9999results about "Speech analysis" patented technology

The invention relates to the technical field of artificial intelligence, can be applied to business scenes of financial science and technology, medical health and the like, and discloses a decision-making method, device, equipment and medium based on multi-modal semantic alignment. Executing cross-modal alignment by taking the voice semantic map as a reference to generate associated information, fusing the voice features, the visual features, the action features and the associated information to generate a fusion feature vector, inputting a decision network to generate a decision feature vector and generate a task execution instruction, obtaining execution feedback information of the task execution instruction, and updating the decision network. According to the method, input is dominated by voice instructions, visual features, action features and semantic map depth alignment and fusion are combined, input naturalness and multi-modal data analysis and decision-making efficiency are improved, and interaction adaptability and decision-making accuracy of the model in a complex scene are improved.

The invention discloses a cross-cultural customer service dialogue quality automatic evaluation method in combination with sentiment analysis, and relates to the technical field of natural language processing, and the method comprises the steps: carrying out the alignment of voice and text based on a transmission matrix in real time, extracting a speech, a metaphor and polarity, and generating a speech tag; constructing an emotion channel and a polite channel, and fusing expression and shielding intensity through sharing attention; comparing and aligning with the same language prototype in a regional culture baseline library to obtain a calibration representation and updating a language offset record table; the potential upgrading probability is represented and recurred according to round aggregation calibration, and a risk vector and a high-risk position are formed; fusing risk and business indexes by a capacity integral kernel, outputting a comprehensive quality score, and giving factors and round attributions; sample recovery is triggered according to score and feedback difference, a micro-weight training data set is constructed, gradient increment training is carried out under low-rank adaptation, and cross-language consistency, early recognition of upgrading risks and interpretable evaluation are achieved through a closed loop.

The invention relates to the technical field of artificial intelligence, can be applied to business scenes of financial science and technology, medical treatment and health and the like, and discloses a violation content auditing method, device and equipment based on multi-modal data and a medium. Inputting the visual semantic features and the composite audio features into a multi-modal model, generating fusion features through model alignment and fusion, and analyzing the fusion features based on a knowledge base to judge whether illegal content fragments exist in the multi-modal data, and when the illegal content fragments exist, positioning the illegal content fragments in the multi-modal data and generating an auditing report. According to the method, the visual semantic features and the composite audio features are fused, cross-modal compliance analysis is realized in combination with the knowledge base, frame-level or time-axis-level positioning is performed on the illegal content segments, and the auditing report containing the evidence is generated, so that the problems of insufficient single-modal detection accuracy and poor positioning capability are solved, and the auditing accuracy is improved.

The invention provides a wind turbine generator voiceprint fault recognition method, and relates to the technical field of wind turbine generator state monitoring and fault diagnosis, and the method comprises the steps: carrying out the noise reduction of an original audio signal through variational mode decomposition, screening a target mode of which the frequency, energy and kurtosis accord with features, and reconstructing the signal; extracting a Mel frequency cepstrum coefficient and a sensing noise robust coefficient, and generating multi-dimensional voiceprint data in combination with statistical characteristics such as a frequency spectrum gravity center, a spectrum entropy, energy, kurtosis and a zero-crossing rate; constructing a support set based on the prototype network, realizing small sample fault classification by calculating the Euclidean distance between the feature vector and the prototype vector, and outputting a preliminary result; judging whether the voiceprint is abnormal according to a preset threshold value, if so, storing the voiceprint into a dynamic abnormal voiceprint knowledge base; frequently occurring abnormal samples are manually labeled and added into a support set, the prototype network is retrained to update the model, and continuous optimization of the fault recognition capability is achieved.

The embodiment of the invention provides a digital human interaction system and method based on multi-modal emotion recognition, and belongs to the technical field of digital human interaction. The system comprises a multi-modal sensing module used for collecting multi-modal data and preprocessing the multi-modal data to generate a standardized data stream; the cross-modal fusion and emotion recognition module is used for carrying out interactive modeling on the multi-modal features and outputting a current emotion label and emotion intensity; the reaction planning module is used for generating a composite reaction strategy; and the digital human rendering module is used for mapping the composite reaction strategy into control signals corresponding to the voice, the facial expression and the action respectively, and driving a digital human to execute corresponding voice output, facial expression change and limb action through the control signals so as to realize interaction. According to the method, multi-modal data are deeply fused through the cross-modal graph neural network and comparative learning, the weight is dynamically adjusted in combination with the modal confidence, and the emotion recognition accuracy and robustness are improved.

A method for training a machine learning (ML) model using a large language model (LLM) is provided. A system for detecting fraud which utilizes the LLM-trained ML model trained is also provided. An artificial intelligence (AI)-based method for monitoring alerts is also provided. The method for training an ML model using an LLM includes receiving tabular data for training the ML model, generating one or more natural-language strings comprising information from the tabular data, generating, via a base LLM, one or more prompts and completions based on the one or more generated natural-language strings, pre-training the base LLM using a plurality of generated prompts and completions, updating the base LLM via supervised learning using a cross-entropy loss function with ground-truth labels, and fine-tuning the updated LLM via reinforcement learning with human feedback using a reward model and a proximal policy optimization model to produce the LLM-trained ML model.

The invention relates to a microphone array sound source positioning method and system based on cross-correlation-beam forming closed-loop optimization, and belongs to the technical field of sound source positioning. The method comprises the following steps: collecting multichannel sound signals through a microphone array and preprocessing the multichannel sound signals to extract time-frequency features and suppress noise interference; time delay information among the microphones is estimated by adopting a generalized cross-correlation phase transformation algorithm, and an optimization strategy is introduced to improve estimation stability and anti-interference performance; enhancing the target sound source signal in combination with a minimum variance undistorted response beam forming algorithm and an adaptive Kalman filtering mechanism; constructing a closed-loop feedback optimization mechanism based on the beam output signal to realize feedback adjustment; and adopting a hybrid network architecture, taking the beam output signal amplitude spectrum as input, and outputting the frequency spectrum or mask of the obtained target sound source signal. The method has the advantages of high calculation efficiency, high positioning precision and strong anti-interference capability, and is suitable for real-time acoustic signal processing in a complex environment.

The invention discloses an equipment fault intelligent early warning system based on abnormal voiceprint AI analysis of energy equipment, and relates to the technical field of equipment health management, the equipment fault intelligent early warning system comprises an equipment fault early warning platform, and the equipment fault early warning platform is in communication connection with the following modules: a data sensing fusion module, a voiceprint AI analysis module and a voiceprint AI analysis module; the data acquisition module is used for acquiring high-frequency voiceprint signals, temperature field distribution and vibration data in real time during operation of energy equipment through a distributed sensor network to form a comprehensive data set; and the voiceprint AI analysis module is used for extracting voiceprint features from the comprehensive data set, and identifying whether the equipment emits abnormal voiceprints or not by using a pre-trained voiceprint AI model. According to the method, early abnormity is identified through the high-precision AI model, sudden equipment faults are effectively prevented, equipment physical field interaction is simulated in combination with the digital twin technology, and a fault evolution path is dynamically deduced, so that operation and maintenance personnel can take intervention measures at the initial stage of the faults, the stability and reliability of equipment operation are remarkably improved, and the non-planned downtime is shortened.

The invention discloses an enteritis patient condition severity risk early warning method, a medium and equipment. The method comprises the following steps: acquiring a borborygmus original signal and a clinical multi-dimensional physiological parameter sequence through a sensing device; constructing a borborygmus dynamic characteristic spectrum based on the borborygmus original signal to generate an acoustic biomarker time sequence; inputting the acoustic biomarker time sequence and the clinical multi-dimensional physiological parameter sequence into a multi-modal fusion early warning model to obtain an intestinal inflammation risk index; executing a signal quality self-evaluation process and generating a data quality warning code when the signal quality is abnormal; triggering a multi-node collaborative monitoring mechanism based on the risk index to generate an intestinal state multi-dimensional situation map; establishing an individualized risk baseline and generating a graded early warning instruction; and finally outputting a comprehensive early warning report. According to the method, multi-modal fusion analysis of the borborygmus signal and the clinical parameters is realized, the accuracy and timeliness of illness state early warning are remarkably improved through dynamic risk assessment and signal quality monitoring, and a reliable basis is provided for clinical decision making.

The invention discloses a sound acquisition and processing system based on cooperation of multiple microphone arrays. The system comprises a sound acquisition module, a multi-channel signal preprocessing module, a sound signal feature extraction module, an abnormal sound recognition module, a sound source positioning module and an alarm module. A multi-channel mixed data signal is collected through a circularly-arranged multi-microphone array formed by a plurality of microphones, after echo cancellation, wave beam domain noise reduction and multi-sound-source separation, a single-sound-source feature vector is extracted, according to the single-sound-source feature vector, abnormal sound including explosion, screaming or glass breakage is recognized through a BiLSTM and an attention mechanism model, and the abnormal sound is recognized through an attention mechanism model. And the GCC-PHAT and MDS-MUSIC algorithms are combined to position abnormal sound production, and alarm information is generated. According to the invention, accurate identification, positioning and alarm of the abnormal sound can be realized, and the real-time performance, the accuracy and the multi-target processing capability of abnormal sound monitoring in a complex environment can be obviously improved.

Provided are a method for training a transformer fault detection model, a fault diagnosis method, and a related device. The method includes: obtaining an initial voiceprint signal of a transformer and a fault type corresponding to the initial voiceprint signal; preprocessing the initial voiceprint signal to obtain an input signal, and establishing an input signal dataset; performing feature extraction on a first input signal in the training dataset based on a preset feature extraction algorithm to obtain a first voiceprint feature; training an initial detection model based on the first voiceprint feature and a first fault type corresponding to the first input signal to obtain a first training result; determining a loss function based on the first training result and the first fault type; and iteratively adjusting a weight value of the initial detection model until the loss function converges to obtain a fault detection model.

The invention relates to the technical field of voice processing, can be applied to business scenes such as financial science and technology and medical health, and discloses a multi-speaker dialogue voice analysis method, device and equipment and a medium, and the method comprises the steps: obtaining a to-be-analyzed multi-speaker dialogue voice, determining a naturalness score based on an acoustic feature, and obtaining a multi-speaker dialogue voice analysis result; determining a semantic consistency score based on voice embedding and semantic embedding corresponding to a preset text, determining a speaker consistency score based on embedding of a plurality of speakers of the same speaker, determining an interaction rationality score based on voice alternate overlapping duration, determining a diversity score based on a variance of voice features, and fusing the scores, the comprehensive mass fraction is obtained. According to the invention, through quantitative evaluation of five dimensions of naturalness, semantic consistency, speaker consistency, interaction rationality and diversity, a comprehensive quality scoring system is established, so that the evaluation result simultaneously reflects voice fluency, content matching degree, identity stability, interaction rhythm rationality and feature richness.

Artificially intelligent systems and methods for financial coaching provide personalized, fiduciary-compliant financial guidance through advanced machine learning architectures with measurable performance criteria. The systems implement privacy-preserving processing pipelines that detect personally identifiable information using multi-layered pattern recognition including regular expressions for formatted data sequences, named entity recognition with confidence thresholds above 0.85, and contextual analysis algorithms. A multi-step artificial intelligence processing workflow includes automated language detection, emotional tone classification with confidence scoring, financial profile transformation using predefined templates, context-aware question rephrasing, and semantic similarity matching employing vector embeddings with financial domain vocabulary weighting applying multiplier values between 1.3-2.0. Specialized training methodologies expand datasets through mathematical transformation functions utilizing statistical standard deviations with incremental variations between 0.5-2.0. Mood-based escalation logic automatically transfers users to human advisors when emotional indicators exceed confidence thresholds above 0.8. The systems maintain response times below 5 seconds while providing regulatory compliance through curated content sources and predefined fiduciary instruction parameters.

The invention provides an underwater acoustic target recognition method based on autonomous task perception, and belongs to the field of underwater acoustic signal processing and artificial intelligence. S3, performing feature extraction on the time-frequency spectrogram by the task type extraction network to obtain a task embedding vector, calculating a similarity score, when the maximum similarity score is greater than a set threshold value, outputting a task representation vector, and entering S3, otherwise, inputting features output by the last Transform layer into a classifier; s3, selecting a router according to the task representation vector, selecting a trained expert network by the router, calculating a door control weight, processing universal acoustic features in parallel by the expert network, fusing output of the expert network, fusing deep features and fused features to obtain enhanced features, and enabling the enhanced features output by the last Transform layer to enter a classifier; the invention further provides a system. The problems that the task category autonomous recognition capability is insufficient and the correlation between tasks is ignored are solved.

The invention discloses a multi-modal emotion recognition method and device, electronic equipment and a storage medium. The method comprises the following steps: acquiring text, video and audio data of a user and respectively performing feature extraction to obtain text features, audio features and facial features; the three features are input into a pre-trained multi-modal emotion recognition model, the multi-modal emotion recognition model comprises a first fusion module, a second fusion module, a third fusion module, a fourth fusion module and a classification module, the audio features and the text features are fused through the first fusion module, and audio text features are obtained; fusing the facial features and the text features by using a second fusion module to obtain facial text features; performing feature enhancement on the text features by using a third fusion module to obtain enhanced text features; fusing the audio text features, the face text features and the enhanced text features by using a fourth fusion module to obtain multi-modal features; and classifying the multi-modal features by using a classification module to obtain a sentiment classification result of the user.

The invention belongs to the technical field of intelligent voice processing, and particularly relates to a spoken language pronunciation training and correcting system based on intelligent equipment, which acquires a user rhythm feature set including pitch change rate, accent intensity, pause duration and intonation contour by acquiring spoken language audio data sent by a user for a target text, and corrects the spoken language pronunciation training and correcting system. The method comprises the following steps: acquiring spoken language audio data, converting the spoken language audio data into a phoneme sequence aligned with target text time, generating a rhythm deviation degree report according to comparison evaluation of a user rhythm feature set and the phoneme sequence, an execution intonation mode, accent distribution, speech stream sound change and speech speed rhythm, and determining the rhythm deviation degree according to a rhythm problem type in the report in combination with user historical learning data. And forming and outputting a correction scheme including a text prompt, a targeted minimum contrast training unit and a listen-and-read simulation task, solving the problem of weak capability of correcting hyper-phoneme in a second language spoken language of an adult, and improving the authentic and fluency of the spoken language, thereby realizing efficient personalized learning.

The invention discloses a music stave sentiment classification method and system based on multi-level distillation, and belongs to the technical field of music information processing. The method comprises the steps of firstly collecting music stave data and converting the data into stave data vectors, then performing feature extraction by using a long short-term memory network, then constructing a teacher network and a student network for knowledge distillation, and realizing multi-level knowledge transmission through temperature scaling, KL divergence loss and mask feature distillation. And finally, training a lightweight classification model to complete sentiment classification. The knowledge distillation technology is creatively applied to staff sentiment classification, the classification accuracy is effectively improved through an online multi-level distillation mode, and the technical problems that a traditional method lacks semantic information and a self-supervised model is not suitable for sentiment tasks are solved. The method has the main advantages of high classification precision, light model weight, capability of effectively capturing music emotion features and the like.

According to the BIM-based construction process monitoring and early warning system provided by the invention, the data acquisition dimension and precision are remarkably improved through multi-source sensing fusion of millimeter-wave radar, multispectral imaging and voiceprint recognition; feature vector voxel units carrying material characteristics and process constraints are adopted, so that risk early warning has space-time relevance and process interpretability; through a composite risk calculation model containing environmental interference correction and time-varying gradient, the problem that a traditional threshold value method is poor in adaptability to complex working conditions is solved; the holographic early warning mechanism realizes upgrading from sound-light alarm to touch sense-three-dimensional projection cooperative interaction; the double-circulation self-optimization system not only guarantees real-time control, but also realizes block chain evidence storage of empirical data, and finally forms a construction monitoring closed-loop system with space-time perception, intelligent decision, accurate early warning and sustainable evolution capabilities.

The invention relates to the technical field of communication network fault diagnosis, and discloses a communication information processing method and device. The method comprises the steps that a voiceprint sensor collects an original signal, and a quantum coding time-frequency matrix is generated through time-frequency conversion and quantum coding; the matrix and real-time circuit diagram data are fused, and topology-fault joint features are generated through feature alignment and wavelet-convolution joint extraction; analyzing the features to reconstruct a node relation graph, and generating real-time updated circuit diagram data through graph optimization; loading an incremental parameter adapter initialization graph neural network based on the data, and executing fault propagation deduction to output fault coordinates and confidence data; generating an incremental parameter adapter according to confidence data federal optimization; and analyzing the fault coordinate matching topology library based on the adapter, and associating the maintenance knowledge base to output a visual repair suggestion. According to the method, fault dynamic deduction is realized, the cross-line generalization ability is enhanced by an incremental learning mechanism, bandwidth occupation is reduced by edge-cloud cooperation, and the fault positioning efficiency and accuracy are remarkably improved.

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating a video output using an autoregressive token generation neural network model In one aspect, a system comprises obtaining a model input, processing the model input to generate an input sequence of embeddings that represents the model input, autoregressively generating a plurality of output sequences of tokens, wherein each output sequence of tokens corresponds to a respective output modality of tokens from a set of a plurality of modalities that includes a video modality and one or more other modalities, and generating a model output that includes a video output of the video modality by decoding the sequence of tokens.

The invention relates to the cross technical field of signal processing and artificial intelligence, in particular to a multi-source noise removal method and system based on DAE, and the method comprises the steps: 1, carrying out the data collection and feature extraction of multi-source noise and pure signals; step 2, constructing a de-noising recognition knowledge base based on feature analysis; step 3, constructing a deep denoising auto-encoder model based on a knowledge base; step 4, hierarchical training and optimization guided by a mixed loss function of the deep denoising model; step 5, denoising processing of a target signal and output evaluation based on a discrimination model; according to the invention, noise data in multiple fields such as electromagnetism, remote sensing and biological signals are integrated, a dynamic mixing strategy and a data enhancement technology are adopted, a training set which highly simulates a real environment is constructed, and a unique cross-scene adaptation module can perform adaptive adjustment according to signal characteristics of different application scenes; the problem that a traditional method is poor in scene adaptability is solved.

The invention relates to the technical field of audio processing, in particular to a self-adaptive scene sound effect adjusting method and device and a storage medium, and the method comprises the steps: collecting a mixed audio stream and an original playing audio, and extracting an acoustic fingerprint vector through a pre-trained convolutional neural network; performing acoustic feature deconstruction on the acoustic fingerprint vector to obtain a plurality of acoustic feature components, performing scene recognition judgment according to a preset scene judgment rule based on the plurality of acoustic feature components, and determining an audio playing scene; constructing a time-frequency mask according to the audio playing scene, generating a noise reduction gain matrix based on the time-frequency mask, and performing scene sound field enhancement on the audio playing scene to obtain a sound field optimization matrix; and performing noise reduction and sound effect adjustment on the original playing audio according to the noise reduction gain matrix and the sound field optimization matrix, and outputting the optimized playing audio. According to the invention, sound effect adjustment can be dynamically adapted according to different scenes, and the noise reduction accuracy and the multi-scene sound quality adaptability are improved.

The invention discloses an intelligent customer service dialogue management method and system fused with multi-layer memory, and belongs to the technical field of computers, and the method obtains a three-dimensional context through multi-mode context awareness, achieves the dynamic scheduling of a memory system through a memory coordinator, and combines a dynamic construction mechanism of a service situation portrait, thereby achieving the intelligent customer service dialogue management. The problem that the response of the intelligent customer service system is not matched with the complex and dynamic service demand of the user due to the single situation awareness dimension and the rigid memory scheduling mechanism is effectively solved, and the adaptive personalized service response can be generated according to the real-time change of the dialogue situation.

The invention belongs to the technical field of hydroelectric generating set fault diagnosis, and particularly discloses a hydroelectric generating set cavitation fault diagnosis method based on multi-channel acoustic emission signal fusion. Through the technical means of acquiring the acoustic emission signals at multiple positions, a more comprehensive cavitation phenomenon data basis is provided; the representation of time-frequency characteristics is optimized through the Mel time-frequency diagram, and the problem that the dynamic characteristics of acoustic emission signals cannot be effectively captured through a traditional method is solved; multi-channel feature extraction is performed on the Mel time-frequency graph through a preset deep convolutional neural network model, so that automatic learning of deep cavitation working conditions is realized; and multi-channel fusion and classification identification are carried out on a feature extraction result through the model, so that effective integration and fault classification of features are realized. Compared with the prior art, a more accurate and comprehensive cavitation fault diagnosis effect is realized, and the diagnosis precision is improved.

The invention provides a counterfeited audio detection system and method based on mask reconstruction and time-frequency feature fusion, and the system comprises a data preparation module, a spectrum feature extraction module, a time-frequency feature fusion and prediction module, a training module, and a counterfeited audio detection module. Through a double-branch structure of an audio mask auto-encoder, features are extracted from two angles of reconstruction learning and direct encoding, enhancement and sequence correlation modeling are performed on the double-branch features, the ability of capturing forged clues is enhanced, frequency domain information and time domain context are combined, fusion features with higher discriminative ability are formed, and the accuracy and the reliability of the audio mask auto-encoder are improved. Through unified optimization of the total loss function, effective learning extraction and fusion of audio features are enhanced, the accuracy of forged audio detection is remarkably improved, and the method is suitable for actual forged audio detection application.

The invention relates to the technical field of monitoring, and discloses a multi-modal video public opinion intelligent monitoring method and system, and the system comprises a multi-modal data collection module, a data processing module, a coefficient generation module, an intelligent analysis module, a dynamic threshold module, and a public opinion visual platform. According to the method, a visual, voice and character cross-modal cooperative computing framework is established, and the one-sidedness problem of traditional single-modal analysis is solved; a recognition-tendency-propagation three-dimensional index system is constructed, and accurate quantitative evaluation of public opinion evolution is achieved; designing a static comprehensive coefficient threshold value and a dynamic growth threshold value of the dual-threshold grading system, and cooperating with an imaginary main body filtering algorithm and KOL polarization value calculation, so that the false alarm rate is remarkably reduced, and meanwhile, rapid early warning response is realized; a three-dimensional propagation map is generated through propagation path concentration analysis, content feature evolution tracking and influence radiation, and a complete public opinion evolution chain and hotspot traceability are provided.

The invention discloses an AI-based illumination control system self-adaptive dimming method, which comprises the following steps of: acquiring a two-dimensional coordinate, a motion track curve and voice communication data of a personnel position, and identifying social interaction scene categories of different time slices; determining a social interaction strength score according to the voice communication data, and predicting the most adaptive spectrum configuration data in the current social interaction scene in combination with the social influence factor matrix and the social interaction scene type; according to a historical dimming log and an energy consumption metering curve, identifying a short-term high-frequency dimming segment, and adjusting a real-time response strategy and a disturbance classification processing mechanism of the spectrum configuration data; and adjusting the illumination power distribution values of the high-frequency use area and the low-frequency use area according to the occupation raster data generated by the millimeter-wave radar and the infrared array in combination with the lamp area mapping relation and the power budget constraint. Through multi-source data fusion and a self-adaptive regulation and control mechanism, unification of real-time performance, pertinence and energy-saving performance of spectrum regulation is realized, and the dynamic response capability and the user perception experience of the intelligent illumination system are improved.

The invention discloses an external voice changing method and system, and the method comprises the steps: building a mapping relation between a voice changing scene and voice changing parameters, and enabling the voice changing scene to be used for representing an external environment where a vehicle is located; in response to the user operation, determining a target voice changing scene; obtaining a target voice changing parameter corresponding to the target voice changing scene based on the mapping relationship between the voice changing scene and the voice changing parameter; and performing voice changing processing on voice played by a loudspeaker outside the vehicle by adopting the target voice changing parameter. According to the method, the voice interaction mode of the vehicle and the external environment has scene perception and self-adaptive capabilities, and the voice outside the vehicle can be matched with the specific environment and is easier to accept.

The present invention comprises a novel social media video debating web and mobile application. The platform will provide a space for users to debate uninterrupted by both the audience and the opponent whereby each participant is given a set time to express their thoughts on a subject matter. The online debate platform provides a controlled setting for the participants to have their debates viewed, voted on and subsequently ranked by the other users of the platform. The online debate platform is also monitored by a unique AI system that updates debate “winners,” flags offensive content, and moderates each debate on the platform in real time. The disclosed platform and following figures will provide a space for individuals to debate subjects in a uniformed structure and have real-time results from active user viewership. The online debate platform aims to provide an established place for constructive debating.

The invention relates to the field of biological medicine, and discloses a method and system for analyzing cough sound by using disease characteristics to diagnose respiratory diseases, and the method comprises the steps: deploying a six-microphone annular array to achieve the precise positioning and triggering of a sound source; self-adaptive spectral subtraction and Wiener filtering cascade are adopted to enhance the audio; segmenting a cough segment based on energy envelope; fusing the Mel-cepstrum, the linear prediction residual error, the harmonic energy ratio and the transient zero-crossing rate to construct a pathological feature matrix; extracting local, medium-range and global time sequence features through a three-branch parallel convolutional network; inputting a disease specific classifier to discriminate asthma, pneumonia and laryngitis respectively, and applying a feature decoupling regular term to improve interpretability. The system correspondingly realizes the modularized processing flow. According to the method, the cough sound collection quality and the disease subtype recognition accuracy in a complex environment are improved, meanwhile, the thermodynamic diagram is output to assist clinical decision making, and the diagnosis credibility and practicability are enhanced.