148 results about "Source separation" patented technology

The invention discloses a music source separation method and wearable equipment, and relates to the technical field of signal processing, music source separation is performed by adopting a neural network model of a UNet architecture, the neural network model at least comprises an encoder and a decoder, and the method comprises the following steps: obtaining an original signal of a to-be-separated music source, and converting the original signal to a frequency domain to obtain a frequency domain signal; performing feature coding processing on the frequency domain signal through an encoder to obtain coded feature data, the feature coding processing including feature extraction, and the encoder performing feature extraction processing at least by using a TFC-TDF block of causal convolution; and performing feature decoding processing on the coded feature data through a decoder, and outputting to obtain a music source separation result. According to the method and the device, low-delay music source separation is realized in the end side equipment with limited resources.

The invention relates to the technical field of signal reconstruction, in particular to a complex electromagnetic environment-oriented millimeter wave signal blind source separation and reconstruction system, which comprises a frequency spectrum trend division module, a path fading construction module, an initial cluster label generation module, a multi-solution path screening module and a fusion reconstruction execution module. According to the method, power spectral density sequence processing is carried out on millimeter wave frequency domain data, a multi-dimensional feature group is formed according to a path loss factor, an angle of arrival and the like, and an initial feature cluster is screened through an Euclidean distance, so that the accuracy of signal source classification is effectively improved; after the frequency domain response and the phase contour are continuously subjected to point comparison, path screening is completed by integrating a mean square error residual value, the path misjudgment probability is reduced, time domain resampling and phase frequency offset standardization are executed in fusion reconstruction, the consistency and fidelity of signal reconstruction are enhanced, and the reconstruction precision is improved. The whole process improves the separation accuracy and reconstruction precision of mixed signals in a complex electromagnetic environment.

The invention relates to the technical field of geodetic survey and structural health monitoring, and discloses a GNSS-RTK coordinate domain error correction method, which comprises the following steps: acquiring GNSS-RTK dynamic observation data to form a mixed signal time sequence; executing an improved adaptive noise complete empirical mode decomposition algorithm on the sequence to obtain an intrinsic mode function component; identifying and eliminating high-frequency components representing Gaussian white noise based on an energy coefficient, and reconstructing residual components into a signal sequence after primary noise reduction; inputting the noise reduction sequence into a rapid independent component analysis model for blind source separation; and finally, carrying out sorting and phase and amplitude uncertainty correction on the separated independent components, and outputting a multi-path error model and a structure dynamic deformation signal. According to the method, the problem of blind source separation failure or low precision caused by strong noise covering source signal statistical characteristics is solved through a strategy of first noise reduction and then separation, and an effective physical signal can be accurately extracted from a strong noise background.

The invention provides a GIS partial discharge positioning method based on deep learning and acoustoelectric combination, and relates to the field of data processing. The method comprises the following steps: firstly, acquiring an original ultrahigh frequency signal and an original ultrasonic signal of a target detection entity, constructing an acoustoelectric signal matrix, and performing short-time Fourier transform to obtain a complex spectrum tensor and a power spectrum tensor; and then, extracting a high-dimensional feature embedding tensor by using a double-branch deep learning network, and generating a time-frequency masking tensor and a source confidence tensor through a multi-source sub-ion network to complete source-by-source spectrum reconstruction. And a time difference of arrival matrix is calculated by using a phase transformation generalized cross-correlation method, a sensor space coordinate is combined to be input into the depth regression positioning sub-network, and a discharge source is output to predict three-dimensional coordinates and regression uncertainty. And finally, fusing the regression uncertainty and the source confidence tensor to construct a confidence interval, and generating a visual positioning result with an ellipsoid confidence boundary. By implementing the technical scheme provided by the invention, the accuracy of partial discharge positioning is improved.

The invention relates to an array microphone noise reduction recording method based on cascade noise reduction and blind source separation, which belongs to the technical field of voice signal processing and recording, and comprises the following steps: configuring a multi-channel array microphone, ensuring that the amplitude and phase of a channel signal are consistent, and collecting an original multi-channel voice signal; a weighted kernel function blind source separation algorithm is adopted, signal-to-noise ratio distribution characteristics of signals are extracted, kernel function weights are given, and target voice and interference signal components are obtained through decoupling of an independent component analysis model; executing target-oriented adaptive cascade noise reduction, locking the voice of a keynote speaker through directional pickup, reducing noise, filtering out reverberation, and enhancing the voice of a far-field target by combining a voice mask neural network with a far-field pickup algorithm in sequence; and processing the target voice through voice feature perception lossless coding and storing the target voice. According to the invention, stable acquisition of multi-channel signals, accurate separation of mixed signals and layered suppression of noise reverberation are realized, the signal-to-noise ratio and definition of far-field voice are significantly improved, and the method is suitable for single-person speaking or multi-person dialogue scenes.

The invention relates to the technical field of smart home control, and particularly discloses a smart home central control system and method based on multi-mode perception, and the system comprises the steps: synchronously collecting a voice audio signal, a gesture image signal, an infrared thermal imaging signal and a millimeter wave radar signal through a plurality of groups of sensors; performing blind source separation processing on the voice and gesture signals, extracting a voice command component and a gesture action component which are independent in statistics, and performing space-time alignment and Kalman filtering fusion on the infrared and radar signals to generate a dynamic environment sensing graph; voice intention features, gesture track features and environment anomaly features are extracted through Mel frequency cepstrum coefficient analysis, skeleton key point tracking and multi-level convolution processing; constructing a three-dimensional decision matrix based on the features, performing weighted evaluation through a fuzzy logic rule base to generate a control instruction priority sequence, and dynamically adjusting an equipment operation mode according to the priority; according to the invention, the problems of control conflict and response delay caused by multi-mode signal coupling are solved.

The invention relates to the technical field of electric power operation and maintenance, in particular to a transformer fault diagnosis method, device and equipment and a storage medium, and the method comprises the steps: collecting a transformer acoustic monitoring signal based on an acoustic microphone array, carrying out blind source separation and decoupling, and constructing an acoustic signal feature set; transformer operation condition parameters are extracted, mechanical state degradation evaluation is carried out according to the acoustic signal feature set, and a mechanical state health degree index is generated; and calculating an acoustic signal receiving timestamp of the acoustic microphone array, performing sound source distribution mapping, and constructing a sound source spatial distribution diagram. According to the method, blind source separation decoupling, mechanical state degradation evaluation, sound source distribution mapping and fault situation diagnosis are performed by acquiring the acoustic signals and the operation condition parameters, so that tiny mechanical faults in the transformer can be accurately identified, intelligent maintenance decision is realized, and the method has the advantages that the tiny mechanical faults in the transformer can be detected earlier; the real-time performance and reliability of fault diagnosis are improved, and the accident risk caused by diagnosis lag is reduced.

A conferencing system includes a plurality of microphones and an audio processing system that performs blind source separation operations on audio signals to identify different audio sources. The system processes the separated audio sources to identify or classify the sources and generates an output stream including the source separated content.

The invention provides a monitoring method and system for a cutting penetration state in a water-jet guided laser processing process based on sound signal blind source separation, and the monitoring method comprises the steps: S1, in the water-jet guided laser processing process, collecting a plurality of observation sound signals in real time, the observation sound signals being mixed sound signals composed of a plurality of independent sound source signals; s2, estimating and solving a plurality of independent sound source signals by using a blind source separation algorithm according to the plurality of collected observation sound signals; s3, the frequency spectrums of the multiple independent sound source signals are compared with sound signals generated during material laser processing, so that sound signals generated during water-guided laser processing due to material removal are determined from the frequency spectrums; and S4, the water-jet guided laser machining process is judged according to the relation that sound signals generated due to material removal change along with time. Compared with a traditional coaxial visual monitoring scheme, the method is not affected by factors such as water flow and light intensity, and can be applied to monitoring of the water-guided laser machining process.

The invention relates to the technical field of power equipment fault monitoring, in particular to a distribution box fault monitoring method, and aims to solve the problem that a traditional monitoring method in the prior art depends on a single filtering or fixed threshold strategy and is difficult to effectively separate fault features. According to the method, mechanical vibration and electromagnetic interference signals can be effectively separated through an adaptive blind source separation algorithm, the problem of early fault signal identification in a mixed interference environment is solved by combining multi-scale feature fusion and a dynamic environment compensation mechanism, and the method has the advantages of improving monitoring accuracy and realizing accurate quantification of fault energy.

The invention discloses a leakage monitoring method based on an LNG fuel gas system. The leakage monitoring method comprises the steps that a self-adaptive frequency modulation continuous wave active acoustic scanning network is established; blind source separation is carried out on mixed signals in acoustic scanning network abnormal events by adopting self-adaptive kernel independent component analysis; constructing a leakage feature mapping model of the physical information neural network; leakage source accurate positioning and quantification based on acoustic tomography and Bayesian reasoning are carried out; multi-modal decision fusion is carried out based on the multi-dimensional data sources received in parallel, a false alarm suppression mechanism is set, and time continuity verification and space consistency verification are carried out; establishing a reinforcement learning model for autonomously optimizing a monitoring strategy according to environment change and system state, and realizing adaptive optimization; the strategy network after self-adaptive optimization is deployed at the cloud, actions are generated regularly according to the current state, the actions are issued to the regional gateway and the edge node for execution, and iterative updating is carried out, so that the monitoring accuracy in a complex environment is improved, and the false alarm rate is reduced.

The invention discloses a phase modifier bearing fault diagnosis method, which belongs to the technical field of power equipment fault diagnosis, and comprises the following steps of: analyzing independent components, performing blind source separation on a received phase modifier bearing vibration signal, and extracting three types of independent source signals of impact, abrasion and noise; the gradient driving window length is self-adaptive, and the window length of short-time Fourier transform is dynamically adjusted based on the instantaneous frequency gradient of the independent source signal; wavelet packet frequency band energy screening: performing wavelet packet decomposition on the signal after window length adaptive processing, screening a fault characteristic frequency band based on an energy contribution rate, and reconstructing the signal; and enhancing stochastic resonance, and inputting the reconstructed signal into a stochastic resonance system. According to the method, independent component analysis, gradient driving window length self-adaption, wavelet packet frequency band energy screening and stochastic resonance enhanced fourth-order diagnosis chain are constructed, so that multi-stage cooperative processing of phase modifier bearing faults is realized, and the technical problems of time-frequency resolution contradiction, insufficient feature decoupling and weak generalization ability are effectively solved.

The invention provides a method for determining a noise component demarcation point in an electric signal of power transformation equipment, which belongs to the technical field of power transformation equipment, and comprises the following steps of: firstly, carrying out time domain sampling and standardization on the electric signal through a data preprocessing module of a hardware unit, and then extracting a frequency domain feature by utilizing fast Fourier transform; blind source separation is carried out through independent component analysis, and an adversarial neural network model is constructed to carry out deep analysis on independent component signals. The generator network generates a noise feature vector, and the discriminator network evaluates a noise component proportion and sets an adaptive threshold to mark a noise dominant signal. And finally, time-frequency energy density distribution characteristics are obtained through wavelet transformation, the energy concentration degree and the frequency bandwidth are calculated, noise component demarcation points are accurately determined, accurate recognition and processing of the electric signal noise of the power transformation equipment are achieved, and the problems that in the prior art, an artificially designed characteristic extraction algorithm is often depended on, self-adaptability is lacked, and the noise is poor are solved. And the effect is not good in a complex and changeable noise environment.

The invention discloses an improved SVMD combined SOBI bridge dynamic strain signal adaptive noise reduction method, which comprises the steps of obtaining an original signal of bridge dynamic strain through a strain sensor, improving an SVMD algorithm by taking kurtosis and a decomposition error as stopping criteria, and further decomposing the original signal; classifying the intrinsic mode function components through a K-means clustering algorithm so as to reconstruct a noise component signal and an initial noise reduction signal; constructing a three-channel hybrid observation matrix containing an original signal, a noise component signal and an initial noise reduction signal; performing blind source separation on the hybrid observation matrix by using an SOBI algorithm, and estimating to obtain a separated source signal and a hybrid matrix; noise components are identified based on kurtosis clustering, signals are reconstructed after frequency points are set to be zero, and high-precision noise reduction is achieved. According to the method, a deep collaborative architecture is constructed by introducing an adaptive decomposition mechanism, intelligent clustering recognition and accurate reconstruction are adopted, and efficient suppression of noise and high-fidelity retention of signal features are realized.

The invention discloses a music source separation method and system based on a hybrid expert self-attention network, and belongs to the technical field of audio signal processing and deep learning. The method comprises the following steps: receiving a time domain mixed audio signal, and obtaining a complex frequency spectrum through short-time Fourier transform; the method comprises the following steps: estimating a complex ideal proportion mask through an improved separator network MoEFormer, respectively modeling a time domain dependency relationship and a frequency domain dependency relationship in the separator network by adopting an axial attention mechanism, and replacing a feedforward network in a standard Transformer with a hybrid expert layer; the hybrid expert layer comprises an expert network specially designed for drum sound, Bass and human voice characteristics, and a self-adaptive weight distribution mechanism based on gating routing; and finally, reconstructing the separated time domain signal through inverse short-time Fourier transform. Through expert specialization and feature adaptive fusion, the problem of multi-sound-source feature confusion is effectively solved, and low calculation complexity is kept while the separation precision is improved.

The invention discloses a belt conveyor safety online monitoring method and device, a medium and a product, and relates to the field of belt conveyor safety monitoring, and the method comprises the steps: collecting a multi-vibration-source signal and a temperature signal of a belt conveyor through a vibration reduction and sensitization acoustic vibration optical cable device; the vibration reduction and sensitization acoustic vibration optical cable device is of a multi-core optical fiber spiral winding structure, and a gradient vibration reduction layer is integrated on the outer layer. Sequentially carrying out wavelet packet decomposition, blind source separation and deep learning optimization on the multi-vibration-source signal, and determining a single-vibration-source signal corresponding to each part; performing time synchronization on the single vibration source signal and the corresponding temperature signal by adopting a dual-optical fiber redundant link and an optical fiber gyroscope; and performing multi-source data fusion diagnosis by adopting a D-S evidence theory according to the synchronized single vibration source signal and the corresponding temperature signal in combination with a sound signal, and generating a fault early warning signal. The distributed optical fiber monitoring accuracy and stability of the belt conveyor can be improved.

A computing system including one or more processing devices configured to receive a first music sample and a second music sample. At a music source separation machine learning (ML) model, the one or more processing devices extract a first vocal component from the first music sample and a second vocal component from the second music sample. At a voice embedding ML model, the one or more processing devices extract one or more first embedding vectors from the first vocal component and one or more second embedding vectors from the second vocal component. The one or more processing devices compute a similarity value between the first and second embedding vectors and determine whether the similarity value is above a predefined similarity threshold. Based on the determination, the one or more processing devices output an indication of whether the first music sample has a same singer as the second music sample.

The invention discloses a graph signal blind source separation method in a white noise environment, and belongs to the technical field of signal processing. Aiming at the problem that the traditional blind source separation performance is reduced due to additive white noise, the invention provides a joint optimization scheme combining blind compression nonlinear noise suppression and image smoothing regularization. Firstly, a blind compression function is applied to a noisy observation image signal for preprocessing; secondly, carrying out mean value removal and whitening processing on the compressed signal; then, constructing a graph Laplacian matrix as a smoothing operator, establishing a joint diagonalization objective function fusing graph autocorrelation, a FastICA item and a graph smoothing regular item, and adopting a Givens rotation algorithm to iteratively optimize and solve a separation matrix; and finally, reconstructing a source signal through multiplication of the separation matrix and the whitening data. According to the method, noise interference is effectively suppressed through blind compression, signal structure consistency is maintained by using graph smoothing prior, and separation robustness, precision and convergence stability in a strong noise environment are remarkably improved.

A method for processing seismic data includes receiving blended seismic data from one or more seismic sources. The method also includes applying a transform to the blended seismic data to decompose the blended seismic data into different parameters. The method also includes applying one or more independent sparse inversions to the different parameters. The method also includes defining a set of prior information techniques to be used within the one or more independent sparse inversions. The method also includes determining an energy part of the blended seismic data that is greater than a first predetermined threshold based at least partially upon the multiple independent sparse inversions, the set of prior information techniques, or both. The method also includes removing the energy part from the blended seismic data to produce modified seismic data.

The application discloses an end-to-end audio-visual source separation method based on dynamic gating fusion. The method uses a dynamically adjusted fusion mechanism to effectively fuse audio and visual features through a dynamic gating fusion module, and further separates the mixed audio signal through an audio-visual Transformer decoder. The method can accurately separate the target audio in a complex multi-source environment, significantly improving the quality of audio-visual source separation, especially in the presence of background noise and multiple sources. Experimental results show that the performance of the method on multiple benchmark datasets has been significantly improved, verifying its effectiveness and advantages in the audio-visual source separation task.

The application discloses a non-invasive load decomposition method based on CEEMDAN and FastICA, and belongs to the technical field of load monitoring. The method comprises the following steps: S1, collecting active power of total load and various single loads and preprocessing; S2, constructing a completely adaptive noise ensemble empirical mode decomposition (CEEMDAN) model, and decomposing total load power; S3, estimating the number of sources based on Bayesian information criterion; S4, performing dimension reduction by using maximum information coefficient (MIC); S5, performing load decomposition by using FastICA blind source separation; and S6, evaluating the approximation degree of decomposed signals and source signals. The application realizes load decomposition from the perspective of signal blind source separation, reduces the cumbersome load information feature extraction, and obtains complete load information through decomposition. Compared with deep learning, the model training time is greatly reduced.

This application provides a signal processing method, chip, electronic device, and storage medium. The method includes: acquiring an input signal, wherein the input signal is a speech signal received by multiple microphones; estimating the covariance matrix of the Nth frame of the input signal to obtain a target covariance matrix of the Nth frame; updating the demixing matrix based on the target covariance matrix of the Nth frame to obtain target elements in the demixing matrix of the Nth frame; performing amplitude demixing based on the target elements in the demixing matrix of the Nth frame to obtain a target demixing matrix of the Nth frame; and performing signal separation based on the target demixing matrix of the Nth frame and the input signal of the Nth frame to obtain an output signal of the Nth frame. The method provided in this application helps to balance the performance and robustness of blind source separation algorithms and improve the signal-to-noise ratio of speech signals.

This application discloses a music source separation method and wearable device, relating to the field of signal processing technology. It employs a UNet architecture neural network model for music source separation. The neural network model includes at least an encoder and a decoder. The method includes: acquiring the original signal of the music source to be separated; converting the original signal to the frequency domain to obtain a frequency domain signal; performing feature encoding processing on the frequency domain signal through the encoder to obtain encoded feature data, wherein the feature encoding processing includes feature extraction, and the encoder performs feature extraction processing at least by using causal convolutional TFC-TDF blocks; and performing feature decoding processing on the encoded feature data through the decoder to output the music source separation result. This application achieves low-latency music source separation in resource-constrained edge devices.

This invention relates to a blind source separation method for jointly stationary correlated source signals, belonging to the field of signal processing. Firstly, based on a linear model for blind source separation, this invention proposes fundamental assumptions about the model. Then, it decomposes the observed signal into a regular component and a predictable component, and studies a method for extracting useful information from the predictable component. Based on this method, the mixing matrix is ​​estimated, and the separation of correlated source signals is achieved. This invention decomposes the observed signal into a regular component and a predictable component, extracting useful information from the predictable component, thus solving the problem that traditional blind source separation methods, based on the assumption that source signals are independent or uncorrelated, cannot separate correlated source signals.

The invention discloses a streaming spatial audio separation method, streaming spatial audio separation equipment and a vehicle-mounted audio system, and relates to the technical field of audio processing. The streaming spatial audio separation method comprises the following steps: extracting a time-frequency feature based on a target sound source signal, performing multi-scale music feature extraction and frequency division compression on the time-frequency feature to obtain a compressed feature, and performing hierarchical long-time sequence dependence modeling on the compressed feature to obtain a multi-stage fusion feature, and performing frequency division decoding on the multi-level fusion features to obtain separated music elements. By adopting the scheme of the invention, high-fidelity, low-delay, multi-source separation and spatialization output of the music source under a resource-limited platform can be realized.

The application discloses a time-frequency domain underdetermined blind source separation method and system based on double sensors, which comprises two stages: in the first stage, a single source point with high clustering characteristics is detected by using a clustering and matching tracking algorithm, and a mixing matrix is estimated, so that high-precision mixing matrix estimation is realized; in the second stage, a time-frequency domain signal recovery problem is converted into a sparse recovery model with a relaxed sparse condition, the constraint of the number of sources on the self-source point in the traditional blind source separation algorithm is broken, and good separation effect is realized. The application is suitable for the double-sensor underdetermined blind source separation scene with more source numbers, low mixing signal-to-noise ratio and serious aliasing, and has strong applicability and practicability.

The application discloses a speech privacy protection method and device based on high-frequency sound signal mixing, first, user speech data is collected to build a personalized user corpus; then, the user customizes the scene requirement, selects the text theme of speech synthesis, and expands the user corpus; then, the user selects the noise theme related to the interference from the selected text theme according to the specific life situation, randomly extracts the speech fragment from the user corpus corresponding to the noise theme to generate the interference noise; the interference noise is modulated to the preset frequency band interval; finally, when it is detected that the user is talking, the controller sends the modulated interference signal to the jammer, and the jammer injects the interference signal into the nearby microphone. The application has practicability and safety, can effectively protect the user privacy, and can resist attacks of various de-noising technologies such as speech enhancement and blind source separation.

The application discloses an InSAR surface deformation multi-source vertical contribution analytical method based on deformation signal spectrum decomposition and belongs to the technical field of surface deformation monitoring. The steps are as follows: first, long-time SAR images are acquired, vertical deformation data and related rates and deformation variables are obtained through SBAS-InSAR technology processing; then, a plurality of auxiliary conditions are fused, regional spatial partitioning is realized through K-Means clustering; subsequently, CEEMDAN decomposition and mean threshold multi-scale reconstruction are performed on deformation signals of each partition, high-frequency items, low-frequency items and trend items are obtained; finally, ICA blind source separation and variance contribution rate analysis are performed, and shallow, medium and deep surface deformation information is inversed through directional reorganization; the K-Means clustering algorithm is adopted to perform spatial partitioning on the regional deformation field, the homogeneity and precision of subsequent signal processing are improved; and the CEEMDAN decomposition and multi-scale reconstruction method is introduced, adaptive and fine scale separation of deformation time series signals is realized; the blind source separation technology is applied to multi-scale component reorganization, and the physical attribution of the vertical position of the subsidence source is realized.

The invention provides a signal noise reduction method for acoustic monitoring of a tool wear state in a milling process, which relates to the technical field of intelligent manufacturing, and is characterized in that main shaft noise is eliminated through spectral characteristics of IMF components obtained through adaptive noise complete set empirical mode decomposition in combination with spectral characteristics of signals acquired by a multi-scene milling test; first-stage noise reduction is realized; performing fast independent component analysis on the residual IMF components after the first-stage noise reduction to realize blind source separation, and identifying and removing random noise components in combination with a fuzzy entropy threshold criterion to realize second-stage noise reduction; reconstructing a de-noised signal by using the residual signal components after the second-stage noise reduction, and screening out features conforming to a tool wear trend through a Kendall rank correlation coefficient; and inputting the screened features conforming to the tool wear trend into a convolutional neural network to realize high-precision intelligent monitoring of the tool wear state based on the sound signal. According to the invention, stable denoising performance is maintained under different working conditions.

This disclosure provides methods, devices, and systems for speech enhancement. The present implementations more specifically relate to utilizing multiple modalities to suppress audio originating from a distractor audio source without distorting audio originating from a target audio source. In some aspects, a speech enhancement system may receive a multi-channel audio signal via a microphone array and may further receive an image associated with a respective frame of the audio signal. The speech enhancement system detects one or more target faces in the image and determines whether the audio frame originates from a target audio source. For example, the speech enhancement system may compare a respective direction of each target face with a direction-of-arrival (DOA) of the audio frame. The speech enhancement system may selectively steer a beam associated with a multi-channel beamformer toward the DOA of the audio frame based on whether the audio frame originates from a target face.