82results about How to "Improve naturalness" patented technology

The invention relates to a real-time low-energy-consumption unmanned aerial vehicle multifunctional automatic inspection method and system, and belongs to the field of automatic inspection. The method comprises the following steps: performing semantic field conversion based on a fuzzy geographic instruction, and outputting a geographic semantic field; performing vision-field coupling navigation based on a geographic semantic field, and outputting a real-time flight path and environment preliminary perception data; topological risk identification is carried out based on the real-time flight path and the environment preliminary perception data, and a 3D environment model with risk annotation is obtained; fractal energy consumption optimization is carried out based on the 3D environment model with the risk label, and an optimized waypoint sequence is output; and carrying out multi-modal data acquisition and real-time risk verification based on the optimized waypoint sequence. According to the invention, intelligent inspection of understanding natural language instructions, fusing multi-sensor information, dynamically evaluating risks and optimizing energy consumption is realized.

The application relates to an information processing method based on a large language model, a receiving method and related devices thereof, which comprises the following steps: obtaining an initial message to be processed, a target key and a current context; inputting the initial message to be processed into a target large language model, and obtaining a hash value corresponding to a current token in the initial message in a process of generating a token of a target text, and obtaining at least two candidate token sets according to the hash value; determining a target token set from the candidate token set according to the current token, and performing sampling control on candidate tokens in the target token set based on an original probability distribution of the current token to obtain a target token corresponding to a current generation position of the target text of the current token, and obtaining the target text. The scheme provided by the application can embed information by using a hash algorithm and a sampling control mechanism in the process of generating the text, and can improve the naturalness of the output text and the unbiasedness of the statistical distribution.

The application is suitable for the technical field of speech synthesis, and particularly relates to a speech synthesis method and device based on artificial intelligence, computer equipment and a medium. The application extracts a text feature vector of a target text through a feature extraction model, predicts the text feature vector through a stress predictor, outputs a stress prediction vector, adds the stress prediction vector to the text feature vector to obtain a text stress vector, predicts the text feature vector through a pause predictor, outputs a pause prediction vector, adds the pause prediction vector to the text feature vector to obtain a text pause vector, predicts the text stress vector and the text pause vector through a prosody predictor, outputs a text prosody vector, matches the text prosody vector with a phoneme sequence of the target text, obtains a phoneme sequence with prosody labels, performs speech conversion on the phoneme sequence with prosody labels, obtains synthesized speech, and through the prediction of stress, pause and prosody, the expressiveness, naturalness and accuracy of the synthesized speech are improved.

The application relates to an intelligent voice communication system and method for real-time interaction between a device end and a large language model cloud end. It solves the problems of high interaction delay, lack of true full-duplex voice stream support, separation of audio stream and instruction execution, and poor interaction efficiency in the prior art. It comprises a device end and a cloud end. The device end comprises a microcontroller running a FreeRTOS real-time operating system, the microcontroller is connected with a signal acquisition and processing unit, and the microcontroller is connected with an NFC controller for device end interaction and communicates data with the cloud end through a wireless communication module. The cloud end comprises a full-duplex voice gateway, a GPU array, an AI acceleration card and a device perception scheduler. The application has the advantages that natural continuous conversation of voice in and out is realized; user instructions are recognized and executed in real time during the conversation process, voice feedback is returned synchronously, and low-delay, low-power-consumption, high-concurrency full-duplex voice interaction is realized.

This invention relates to the field of robot interaction technology, specifically to a method for recognizing the intensity of human-computer interaction intentions in real time. The method includes the following steps: S1, acquiring video of pedestrians through a collection unit mounted on the robot, performing frame extraction on the video, and using a 3D skeleton extraction model to extract the skeleton information of each pedestrian in each frame in chronological order; S2, based on the skeleton information of each pedestrian in each frame, using a preset lightweight motion behavior recognition model to identify the interaction reference information of each pedestrian; S3, analyzing and processing the three types of interaction reference information—straight-line distance, orientation information, and motion behavior—to obtain the real-time interaction intention intensity of the pedestrian relative to the robot. This invention can accurately identify the interaction intentions of pedestrians, thereby improving the initiative and naturalness of social robots when interacting with humans.

The application provides a zero-shot song conversion method and device, equipment and storage medium, relates to the technical field of audio processing, and includes the following steps: obtaining a source song waveform and a reference song waveform; performing source condition extraction on the source song waveform to obtain a source song condition sequence; performing timbre coding on the reference song waveform to obtain a target timbre sequence; performing flow matching sampling on the source song condition sequence and the target timbre sequence by using a preset diffusion transformer to obtain a target continuous latent sequence; and decoding the target continuous latent sequence to obtain a target song waveform. The application can improve the similarity of song conversion in a zero-shot scenario.

The invention provides a multi-round dialogue intention classification method and device, and the method comprises the steps: defining and extracting a single-valued slot which is a core keyword contained in an FAQ intention; constructing intention clusters based on the single-valued slots, wherein the intention clusters of the same entity type and the intention clusters of the same single-valued slots are constructed; a single-valued slot inheriting slot mechanism is introduced into multiple rounds of dialogues, and different dynamic score adding strategies are performed on the intention clusters of the same entity type and the intention clusters with the same single-valued slot based on the recognition result that the hit intention has the single-valued slot; and performing intention classification again according to the intention score after score addition. The intention recognition accuracy and effectiveness can be improved.

The embodiment of the invention provides an intelligent film picture repairing method and device based on optical flow propagation. The method comprises the following steps: acquiring accumulated displacement information between a target frame and a source frame; the accumulated displacement information represents the corresponding position of each pixel in the target frame in the source frame; the accumulated displacement information is obtained according to a spatial transformation optical flow of each frame between the target frame and the source frame, and the spatial transformation optical flow is an optical flow obtained by taking the source frame as a spatial transformation reference; extracting pixel data from the source frame according to the accumulated displacement information; and taking the pixel data as a filling result of the target frame, and performing repair filling on the target frame to obtain a repair frame. The method is used for achieving the effect of improving the movie picture repairing quality.

The application discloses a kind of humanoid posture constraint under redundant mechanical arm inverse kinematics solving method, belong to robot control technical field;The method includes the following operations: S1, in offline training stage, construct humanoid posture model;Step S1 includes the following operations: S11, human body double-arm joint angle data extraction;S12, sine / cosine coding and standardization processing;S13, variational autoencoder humanoid prior training;S2, online solving stage, utilize the inverse kinematics solving of constructed humanoid posture model, in combination with particle swarm optimization algorithm PSO;The application is constructed by constructing humanoid posture model, and the implicit space distribution of humanoid posture is learned from human motion data using variational autoencoder, and human motion feature is quantified as the humanoid penalty that can be calculated, which provides the evaluation basis for the naturalness of the posture of the robot arm, solves the problem that the posture of the inverse kinematics solving result is not natural in the prior art, and the human motion data and the kinematic characteristics of the robot arm are not matched.

The present disclosure provides an image processing method, device, electronic equipment and storage medium, the method comprising: determining a feature image corresponding to a target spectral range in a first image; preprocessing a second image to obtain a preprocessed image; wherein the second image and the first image are images corresponding to the same scene; the first image has more spectral channels than the second image; and fusing the preprocessed image and the feature image to obtain a target image. The technical solution of the present disclosure can improve the natural effect of image beautification.

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 embodiment of the invention provides a voice signal enhancement method, electronic equipment and a storage medium. The method comprises the following steps: acquiring an input voice signal, and sequentially performing signal processing and frequency band division processing on the input voice signal to obtain a low-frequency band and a high-frequency band corresponding to the input voice signal, low-frequency band data corresponding to the low-frequency band and high-frequency band data corresponding to the high-frequency band; determining N frequency points adjacent to the high frequency band in the low frequency band as a first frequency band, and determining first frequency band data corresponding to the first frequency band in the low frequency band data; performing first enhancement processing on the low-frequency band data through a first enhancement network to obtain low-frequency enhanced data, and performing second enhancement processing on the first frequency band data and the high-frequency band data through a second enhancement network to obtain high-frequency enhanced data; smooth processing and splicing processing are performed according to the low-frequency enhancement data and the high-frequency enhancement data to determine the enhanced voice signal, so that the voice enhancement effect is improved.

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 invention discloses a sound adjusting method and device for an intelligent cabin, equipment and a storage medium, and the method comprises the steps: processing obtained data, and obtaining an in-vehicle member identity label, a current road scene label and a user behavior label; determining a first gain and a mode change state of a preset area in the vehicle according to the acquired data of the environment inside and outside the vehicle and the current sound output mode; performing mode judgment according to the in-vehicle member identity label and the user behavior label to obtain an initial sound output mode and a second gain; compensation is carried out according to the in-vehicle member identity, the initial sound output mode, the current road scene label and the in-vehicle and out-vehicle environment data, and a compensation gain is obtained; making a decision according to the current sound output mode, the first gain, the mode change state, the initial sound output mode, the second gain and the compensation gain to obtain a target sound adjustment scheme; and adjusting the in-vehicle sound output mode and the sound area volume according to the target sound adjustment scheme.

This application relates to a method for generating singing voice, a computer device, and a storage medium. The method includes: acquiring syllable information, note information, and tether information of a target musical score; determining the musical score phoneme encoding features, musical score phoneme duration, and phoneme-level fundamental frequency based on the syllable information, note information, and tether information; generating the musical score audio of the target musical score based on the musical score phoneme encoding features, musical score phoneme duration, and phoneme-level fundamental frequency; acquiring the frame-level fundamental frequency corresponding to the musical score audio and the musical score phoneme information corresponding to the target musical score; and generating the target singing voice audio based on the frame-level fundamental frequency and the musical score phoneme information. This method can determine musical score information by parsing the musical score, determine musical score features based on the musical score information, generate the musical score audio of the target musical score based on the musical score features, and generate a singing voice of the target musical score with a preset timbre based on the musical score audio and the phoneme information of the target musical score, thereby improving the naturalness of the synthesized singing voice.

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.

The invention relates to an intelligent replacement method for a short play video advertisement article, which relates to a video fusion technology, and comprises the following steps: carrying out multi-frame dynamic image acquisition and advertisement area three-dimensional space information extraction on a to-be-processed short play video, and establishing a physical attribute inversion and dynamic physical attribute field model; rendering parameter fitting and standardized input are carried out on the advertisement three-dimensional model according to physical parameters such as dynamic illumination, materials and roughness of an original scene, a three-dimensional space self-adaptive physical rendering engine is driven, and synchronous physical reconstruction of advertisement content and continuous frame scene illumination and material changes is achieved. In combination with multi-foreground-layer dynamic shielding and space superposition, in cooperation with the steps of boundary adaptive hybrid rendering, time sequence image quality optimization and the like, the advertisement content can be dynamically and accurately transformed according to the environment and the foreground, and finally high consistency and seamless natural fusion of the advertisement and the movie video content in physical attribute and visual perception are achieved.

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 invention belongs to the technical field of human body posture recognition, and relates to a human body posture estimation method using diffusion model reconstruction, which comprises the following steps: acquiring data sets, integrating the data sets, uniformly converting the data sets to an SMPL coordinate system to form a fusion input vector, and constructing local input vectors of a left ear and a right ear; embedding the local input vectors of the left ear and the right ear into a feature space of a specific dimension, and stacking to obtain embedded feature vectors of the left ear and the right ear; the embedded feature vectors of the left ear and the right ear are sequentially input into a time modeling module and a space modeling module to obtain output after interaction of two channels, then the output of the two channels is spliced and mapped into a unified space-time fusion feature, and a feature sequence is output after all time steps are stacked; and inputting the feature sequence into a de-noising diffusion probability model by adopting the de-noising diffusion probability model, and obtaining an output model after forward diffusion and backward diffusion. The frame can complete space-time modeling and reconstruction of human body postures only by depending on the two IMU sensors worn on the two ears.

The invention relates to an image processing method, and a color correction information calibration method, device and equipment, and belongs to the technical field of image processing. The method comprises the following steps: in response to a shooting instruction, determining each piece of actual response information of a camera module for a shooting object; determining a first target hue interval to which any piece of actual response information belongs from the candidate hue intervals; and performing color correction on the corresponding actual response information based on color correction information of the camera module for the first target hue interval under the reference light source to obtain target response information. Therefore, different color correction information can be adopted in different candidate tone intervals, color correction can be performed in different regions, the diversity of the color correction information is improved, the color correction precision is improved, for example, the overall color accuracy and naturalness are remarkably improved, the problems of color distortion and the like caused by using single color correction information can be avoided, and the color correction efficiency is improved. And the visual experience is optimized.

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.

The application relates to the technical field of intelligent control, and particularly discloses a self-adaptive control method for a trolley case based on holding force and motion trajectory identification. The method comprises the following steps: collecting a holding force time sequence signal and a motion trajectory time sequence signal, and constructing a holding force-motion trajectory coupling time sequence matrix; performing elastic matching on the holding force-motion trajectory coupling time sequence matrix and an intention mode library to obtain an initial intention identification result; adjusting the fusion weight of the coupling matrix according to a response deviation fed back by a motor; and generating a control instruction according to the identification result and the adjusted weight. The application reconstructs user intention identification into an elastic matching problem of a space-time coupling mode, solves the technical defects of high misjudgment rate and response delay caused by the inability of the prior art to distinguish between active force exertion and passive posture change, and realizes accurate prediction and self-adaptive assistance control of user intention.

The application provides a scheduling method and system for multi-digital avatar synchronous screen driving, wherein the method comprises: collecting real-time facial key point data and emotion feature vectors of a driving source, the facial key point data comprising key point coordinates, eye opening degree, mouth opening degree, FACS activation intensity, and the yaw angle, pitch angle and roll angle of the head posture; inputting the emotion feature vectors into an emotion-action mapping model to generate corresponding action control parameter sequences; distributing the action control parameter sequences to multiple digital avatars on the same screen through a timestamp alignment mechanism, so that each digital avatar completes synchronous update of expressions and actions within the same frame; and dynamically allocating scheduling priorities of each digital avatar to control end-to-end driving delay within a preset threshold. The application can greatly reduce the inter-frame synchronization error of multiple digital avatars, significantly improve the naturalness of digital avatar expression transition, and improve the visual continuity of multi-digital avatar synchronous screen driving.

This invention discloses a method, apparatus, device, and storage medium for generating adversarial audio examples. The method includes: acquiring text information and noise; inputting the text information into a random duration predictor to obtain the duration information of each text, aligning the duration information with the noise to obtain an alignment result; inputting the text information and alignment result into a prior encoder to encode intermediate features; inputting the intermediate features and noise into a conditional distribution generator, using noise as a condition, passing through multiple parallel long short-term memory networks and fully connected layers to obtain multiple frame-level distributions of the intermediate features under noise conditions; converting the distributions through a flow layer to an approximate distribution with the posterior distribution of the audio frames, and sampling latent variables; and decoding the latent variables to obtain the audio adversarial examples. This invention utilizes the principle of distribution sampling to conditionally add noise to a region inaudible to the human ear, making the generated audio adversarial examples more effective at deceiving the human ear.

The invention discloses a voice control method, system and device, a storage medium and a program product, and relates to the technical field of industrial equipment control and artificial intelligence crossing, and the method comprises the steps: obtaining voice data; performing text conversion and semantic analysis on the voice data through a pre-trained NLP semantic understanding model to obtain a semantic analysis result; generating an equipment control instruction in combination with the semantic analysis result and a preset verification mechanism, and sending the equipment control instruction to the target equipment; the state or data returned after the target equipment executes the equipment control instruction is received, and result feedback is generated based on the state or data, so that the limitation of a traditional fixed instruction can be broken through, a user is supported to interact with industrial equipment in a natural language, and the operation safety is confirmed and guaranteed through a preset verification mechanism in combination with semantic and instruction mapping. And industrial equipment interaction under multi-scene coverage is supported.

The invention discloses a digital human generation method and system combining acoustic driving and personality enhancement. The method comprises the following steps: carrying out multi-dimensional perception separation and semantic feature coding on an audio provided by a user by utilizing an autoregressive continuous action generation module based on acoustic driving, and further carrying out linkage generation of multi-modal face and trunk continuous actions through context association coding and decoding; a diffusion generation module based on structure strengthening is utilized to perform diffusion generation of the digital human video according to the generated face and trunk continuous actions, and pixel-by-pixel trajectory tracking and correction are performed to optimize the digital human video quality; and constructing a customized fusion scheme according to a multi-scene feature element set such as user audio, character, habitual language and the like by utilizing a character-based audio and picture style guiding module, and obtaining a digital human video with user characteristics. According to the method, the generation quality and authenticity of the digital human video can be improved from the aspects of semantic-action consistency, digital human image quality and the like.

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 invention discloses natural herbal shampoo with hair growing and strengthening functions and a preparation method of the natural herbal shampoo, and relates to the technical field of daily chemicals. Comprising the following raw materials in parts by weight: saponin, soapberry, tea bran powder, angelica sinensis, cacumen biotae, folium mori, glossy privet fruit, radix polygoni multiflori preparata, ginger, folium artemisiae argyi, moutan bark, eclipta, garden balsam stem, astragalus membranaceus, radix sophorae flavescentis, fructus cnidii, ligusticum wallichii, radix rehmanniae, fructus kochiae, formosan lattuce herb, mint, hibiscus leaves, rhizoma drynariae, white mulberry root-bark, golden cypress, cortex dictamni, lucid ganoderma and ginseng. According to the natural herbal shampoo with the hair growing and strengthening functions and the preparation method of the natural herbal shampoo, more than twenty Chinese herbal medicines including Chinese honeylocust fruits, Chinese soapberries, Chinese arborvitae twigs and leaves, radix polygoni multiflori preparata, ginseng and lucid ganoderma are carefully selected and compounded, so that all the raw materials have a synergistic effect on different aspects of decontamination, cleaning, dandruff removing, itching relieving, blood activating, nourishing, hair strengthening and strengthening and the like; therefore, the comprehensive hair growing and fixing effects of deep cleaning, scalp environment conditioning and hair follicle nutrition supply can be achieved in a single product at the same time.

The application discloses an underwater image enhancement method and system based on dynamic multi-channel compensation. The method comprises the following steps: acquiring an underwater image to be processed, limiting the exclusive stretching intensity of a red channel, performing linear dynamic range stretching on a red-green-blue three-channel based on high and low quantile points, performing defogging based on inverse operation of an atmospheric scattering model, performing color correction on a blue-biased image and a green-biased image by using a green-channel-dominant differential color compensation strategy through an iterative loop, performing mean centering fine adjustment on a and b channels in a Lab space, performing strengthening processing on a brightness channel in an HSV space, extracting feature details of different levels and performing edge detection, fusing the details and edge information with the original brightness channel, and obtaining an enhanced underwater image. The application effectively recovers the lost information of the red channel by adaptively processing the blue-biased or green-biased image, significantly improves the visual effect of the underwater image, and enhances the detail presentation, color accuracy and overall clarity of the image.