20results about How to "Improve naturalness" patented technology

The application provides a kind of based on user behavior perception's augmented reality content recommendation and expressive presentation method, including constructing the state representation of multiple virtual roles in augmented reality scene;Joint objective function for evaluating virtual role layout rationality is constructed;According to the attribute information of virtual role, the adaptive weight is generated by the weight prediction model constructed in advance, and the adaptive weight is used to adjust the cost item related to the importance of role in the space cost function;Based on the joint objective function, the position and orientation of all virtual roles are iteratively optimized to obtain an optimized layout;Expressive presentation is carried out in the augmented reality scene The optimized layout is realized in three-dimensional space Automatic optimization of role position and orientation, avoid visual conflict, improve the naturalness, coordination and semantic expression ability of layout, thereby significantly enhance the interactive experience and information transmission effect of AR system.

The application discloses a language decoding method and device, electronic equipment and storage medium. The method comprises the following steps: obtaining an electroencephalogram decoding result of a current time step, wherein the electroencephalogram decoding result comprises electroencephalogram decoding scores corresponding to each preset syllable; expanding a historical candidate sentence of a previous time step based on candidate words associated with the preset syllable to obtain a corresponding expanded sentence set; for each expanded sentence, determining an integrated language decoding score of the expanded sentence based on a historical integrated language decoding score of the previous time step, a first language probability score of the expanded sentence and the corresponding electroencephalogram decoding score; determining a candidate sentence sequence of the current time step based on the integrated language decoding scores of the expanded sentences; performing sentence end detection based on the candidate sentence sequence of the current time step, and determining language content to be output in the current time step according to a detection result of the sentence end detection. The application improves the accuracy, intelligibility and naturalness of interaction of the output language content.

The application discloses a multi-speaker multi-language speech synthesis system based on self-learning text representation, self-learning multi-language text representation, and is embodied in two modules, namely a text-to-SMTR prediction module and an SMTR-to-multi-language acoustic spectrum prediction module. Specifically, the application comprises the following steps: constructing an SMTR extraction method based on a self-learning system; constructing a multi-language text-to-SMTR prediction method; constructing an SMTR-to-multi-language acoustic spectrum prediction method; and constructing an end-to-end multi-language speech synthesis method based on SMTR fusion. The application can improve the accuracy of multi-language speech synthesis.

This utility model relates to a shoulder joint trainer, including a positioning mechanism, a shoulder ladder assembly, and a gripping assembly. The positioning mechanism includes a first surface and a second surface along its thickness direction. Along the height direction of the positioning mechanism, the first surface has a plurality of fasteners arranged in an array, and the second surface is used to fit against a wall. The shoulder ladder assembly is detachably connected to the positioning mechanism. The shoulder ladder assembly includes a third surface and a fourth surface along its thickness direction, with the third surface opposite to the first surface. The third surface is provided with a connector adapted to the fasteners. When the fasteners and connectors are connected, the shoulder ladder assembly is fixed to the positioning mechanism. The gripping assembly is slidably connected to the shoulder ladder assembly. This utility model can connect the connectors to fasteners of different heights. For patients in the early stages of rehabilitation, the trainer can be adjusted to a lower position to easily complete basic lifting movements. As the recovery progresses, the height can be gradually increased to increase the training difficulty.

The application relates to the technical field of artificial intelligence, and discloses a speech synthesis method and device, computer equipment and a storage medium, which comprise the following steps: obtaining reference speech data and speech synthesis text; performing emotion extraction on the reference speech data to obtain reference emotion embedding; fusing the reference speech data, the speech synthesis text and preset noise features to obtain original fusion features; determining a dynamic emotion injection coefficient based on a preset denoising time step parameter; performing denoising processing on the original fusion features based on the dynamic emotion injection coefficient and the reference emotion embedding to obtain target synthesis spectral features; and performing audio conversion based on the target synthesis spectral features to obtain target speech data. The application can be applied to the speech synthesis scene of financial technology and medical health, and improves the emotion expression accuracy of speech synthesis.

This invention provides an AI-powered companion robot control system and method based on mobile communication devices, relating to the field of intelligent robot control technology. It collects multi-dimensional behavioral data of users using mobile communication devices and extracts behavioral feature vectors. Based on the behavioral feature vectors, it matches them with historical behavioral patterns to generate clone commands containing predicted actions and their confidence levels, emotional state values, and their confidence levels. The robot itself can execute the predicted actions based on these clone commands, providing proactive reminders and companionship to the user. Furthermore, by dynamically evaluating the computing power contribution level using the operating status parameters of the mobile communication device, it performs differentiated task scheduling, achieving the coordinated utilization of three levels of computing resources. This protects user privacy data, reduces the hardware cost of the robot itself, and organically unifies the high intelligence of cloud-based large-scale models with low-latency, secure local control, improving resource utilization, saving computing power, and achieving a three-in-one collaborative system.

The present application belongs to the technical field of image processing, and in particular to a multi-scale attention weak light image enhancement method and system. The present application comprises the following steps: step 1, constructing a network model: constructing a multi-scale attention weak light image enhancement network model, which comprises a decomposition network, a reflection image restoration network, and an illumination image enhancement network; step 2, preparing a data set: using the LOL-v1 data set to divide the training set and the test set and to perform preprocessing; using the LOL-v2 data set to fine-tune the model; step 3, training the network model: inputting the LOL-v1 data into the network for training until the preset training number or the loss function reaches the standard step; step 4, fine-tuning the model: using the second weak light image data set to further train and fine-tune the network model; and step 5, saving the model: solidifying and saving the final model parameters to realize the conversion output of weak light images into high-quality images. The present application designs an efficient multi-scale attention enhancement network for improving the performance of image features. This method can significantly improve the quality and clarity of the image, making the illumination and reflection details more distinct.

This invention discloses a real-time piano timbre simulation method and system based on machine learning, relating to the field of audio signal processing technology. The method includes: data acquisition and multi-dimensional annotation, acquiring multiple types of piano audio, covering techniques and seven dynamic levels, and simultaneously acquiring information such as key presses and techniques; audio preprocessing, including pre-emphasis compensation for high frequencies, Hanning window framing, Fourier transform to frequency domain, spectral subtraction for noise reduction and normalization; multi-dimensional feature extraction, extracting static features such as MFCC and spectral parameters, dynamic features such as first- and second-order differences, and overtone structures; two-stage model training, using stacked autoencoders for dimensionality reduction; real-time parsing, filtering and converting acquired performance data into parameter sequences; timbre synthesis, where the model generates a spectrum and performs an inverse Fourier transform into a waveform; and dynamic optimization, receiving user feedback. This invention solves the problems of traditional simulation methods; the two-stage model enhances timbre coherence, dynamic control achieves low latency, and multi-scenario adaptation and feedback optimization meet specific needs.

The present application relates to a kind of speech synthesis method and system based on bone conduction signal and lip image fusion, comprising the following steps: bone conduction signal, lip movement image signal are synchronously acquired when user speech input is collected;Determine the single-mode data characteristics of time domain and spatial domain based on bone conduction signal, lip movement image signal;Based on the two-source single-mode data characteristics of time domain and spatial domain determined, apply the generative adversarial network of cross-modal attention mechanism and mel-spectrogram fusion method, establish speech model, obtain modal collaborative feature expression;Based on the modal collaborative feature expression obtained, it can be recognized as specific phrase and instruction output by neural network model, and speech synthesis is realized using vocal synthesis model.The above algorithm realizes the commonality of modal collaborative representation, makes up the representation defect problem of single-mode independent existence, optimizes the effect of speech synthesis under high noise interference or mute mode, so as to expand the realizability of speech interaction.

This application discloses a method, device, equipment, and storage medium for displaying vehicle HUD information. It collects multi-source data including voice interaction, vehicle environment, and driver status, and performs deep processing of voice commands based on an improved Context-Former model to achieve full-process recognition of command status and association with multi-turn dialogue context. Based on this, it comprehensively decides and generates optimal display parameters by combining the current driving scenario, command priority, and driver's real-time status, thereby achieving accurate projection of interactive information. This application achieves dynamic matching of display strategies with driving scenarios, command importance, and driver attention through multi-dimensional collaborative adaptation, effectively reducing information interference. While ensuring clear communication of key information, it significantly reduces the cognitive load on the driver, improving the naturalness and safety of human-computer interaction. The technical solution of this application can be widely applied in the field of vehicle technology.

This invention discloses a virtual human-computer interaction system and method for brand promotion, relating to the field of human-computer interaction technology. The system includes: S1: collecting multimodal data from users during the interaction process via a user terminal and performing real-time analysis to generate real-time user status information; S2: generating comprehensive interaction context instructions; S3: based on the comprehensive interaction context instructions, executing a virtual human adaptive generation step and a brand content dynamic generation step in parallel; S4: forming and outputting an interaction response for the current user; S5: collecting user behavior data related to the interaction effect and updating and optimizing rules based on the user behavior data. The advantages of this invention are: through multimodal perception and context analysis, it achieves real-time personalized dynamic generation of virtual human image, interaction style, and promotional content, enabling brand promotion to accurately adapt to the real-time status and historical preferences of different users, greatly improving the naturalness, attractiveness, and user resonance of the interaction.

The application discloses a kind of based on deep learning's Mars exploration panoramic image generation method, it is related to image processing technical field.The method comprises the following steps: obtaining Mars exploration image dataset and constructing degradation pool;Multi-branch pyramid feature extraction network and the module based on the transformation of multi-scale homography estimation are trained;IAR-;Net network model comprising image alignment module and image reconstruction module is constructed;the panoramic image is generated by training network model.The application designs the unsupervised image splicing framework of fusion image alignment and depth reconstruction, improves the geometric alignment accuracy between images by multi-scale feature extraction and cross-image attention fusion mechanism, and realizes seamless fusion of image by combining primary reconstruction network and advanced reconstruction network, can effectively improve the registration accuracy and reconstruction quality of Mars exploration image in panoramic generation, solve the problem of poor generation effect caused by image texture sparsity, large view angle difference and lack of labeled data under complex environment of Mars.

This invention provides a badminton serving machine control method and system based on visual perception, posture recognition, trajectory prediction, and adaptive decision-making, belonging to the field of serving control. It addresses the problems of limited serving patterns, poor equipment perception capabilities, and ineffective training due to a lack of human-like game strategy. This invention predicts the badminton flight trajectory and landing point based on a rigid body aerodynamics model; extracts skeletal joints based on a geometric constraint loss function, calculates the athlete's posture state, and constructs a dynamic defensive potential energy field representing the range of attention coverage; generates adaptive control decisions including the target landing point and serving pattern based on expert knowledge rule-driven and potential energy field extremum optimization methods; and drives the actuator to complete the serve through inverse kinematics. This invention can perceive the athlete and ball state in real time, simulate real-person game thinking, and adaptively serve to target the athlete's defensive weaknesses, significantly improving the intelligence level and practical effectiveness of badminton training.

This application provides a retail shopping guide robot and an interactive control method for the robot. The method includes: acquiring a first image and a user's first interactive voice; inputting the first interactive voice into a shopping guide language model to obtain the user's purchase intent information and determine the user's shopping guide service type; if the shopping guide service type is product guidance, then generating an interactive action strategy based on the first interactive voice using an emotion strategy library; determining the target product information based on the purchase intent information; inputting the first image into a cascaded visual model to obtain the product placement state and position; generating a movement and grasping control strategy for the shopping guide robot based on the product placement state and position; generating an action interaction command to grasp the target product based on the movement and grasping control strategy and the interactive action strategy; and controlling the shopping guide robot to interact with the user according to the action interaction command. This can enhance the user's shopping experience.

The utility model discloses a continuous type rubbing machine, including the box, the inside rotation of box is equipped with rubbing cylinder, is provided with the driving force device of rubbing cylinder rotation on the box, and the circumferential of rubbing cylinder cylinder wall is equipped with a plurality of air inlets, the inner surface of rubbing cylinder is evenly distributed with a plurality of baffle strips, be equipped with at least two hot -blast circulating unit on the box, hot -blast circulating unit includes the heating device of setting at the hot -blast exchange mouth of box upper portion, the upper portion of box is provided with the hot -blast circulating fan of heating device inside intercommunication, the inside of box is provided with the heating oven of rubbing cylinder's outside, is provided with a plurality of hot -blast air -blast pipe of facing rubbing cylinder on heating oven, is equipped with a plurality of air outlet pipes on hot -blast air -blast pipe, and air outlet pipe is inclined to the place of being close to hot -blast air -blast pipe along the direction of raw material discharge.

The application provides a robot visual operation control method and system based on eye tracking, and belongs to the field of robot vision and artificial intelligence. The method comprises the following steps: a head-mounted device with eye tracking function is used to synchronously collect data streams when an operator performs an operation task, wherein the collected data streams comprise eye movement data streams and scene visual data streams when the operator performs the operation task; for the preprocessed data streams, a dynamic Bayesian network is used to extract gaze-intention coupling features; the preprocessed data streams and the gaze-intention coupling features are input into a robot operation model with a double-path attention fusion that has been constructed, so as to obtain online attention results, and the robot performs corresponding operations according to the online attention results. The method of the application can quickly focus on key areas in complex scenes such as occlusion and interference, and improves the task success rate.

The application provides a humanoid robot gait naturalness intelligent evaluation training method and device, comprising: acquiring corresponding reference force exertion timing records from a historical force exertion timing database for the working condition label, and determining the feasible range of force exertion timing; when the step speed variation amplitude exceeds a preset safety threshold and the friction resistance level is lower than a preset stability level, adjusting the force exertion timing delay; updating the initial force exertion timing using the adjusted force exertion timing delay, and iteratively comparing and correcting the feasible range of the force exertion timing to obtain an optimal take-off force exertion time; generating a force exertion timing dynamic adjustment scheme according to the comparison between the optimal take-off force exertion time and the metatarsal flexion torque peak time; and performing propulsion power and slipping risk evaluation on the force exertion timing dynamic adjustment scheme, and outputting the verified force exertion timing scheme after confirming that the requirements are met.

The application discloses a digital role animation control method, a modeling method, equipment and a storage medium. The method comprises the following steps: acquiring an animation switching instruction corresponding to a digital role, the animation switching instruction comprising a target bone parameter vector corresponding to a target bone node; processing the target bone parameter vector and a current bone parameter vector based on a parameterized bone model to determine a bone transformation matrix; transforming a current vertex attribute corresponding to a target mesh vertex based on the bone transformation matrix and a first bone skinning weight to determine a target vertex attribute corresponding to the target mesh vertex; transforming a current Gaussian attribute corresponding to a target Gaussian point based on the bone transformation matrix and a second bone skinning weight corresponding to the target Gaussian point to determine a target Gaussian attribute corresponding to the target Gaussian point; and determining a target animation image based on the target vertex attribute corresponding to the target mesh vertex and the target Gaussian attribute corresponding to the target Gaussian point. The method can guarantee the synchronization and high detail fidelity of action control.

This disclosure relates to the field of computer technology, and more particularly to a control method and apparatus for an embodied robot. The method includes: acquiring audio and video data collected by a collection device via a dedicated synchronous communication protocol for the embodied robot; processing the audio and video data and environmental perception data using a multimodal large model to obtain a data sequence; identifying and processing the data sequence using a mapping model to obtain control commands corresponding to the embodied robot; executing control operations corresponding to the control commands; and adjusting the software and hardware parameters of the embodied robot using the control commands. Using this disclosure can improve the naturalness, stability, and security of audio and video interaction in embodied robots.