113 results about "Acoustic spectrum" patented technology

The invention relates to the technical field of gas sensing detection, and discloses a gas detection method and system based on a quartz tuning fork enhanced photoacoustic spectrum. The method comprises the following steps: constructing a standardized continuous tuning fork signal data stream; dynamically segmenting the data stream based on the gas concentration interval, and extracting time domain and frequency domain features in parallel in each segment to form an initial feature vector; dynamically adjusting an anomaly judgment threshold according to the jump amplitude of the adjacent segment feature vectors, and scanning and screening out an effective analysis interval of a stable concentration state according to the anomaly judgment threshold; performing cross-domain coupling compensation on the feature vectors in the effective interval to generate enhanced feature vectors; and finally, performing reverse traceability reasoning through a gas leakage propagation model based on the enhanced feature vector, and outputting a leakage source position identifier. The device can adapt to a dynamically changing concentration field, and the reliability and traceability positioning capability of gas detection in a complex environment are improved.

The invention discloses a dynamic encryption transmission method and system for sound-light alarm signals, and relates to the field of data encryption, and the method comprises the steps: obtaining a dynamic environment entropy value; generating a pseudo-random chaotic sequence; obtaining a helium-oxygen mixed gas density parameter in the dynamic environment entropy value; segmenting a preset standard alarm waveform into a plurality of signal waveforms in a time domain, and generating a distorted signal waveform according to the helium-oxygen mixed gas density parameter; performing encryption modulation on the distorted signal waveform to generate an encrypted distorted signal waveform; determining a sending sequence and a sending interval of the encrypted distortion signal waveform by combining a pseudo-random chaos sequence based on an impedance fluctuation value and background acoustic spectrum characteristics in the dynamic environment entropy; generating a dynamic modulation signal based on the transmission sequence and the transmission interval; the dynamic modulation signal is sent to a water surface console through an umbilical cable; and receiving the guide pulse, performing phase conjugation processing on the guide pulse, generating an echo verification signal, and transmitting the echo verification signal back to the water surface console.

The invention belongs to the technical field of geological analysis, and discloses a ground subsidence analysis method and device, and the method comprises the steps: matching a sensor layout strategy corresponding to a foundation subsidence risk and meteorological data of a monitoring region; acquiring land subsidence data of the monitoring area according to a sensor; performing time-frequency analysis on the land subsidence data to obtain a geoacoustic spectrogram corresponding to the land subsidence data; inputting the geoacoustic spectrogram into a preset neural network, and outputting a settlement semantic vector corresponding to the geoacoustic spectrogram; acquiring and analyzing historical land subsidence data, and constructing a subsidence rule corresponding to the historical land subsidence data; and inputting the settlement semantic vector and the settlement rule into a preset space-time analysis model, and predicting to obtain a future settlement trend of the monitored area. By using the method disclosed by the invention, the problem that all risks in a certain area cannot be checked when geological data is monitored based on local monitoring points can be solved, and the problem that the geological risk analysis efficiency is relatively low can be solved.

The invention relates to the field of photovoltaic equipment abnormity early warning, and provides a photovoltaic equipment temperature abnormity prediction method and device based on a digital twinborn model, and the method comprises the steps: obtaining an environment acoustic signal, meteorological data, and the actually measured temperature data of photovoltaic equipment; performing acoustic spectrum feature extraction according to the environmental acoustic signal to obtain a virtual environmental parameter set; sending the meteorological data and the virtual environment parameter set to a preset digital twinborn model for processing to obtain an initial temperature predicted value set; determining a temperature prediction value distribution diagram according to the initial temperature prediction value set and actually measured temperature data of the photovoltaic equipment, and comparing a temperature prediction result with an abnormal temperature threshold value to obtain a temperature abnormal risk point; and processing according to the temperature prediction value distribution diagram, the temperature anomaly risk points and the meteorological data to obtain temperature anomaly early warning information of the photovoltaic equipment, and improving the accuracy and foresight of temperature anomaly prediction of the photovoltaic equipment.

The invention relates to the technical field of livestock and poultry product safety detection, in particular to a temperature-controllable crushing device for livestock and poultry product detection, which is characterized in that sound wave signals generated in a homogenizing process are acquired through a sound sensor arranged outside a detection cup, and spectrum drift and energy attenuation characteristics of the sound wave signals are analyzed by adopting a sound spectrum temperature inversion algorithm; inverting the temperature and distribution of the sample in real time; and meanwhile, a cooling air curtain is formed through nitrogen injection, the gas flow rate and the injection angle are intelligently adjusted through a control unit according to a temperature inversion result, directional focusing cooling of a local overheating area is achieved, and a closed-loop temperature control system driven by acoustic signals is formed. According to the device and the method, cross contamination and sensor adsorption effect caused by contact type temperature measurement are thoroughly avoided, the problems that the sensor is easy to pollute and difficult to maintain in a high-humidity and fat-rich environment are solved, heat-sensitive veterinary drug decomposition caused by friction temperature rise in the homogenizing process is effectively inhibited, and the accuracy, reliability and repeatability of veterinary drug residue detection are remarkably improved.

The invention relates to the technical field of unmanned aerial vehicle detection and tracking, in particular to a low-altitude unmanned aerial vehicle positioning and tracking method and system. The method comprises the following steps: acquiring sound spectrum time sequence data and radar plot data of an unmanned aerial vehicle; extracting a first sound spectrum characteristic parameter and a second sound spectrum characteristic parameter; extracting a first radar characteristic parameter and a second radar characteristic parameter; determining initial position estimation of the unmanned aerial vehicle through a first judgment condition; determining optimized position estimation of the unmanned aerial vehicle through a second judgment condition; and tracking the unmanned aerial vehicle by using an improved particle filtering algorithm to obtain a final position and a tracking trajectory of the unmanned aerial vehicle. According to the invention, through heterogeneous sensor data fusion, environment adaptive feature extraction, multistage intelligent judgment optimization and improved particle filter tracking, the precision, robustness and real-time performance of unmanned aerial vehicle detection in a complex environment are effectively improved, and the regional monitoring and countering capability is significantly enhanced.

The invention relates to the technical field of fire-fighting management and control, and particularly discloses a water level detection and fire-fighting identification method and system based on automatic control, and the method comprises the steps: obtaining the temperature and acoustic multi-modal data of a fire-fighting pipe network, building a pipeline temperature gradient model, and predicting the temperature distribution of each section through combining the environment and the medium temperature; and extracting ice crystal nucleation and fluid anomaly characteristics at low temperature. Based on the temperature gradient, the acoustic spectrum, the water level fluctuation rate and historical data, a multi-parameter fusion congelation risk assessment model is constructed, and risk indexes are calculated and graded; constructing a congelation risk and firelight early warning linkage control strategy, and adaptively adjusting a firelight detector threshold value and a sampling frequency according to a risk level; when the risk reaches an early warning value, partitioned heating is implemented, a temperature recovery curve and water level stability are collected in real time as feedback, model iterative optimization and closed-loop control are achieved, early warning accuracy is improved, energy consumption is reduced, and reliable operation of a fire fighting system is guaranteed.

The invention discloses a defibrillation decision-making method, device and equipment and a computer readable storage medium, and the defibrillation decision-making method comprises the steps: obtaining an electrocardiosignal, and carrying out the filtering and denoising of the electrocardiosignal, and obtaining an electrocardiosignal sequence; performing frequency domain analysis on the obtained electrocardiosignal sequence, and calculating frequency domain characteristic indexes including a sound spectrum flux, an optimized amplitude spectrum area and a sum of average amplitude difference functions; and according to the calculated frequency domain characteristic indexes, a comprehensive decision-making mechanism is adopted to carry out comprehensive decision-making to obtain a decision-making result of a defibrillation state, so that the problems that the defibrillation opportunity is difficult to grasp and the defibrillation effectiveness is limited are solved.

This invention provides a quantum-derived Newton-Raphson optimal fractional-order spectrogram generation method, comprising the following steps: Step 1: Acquire the original audio signal and construct a fractional-order spectrogram based on fractional Fourier transform (FRFT); Step 2: Perform nonlinear scaling compression on the fractional-order spectrogram using a Mel filter bank to generate a fractional-order Mel spectrogram; Step 3: Construct an adaptive optimization framework with information entropy minimization as the objective function to measure the information fidelity between the spectrogram and the original signal; Step 4: Use the quantum-derived Newton-Raphson optimization algorithm (QNRBO) to globally optimize the fractional-order order, frame length, and frame shift hyperparameters to generate the optimal fractional-order spectrogram; Step 5: Input the optimal fractional-order spectrogram into a downstream speech recognition model. This technical solution aims to systematically solve core problems such as insufficient traditional time-frequency representation capabilities, rigid hyperparameter configuration, limited optimization algorithm performance, and feature-task disconnect.

Embodiments of the present disclosure provide a singing evaluation method and device, a storage medium and a computing device. The method comprises: performing feature extraction on obtained singing audio to obtain a sound spectrum feature of the singing audio; inputting the sound spectrum feature into a multi-scale network model for calculation; wherein the multi-scale network model comprises a model based on a convolutional neural network, and a convolutional layer in the multi-scale network model comprises convolutional kernels corresponding to multiple time scales; and determining a final evaluation result based on a probability distribution of different evaluation results output by the multi-scale network model.

The present application relates to the technical field of artificial intelligence, and more particularly to an AI interaction system for ornithological scientific research, which comprises a collection unit, a trigger determination unit, a path determination unit, a dominant determination unit, a prompt generation unit and an adjustment unit. The present application constructs a multi-level behavior determination process through joint analysis of bird song spectrum, image motion characteristics, flight trajectory and environmental factors, and through statistical analysis of the time distribution and spatial distribution of these behavior indicators within a preset period, and correlation analysis with environmental factors such as temperature, humidity, wind speed and illumination, the influence of environmental changes on bird activity frequency and spatial aggregation degree can be identified, so that the acoustic trigger threshold can be dynamically adjusted, so that the system can continuously maintain the sensitive recognition ability of bird activity with the change of the habitat environment, and then realize high-reliable behavior recognition and interaction prompt for ornithological scientific research.

The invention discloses a method, a system, equipment, a medium and a program for intelligently diagnosing faults of multiple parts of a wind power generator cabin, and belongs to the technical field of wind power generation. The method comprises the following steps: collecting a multi-source acoustic signal in a cabin of the wind power generator, preprocessing the multi-source acoustic signal to obtain a preprocessed voiceprint signal, and generating a spectrogram; performing cross-domain feature extraction on the pre-processed voiceprint signal, performing feature dimension reduction and redundancy elimination through principal component analysis, and constructing a voiceprint feature vector; inputting the spectrogram and the voiceprint feature vector into a CNN-LSTM fusion diagnosis model, and performing fault diagnosis to obtain a fault diagnosis result; and matching the fault diagnosis result with the fault-voiceprint knowledge graph to generate a decision suggestion. According to the method, through non-contact voiceprint monitoring and multi-dimensional analysis, high-precision and real-time diagnosis and active decision-making of concurrent faults of multiple components of the wind power generator cabin are realized, and the defects of complex deployment and sound signal interference resistance of traditional vibration monitoring are effectively overcome.

This invention relates to the fields of animal behavior, animal welfare, and mental health conditioning, specifically a music conditioning system and method for pets and laboratory animals. It is a pure PC-based software system, including an audio editing and playback control module, a music library management module, and a music playback module. The output of the audio editing and playback control module is connected to the input of the music playback module, and the music library management module and the audio editing and playback control module have a bidirectional data connection. This invention precisely matches different animals through three parameters: pitch shifting, speed adjustment, and volume adjustment, significantly improving audio compatibility. The effects of emotional stabilization, anxiety relief, and sleep promotion are significantly better than general music. One-click templates are suitable for quick home use. Customizable visual editing is suitable for pet hospitals, laboratory animal centers, and research institutions to establish standardized programs. Waveform graphs, spectrograms, and acoustic spectrograms are displayed synchronously in real time, allowing users to intuitively judge the audio structure. Adjustment accuracy is improved by more than 80%, requiring no professional audio knowledge.

The application relates to the technical field of biomedical engineering, and particularly discloses a closed-loop OSA stimulation control method and device based on multi-modal physiological signal fusion, which comprises a multi-modal physiological signal acquisition module, a control processing module, a stimulation output module and a feedback correction module; by fusing related physiological signals such as respiratory muscle electricity, blood oxygen saturation and snore sound acoustic spectrum characteristics, the blocking risk in the sleep process is evaluated in real time, and the corresponding stimulation time, stimulation channel and stimulation parameter are determined in combination with the breathing phase information and the feedback result after stimulation, so that the precise, real-time and individualized closed-loop control for OSA intervention is realized under the premise of meeting the preset safety constraint. The application can more comprehensively reflect the respiratory state and the upper airway blocking risk in the acquired data, improve the accuracy of OSA related event identification, and determine the stimulation triggering time according to the current respiratory state, thereby improving the matching degree between the stimulation intervention and the actual respiratory demand.

The present application belongs to the technical field of compressor abnormal sound detection, and particularly relates to a compressor abnormal sound detection method and system based on acoustic spectrum pattern recognition, which comprises the following steps: extracting a characteristic signal from a sound signal through wavelet packet transform and kurtosis criterion, locating the occurrence time of an impact event contained in the characteristic signal and calculating the time interval of adjacent impact events, calculating a rhythm stability index based on all time intervals in an analysis time window, intercepting a time domain signal window containing each impact event and calculating its energy envelope, calculating the intrinsic damping factor of each energy envelope according to the logarithmic difference between the energy envelope peak value and the energy value after a set time length, calculating the fault authenticity score of each impact event according to the energy value of the impact event in the characteristic signal, the rhythm stability index corresponding to the analysis time window and the intrinsic damping factor, and performing early warning determination, so as to realize high-reliability and accurate early warning of compressor early faults.

The invention provides a sound spectrum nondestructive testing device for an austenite manganese casting for ship equipment. The sound spectrum nondestructive testing device comprises a base, supporting legs are fixedly mounted at the bottom of the base, a control box is fixedly mounted at the upper end of the base, a placement plate is mounted and connected to the upper end of the base, and two sliding plates are slidably connected to the upper end of the base; a plurality of hydraulic rods are fixedly installed at the bottom of the base, and pulleys are fixedly installed at the bottoms of the hydraulic rods and provided with brake pads. According to the acoustic spectrum nondestructive testing device for the austenite manganese casting for the ship equipment, by arranging structures such as the hydraulic rod, the hydraulic rod realizes height adjustment and flexible movement of the device, and stably supports the equipment, so that the acoustic spectrum nondestructive testing device adapts to a complex detection environment of a ship; the sliding block is matched with the sliding rod to achieve high-precision transverse movement, it is guaranteed that the probe is accurately positioned, and castings in complex shapes are efficiently covered. The second motor drives the placing plate to rotate at a constant speed, diversified detection modes are provided, no dead angle of sound wave detection is guaranteed, the detection efficiency and accuracy are improved, and the durability of equipment is enhanced.

The invention discloses a voiceprint correction method for a wind turbine generator and a related device. The voiceprint correction method comprises the following steps: acquiring operation parameters of the wind turbine generator; calculating a fog drop scattering coefficient, a corrected sound velocity and a corrected sound path distance difference according to the operation parameters of the wind turbine generator; dynamically reconstructing a sound spectrum according to the fog drop scattering coefficient, the corrected sound velocity and the corrected sound path distance difference; and obtaining the corrected voiceprint according to the dynamically reconstructed sound spectrum. The method and the related device can accurately detect the voiceprint signal of the wind turbine generator.

The invention discloses a hierarchical strategy-planner hybrid architecture for audio-visual navigation (AV-Nav) of a body-equipped agent, and aims to solve the challenge of tight coupling of multi-modal reasoning and precise action control in a traditional end-to-end model. According to the high-level strategy, a Transform-based network is adopted, deep fusion is carried out on real-time visual and acoustic features through a token and self-attention mechanism, and a navigation target point on a 3 * 3 local grid is output. The local target is then passed to a low-level classical planner that calculates the shortest path on the internally maintained dynamic navigation map using the Dijkstra algorithm and determines the atomic action to be performed. The method is characterized in that a planner is integrated with an interactive trial and error obstacle avoidance mechanism: once a collision signal Ct is received from the environment, the planner can immediately remove a corresponding blocked path edge from a navigation map, so that a system is forced to automatically re-plan a collision-free path in the next time step, and the navigation efficiency and robustness are remarkably improved. In addition, the architecture may perform end-to-end training in conjunction with novel audio enhancement strategies, including interfering sound sources, mixing or spectrogram masks.

The application discloses a grating sensor for photoacoustic imaging, a photoacoustic probe and a photoacoustic imaging device, and the grating structure is processed by adopting a three-dimensional photoetching process, the design of the external solid structure and the internal periodic structure is combined, the external solid structure and the multiple cavity regions form a significant refractive index difference, and therefore the resonance effect of the grating is enhanced. In the ultrasonic detection process, the periodic resonance effect caused by the refractive index difference of the grating structure can efficiently convert the ultrasonic wave pressure into a light signal and further convert the light signal into an electric signal for photoacoustic imaging. Compared with the conventional ultrasonic sensor, the application provides higher sensitivity and a wider bandwidth, can realize clearer and more accurate imaging effect in high-resolution imaging, and effectively avoids the phenomenon that the ultrasonic sensor in the prior art has poor imaging effect in high-resolution imaging. In addition, the two-photon processing and ultraviolet light curing are adopted to form a significant refractive index difference, and therefore the acoustic spectrum detection capability of photoacoustic endoscopic imaging is improved.

The application discloses a self-adaptive miniature circuit breaker switch state indication system, comprising: a mechanical structure unit; a sensor network unit; a data acquisition and processing unit connected with the sensor network unit, used for collecting, synchronizing and preprocessing sensor data; an algorithm processing unit; a fusion decision unit connected with the algorithm processing unit, used for integrating multi-algorithm output results, calculating a final compensation value, and dynamically adjusting algorithm weights according to environmental conditions; and an execution control unit connected with the fusion decision unit, used for adjusting the position of a state indication mechanism according to the output of the fusion decision unit. The application realizes ultra-high precision, full environmental adaptability and intelligent self-optimization capability of switch state indication through deep fusion of a mechanical deviation correction algorithm, a nonlinear thermodynamic compensation neural network algorithm and a quantum acoustic spectrum vibration suppression algorithm.

The invention discloses an extremely simple auxiliary method and system for judging an electrical fault based on an arc acoustic spectrum, and belongs to the technical field of electrical fire investigation and system evidence collection. Electrical abnormal events are accurately positioned through monitoring video audio track data processing and abnormal event windowing; according to the calculation of the energy sudden increase ratio, the peak count value and the high-frequency proportion increase amount, the calculation process of the electrical abnormal event is simplified, and the calculation efficiency is improved; according to the method, the index threshold value is calculated, simple result judgment is carried out through the index threshold value, various data in the calculation process are recorded, an auxiliary judgment evidence packet is formed, evidence storage traceability of the calculation process is achieved, and due to the simplified calculation process, the requirement for calculation power resources is greatly reduced, and high adaptability is achieved in the actual application scene.

Aspects of the subject technology relate to systems, methods, and computer readable media for estimating acoustic spectra. Acoustic data can be received at a hydrophone array from a first acoustic source and a second acoustic source in a downhole environment. An initial noise spatial correlation matrix estimation can be generated based on the acoustic data. The initial noise spatial correlation matrix estimation can be applied to a beamformer to generate a first source spectra estimation for the first acoustic source and the second acoustic source. A revised noise spatial correlation matrix estimation can be generated based on the first source spectra estimation. The revised noise spatial correlation matrix estimation can be applied to the beamformer to generate a second source spectra estimation for the first acoustic source and the second acoustic source in the downhole environment based on the first source spectra estimation.

The invention provides a coal quality online detection system and method based on multi-modal spectroscopy information, and the system comprises a TRLIBS module which is used for collecting a time-resolved laser-induced breakdown spectrum; the near infrared spectrum module is used for collecting a near infrared diffuse reflection spectrum of the coal sample; the sound spectrum acquisition module is used for acquiring a sound wave signal generated by the laser-induced plasma; the time sequence control module is used for synchronously controlling the acquisition time sequence of the TRLIBS module, the near infrared spectrum module and the sound spectrum acquisition module; and the data fusion processing module is used for carrying out feature extraction and fusion calculation on the received time-resolved laser-induced breakdown spectrum, near-infrared diffuse reflection spectrum and sound wave signals, and outputting coal quality indexes. Aiming at the problem of spectrum fluctuation caused by insufficient detection accuracy of a single spectrum technology and coal flow surface fluctuation in coal quality detection, the method provided by the invention improves the coal quality detection accuracy by integrating atomic spectroscopy, molecular spectroscopy and sound spectrum information and combining a synchronous time sequence control and intelligent fusion method.

The invention belongs to the technical field of artificial intelligence and medical speech analysis, and discloses a voice state intelligent classification method based on speech spectrum features and a reinforcement learning optimization mechanism. The method comprises the following steps: carrying out data preprocessing, acoustic feature extraction and feature optimization on a tested voice sample, and mapping the preprocessed tested voice sample into a two-dimensional Mel spectrogram; realizing voice state classification by using a multi-scale feature extraction structure comprising a first convolution branch, a second convolution branch and a third convolution branch and a bidirectional time sequence feature learning structure; meanwhile, a reinforcement learning optimization mechanism is introduced, feature selection, model structure configuration, training hyper-parameters and an updating strategy are subjected to self-adaptive optimization, confidence coefficient calibration, uncertainty judgment and interpretability result generation are combined, and a final voice state judgment result, calibrated confidence coefficient and an acoustic attention area are output. The voice state classification accuracy, stability and interpretability can be improved.

Systems and methods for treating misophonia include utilizing machine learning within a deep learning processor to allow a user to listen to ambient sounds from their environment without hearing trigger sounds. The method includes the steps of recording ambient sounds with one or more microphones, digitizing the recorded ambient sounds into digital signals, creating spectrographic data for the digital signals, comparing the spectrographic data against a signature library that comprises preprogrammed spectrographic data for the unwanted trigger sounds, identifying the spectrographic data that corresponds to the unwanted trigger sounds, removing the unwanted trigger sounds from the spectrographic data to provide filtered spectrographic data, converting the filtered spectrographic data into a filtered digital signal, converting the filtered digital signal into a filtered audio signal that does not include the unwanted trigger sounds, and playing the filtered audio signal to the user through the one or more speakers on the headset.

An audio generation method, an audio generation device, and a storage medium are provided. The method includes: receiving an audio generation instruction input by a user, wherein the audio generation instruction is used to indicate a two-dimensional image that the user wants to embed into generated target audio; obtaining a target grayscale image of the two-dimensional image in response to the audio generation instruction; converting grayscale data of each pixel in the target grayscale image into frequency-domain data of each pixel in a spectrogram, to obtain a target spectrogram; and generating target audio corresponding to the target spectrogram by using the target spectrogram.

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 present application relates to a kind of sound leakage detection method, system, equipment and storage medium of secret meeting room, its method includes obtaining image data and the visual angle range data corresponding to image data;According to image data, corresponding sound data is called, sound data includes sound spectrum data and sound imaging data, sound spectrum data is the sound spectrum diagram of sound, sound imaging data can reflect the image of sound source distribution position;Based on the matching rule of pre-established, according to visual angle range data and sound imaging data, the sound source position in visual angle range data is matched, and target sound source is marked in sound imaging data;According to the preset image analysis rule and sound imaging data, adjust sound spectrum data, obtain target sound spectrum data;Based on the preset sound spectrum transformation model, target sound spectrum data is inversely transformed, and the target audio data corresponding to image data is obtained.The present application realizes the effect that the sound condition in the limited range outside secret meeting room is monitored.

The application discloses a Tibetan speech recognition method fusing tone perception hybrid experts and retrieval error correction, and belongs to the technical field of signal processing in the electronic industry. The specific steps of the recognition method are as follows: I: Tibetan speech signals are acquired, and corresponding log-mel spectrogram features and fundamental frequency contour features are extracted; II: the tone gating weight is calculated based on the fundamental frequency contour features, and is dynamically routed to the corresponding hybrid expert network to acquire dialect-independent acoustic hidden layer features. The application effectively solves the model interference problem caused by the presence or absence of tones among multiple dialects, avoids the parameter conflict between tone dialects and non-tone dialects, significantly improves the recognition performance of multi-dialect hybrid training, corrects the homonym heterograph error commonly existing in Tibetan, improves the performance of multi-dialect Tibetan speech recognition, can be used for converting Tibetan speech into characters, and is helpful for protecting and mining Tibetan culture.