre file is automatically blocked.

[0064] Example 8: Liveness Detection-Based Tone Licensing Protection This embodiment prevents the use of third-party recordings to impersonate the owner for modeling.

[0065] S801 Random Challenge Response: During the modeling phase, the system randomly generates a temporary text and asks the user to read it aloud, detecting the matching degree between the speech rhythm and the text to complete the liveness verification.

[0066] S802 hardware-level encrypted storage: stores timbre model parameters Stored in a Trusted Execution Environment (TEE) on the client side to ensure that the model file cannot be illegally exported.

[0067] Example 9: Interactive Closed-Loop Feedback and Adaptive Optimization This embodiment enables the device to have an interactive capability that "becomes smarter the more it is used".

[0068] S901 Feedback Effect Detection: After broadcasting, the system continuously monitors the pet's behavior. If "stop barking" or "turn to toy" is detected, it is considered a valid interaction.

[0069] S902 Automatic Strategy Evolution: If the pet reacts to the current tone of voice with "stress" or "indifference", the system will automatically adjust the volume and speech rate parameters in the behavior mapping table, or switch to a soothing tone.

[0070] Example 10: Comprehensive Implementation Scenario of the Entire Process This example illustrates a typical scenario of "office workers accompanying each other in different locations".

[0071] The owner recorded voice messages and completed the modeling before going to work.

[0072] During midday, the pet exhibited "anxious pacing," and the terminal identified low-intensity anxiety (S=0.3).

[0073] The system matches a "gentle and soothing template" and generates a voice message with the owner's voice tone, saying "Baby, I'm here, don't be afraid," which is played at a low volume.

[0074] The sensor detects that the pet has entered a resting state, and the system uploads this successful case to the cloud to optimize strategy parameters.

[0075] This invention ensures data quality by collecting user voice samples and utilizing a multi-metric scoring mechanism (SNR, reverberation, distortion, etc.). Based on high-quality samples, a personalized voiceprint embedding vector for the target user is constructed. During operation, the system identifies the pet's behavior category B and intensity S in real time, and dynamically maps matching tone parameters (speech rate, pitch, volume, etc.) to drive the synthesis of a companion voice with the owner's timbre and appropriate emotion. It features the following technological innovations: 1. A dynamic tone regulation mechanism driven by behavioral intensity The system is not a simple "behavior-speech" matching system, but introduces a continuous behavior intensity value S.

[0076] Dynamic control formula: through formula , and Real-time correction of voice parameters is achieved.

[0077] A tiered intervention strategy: By combining the duration of the behavior (T), the tone template is automatically upgraded from "mild reassurance" to "forceful instructions" and then to "feeding guidance," which significantly improves the effectiveness of behavior correction.

[0078] 2. Quantitative scoring of recording quality and segmented re-recording Quantitative evaluation: A recording quality scoring formula Q based on a weighted linear model is proposed to normalize the calculation of multi-dimensional indicators such as signal-to-noise ratio and pronunciation integrity, so as to ensure the accuracy of modeling from the source.

[0079] Low-cost error correction: It supports a segmented re-recording mechanism that "only re-records low-segment segments", avoiding the inefficiency of re-recording the entire segment in existing technologies.

[0080] 3. Closed-loop feedback and safety compliance system Closed-loop interaction: After the voice broadcast, the system continuously monitors the pet's feedback. If the behavior does not improve, the system automatically adjusts the tone or frequency of the voice to form an adaptive control closed loop.

[0081] Multiple security defenses: Integrating sensitive semantic filtering, liveness detection verification, encrypted model storage, and imperceptible digital watermarking technology, comprehensively reducing the risk of voice abuse and data leakage.

[0082] Figure 15 The "step-by-step intervention strategy" and "behavior intensity-driven tone regulation" were validated.

[0083] Indicator description: Three typical scenarios were selected: "barking control", "stopping destructive behavior", and "relieving separation anxiety".

[0084] Comparison of options: Control group (Baseline): Standard synthesized speech (monotonous timbre, fixed tone) commonly used in existing technologies.

[0085] The proposed invention uses a cloned master voice and combines it with behavioral intensity S to dynamically adjust speech rate, volume, and pitch.

[0086] Technical Results: Experimental data showed that, with dynamic tone regulation and tiered intervention, the compliance rate of pets with high-intensity behaviors such as "destroying the house" increased by approximately 171% (from 28% to 76%), significantly enhancing the effectiveness of behavior correction.

[0087] Figure 16 It aims to support the technical effects of "highly realistic companionship" and "adaptive interactive closed loop".

[0088] Metric Description: Records the average duration for which a pet remains calm (without pacing anxiously or whimpering) after a single interaction with its owner while the owner is away from home.

[0089] Comparison of options: Standard synthesized speech: Pets recognize non-owner voices and remain emotionally stable for only about 18 minutes.

[0090] Recordings with the same owner: Although it is the owner's voice, the lack of emotional feedback can easily cause "auditory fatigue" in pets. The duration of stable recordings is about 45 minutes.

[0091] This invention dynamically generates speech: it restores highly realistic timbre through voiceprint modeling and adjusts the tone of voice in real time according to the pet's feedback, giving the pet the illusion that "the owner is present", and significantly increasing the duration of emotional stability to 115 minutes.

[0092] The present invention also provides an electronic device, including a memory and a processor having a computer program stored thereon, wherein the processor executes the program to implement the method described thereon.

[0093] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described thereon.

[0094] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for generating pet companion voice based on voiceprint modeling, characterized in that, Includes the following steps: Step 1: Obtain the voiceprint features of the target user and build a personalized voice model based on the quality score; Step 2: Obtain pet behavior data collected by the pet companion terminal, identify the pet's behavior categories and the intensity of the behavior indicating the intensity of the pet's behavior; The calculation logic for the intensity of the behavior includes: Audio dimension: determined based on the sound pressure level (dB) or fundamental frequency change rate of the pet's vocalizations; Image dimensions: determined based on the displacement vector of the pet's moving target, the amplitude ratio of the movement, or the movement frequency of the skeletal joints; Step 3: Determine the basic tone template based on the behavior category, and dynamically adjust the speech parameters in the basic tone template according to the behavior intensity and a preset mapping relationship to obtain the target tone parameters; the mapping relationship includes a linear proportional mapping relationship or a non-linear function mapping relationship; the target text is generated in real time by a pre-trained natural language generation model based on the behavior category and environmental context information; Identify the duration of the pet's behavior; If the duration exceeds a preset threshold, the current basic tone template will be upgraded to a target template with a higher intervention level, and a non-voice intervention strategy will be superimposed. Step 4: Based on the target tone parameters, the personalized timbre model, and the target text matching the behavior category, synthesize the target companion voice and send it to the pet companion terminal for playback; The step 1, which involves constructing a personalized timbre model based on quality scoring, includes: 1) Perform multi-index quantitative scoring on the target speech samples recorded by the user, and calculate the quality score Q using a weighted linear model; the indices include signal-to-noise ratio, reverberation, and clipping distortion; 2) Divide the target speech sample into multiple speech slices according to semantics, and score each slice independently; 3) Only prompt the user to re-record for voice segments whose scores are below a preset threshold; 4) Input the verified standard speech segment into the speaker coding network to extract the speaker vector. and scored by quality Aggregate the weights to obtain the global timbre representation vector. .

2. The method for generating pet companion voice based on voiceprint modeling according to claim 1, characterized in that, Also includes: After playing the target companion voice, the pet's feedback behavior data is continuously collected; If the feedback behavior data indicates that the pet's behavior has not improved, the adjustment intensity of the target tone parameter will be automatically increased or the tone template will be switched to form an interactive closed loop.

3. The method for generating pet companion voice based on voiceprint modeling according to claim 1, characterized in that, It also includes security defense measures: Before synthesizing the target accompanying voice, the target text is semantically classified and detected. If sensitive semantics involving financial payments, identity information, or verification codes are identified, the voice generation is blocked. An imperceptible digital watermark containing device identification and timestamp information is embedded in the synthesized target companion voice.

4. A pet companion voice generation device based on voiceprint modeling, characterized in that, For performing the method according to any one of claims 1 to 3, comprising: The timbre modeling module is used to build personalized timbre models for target users; The behavior recognition module is used to identify the pet's behavior categories and the intensity of the behavior, indicating its severity. The tone mapping module is used to determine the basic tone template based on the behavior category and dynamically adjust the speech parameters based on the behavior intensity to obtain the target tone parameters; The speech synthesis module is used to generate and output the target companion speech with the owner's timbre and the target's tone characteristics.

5. An electronic device, characterized in that, It includes a memory and a processor having a computer program stored thereon, wherein the processor executes the program to implement the method as described in any one of claims 1 to 3.

6. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 3.

Images