21 results about "Spectral mask" patented technology

Embodiments are disclosed for converting audio data to studio quality audio data. The method includes obtaining an audio data having a first quality for conversion to studio quality audio. A first machine learning model predicts a set of acoustic features. A spectral mask is applied to the audio data during the prediction of the set of acoustic features. A second machine learning model generates studio quality audio from the set of acoustic features and the audio data.

The present disclosure provides methods, components, devices, and systems for 480 and 640 megahertz (MHz) transmission in Wi-Fi. Some aspects are more particularly directed to including a bandwidth field and a bandwidth extension field in a preamble of a physical layer protocol data unit (PPDU), the bandwidth field and the bandwidth extension field jointly indicating that a channel bandwidth of the PPDU is a continuous 480 MHz channel bandwidth or a 640 MHz channel bandwidth. In some aspects, parameters of the PPDU may be defined based on the PPDU occupying a 480 MHz or 640 MHz bandwidth. The parameters may include, for example, a tone plan, an allowed puncturing pattern, signaling of a puncturing plan, signaling of a resource element allocation, a short training field, a long training field, a pilot signal sequence, a phase shift, a segment parser, and / or spectrum masking.

Various aspects of the present disclosure generally relate to wireless communication. In some aspects, a user equipment (UE) may perform a crest factor reduction on a radio frequency (RF) waveform in multiple passes, wherein the multiple passes each include: a clipping operation to reduce a peak-to-average power ratio (PAPR) associated with the RF waveform according to a threshold; and a filtering operation to remove at least a portion of out-of-band emission from the clipped RF waveform. The UE may transmit the RF waveform over a wireless channel, wherein the RF waveform transmitted over the wireless channel satisfies a spectral mask. Numerous other aspects are described.

This disclosure describes systems, methods, and devices related to reducing noise in and improving speech signals and radio signals. A device may identify a radio frequency signal received by radio frequency circuity at a time; apply a Short Time Fourier Transform (STFT) to the radio frequency signal to generate a STFT signal; identify a signal-to-noise ratio associated with the radio frequency signal at the time; identify a packet-error-rate associated with the radio frequency signal at the time; identify activity when the multiplexed radio frequency signal is received by the radio frequency circuity at the time; select, based on the signal-to-noise ratio, the packet-error-rate, and the activity, a spectral mask to be applied to the STFT signal; apply the selected spectral mask to the STFT signal to generate a clean STFT signal; and apply an inverse STFT to the clean STFT signal to generate a clean radio frequency signal.

The invention discloses a network flow time sequence decomposition method and system based on hierarchical decoupling, belongs to the technical field of time sequence analysis and network flow prediction, and solves the problems of insufficient global trend and local disturbance signal separation capability and the like in the prior art. The method comprises the following steps: processing an input network traffic time sequence by adopting a sliding window moving average algorithm to obtain a trend component, and extracting a seasonal component; dividing into a plurality of windows, and carrying out Fourier transform to obtain an amplitude sequence; after obtaining an amplitude sequence, carrying out averaging processing and sorting, selecting a frequency index with the maximum amplitude, and constructing a frequency spectrum mask; dividing the frequencies of the trend component and the seasonal component into a global frequency and a local frequency based on a spectrum mask, and respectively executing inverse Fourier transform to obtain four decomposition components; and adopting a multi-component time sequence feature fusion model to reconstruct and optimize the four decomposed components to obtain enhanced time sequence feature representation. The method is used for improving network traffic prediction.

The invention discloses a hyperspectral anomaly detection method based on double-domain multi-scale feature reconstruction, which comprises the following steps: firstly, in order to relieve the influence of abnormal pixels and spectral variation on background recovery, reducing the abnormal pixels through a space-spectral mask strategy, thereby avoiding interference on subsequent feature extraction; then, due to the complexity of the background, the invention designs a multi-scale double-domain feature interaction module which is used for extracting multi-scale features of the background in different domains. Specifically, multi-scale spatial features are extracted in a spatial domain through a multi-scale convolution block, and frequency domain features are extracted in a frequency domain through multi-scale wavelet convolution. In addition, the features of different domains are fully fused through feature interaction, and the background modeling capability of the network is improved. And finally, performing anomaly detection on the residual error of the reconstructed background and the input image through the mahalanobis distance. According to the method, features can be effectively extracted for targets with different shapes and multi-scale targets, so that a relatively good detection effect is achieved.

This disclosure provides methods, components, devices and systems for ambient power downlink bandwidth control. Some aspects more specifically relate to generating, by a transmitting device, a spreading waveform by concatenating a sequence of symbols having a first duration, and modulating an on-off keying (OOK) waveform having a different symbol duration with the spreading waveform. In some examples, the different waveforms may have different symbol durations, and symbol boundaries for the OOK waveform and the spreading waveform may or may not align in time. The spreading waveform may include a set of concatenated orthogonal frequency division multiplexing (OFDM) symbols, Barker sequences, or the like. The OOK waveform may include zero-mean OOK signals. The access point (AP) also may apply filtering to ensure that the generated waveform satisfies a spectral mask.

1. Name of the product in this design: Spectral Mask (2563). 2. Purpose of this design: For facial beauty and care. 3. The key design feature of this product is its shape. 4. The image or photograph that best illustrates the design's key points: 3D view 1.

The invention designs an innovative method suitable for link prediction tasks in a dynamic graph convolutional network. Most of existing dynamic graph link prediction methods are focused on independent coding of interaction sequences in a time domain, the potential causal relationship between node interactions cannot be effectively mined, and time migration existing in the interaction process is difficult to capture. The invention designs a multi-view adaptive frequency domain dynamic graph convolutional network. Firstly, multi-view information including nodes, edges, time and node interaction frequency is coded by using a time sequence interaction perception graph embedding module; wherein the node interaction frequency is coded by a neighbor interaction frequency method based on a matrix and is used for describing an interaction evolution rule between nodes. Fusion features are subjected to feature enhancement through a dual-channel spectrum mask module and then transferred back to a time domain, nonlinear features are captured and aggregated by using MLPmultiplier, efficient and accurate prediction of future interaction relationships is realized, and the method has both performance and speed advantages and has wide application value.

The invention relates to the technical field of memory control and artificial intelligence, and discloses a 3D stacked memristor array optimization method based on AI data analysis, and the method comprises the following steps: S1, extracting frequency domain impedance response data containing a resistance component and a reactance component, and constructing a local state vector reflecting a physical state; s2, in combination with the time evolution characteristics of the historical operation sequence, outputting a sensitivity graph representing the interference susceptibility degrees of the target unit and the reference unit to different frequencies; s3, determining the optimal write-in carrier frequency and write-in voltage amplitude of each layer through game iteration; and S4, generating a write-in voltage waveform subjected to frequency spectrum mask filtering, and applying the write-in voltage waveform to the target unit to complete write-in operation. According to the method, the frequency domain impedance is analyzed through the AI to predict interference, the orthogonal frequency is distributed by using the game theory, and the reverse offset signal is applied, so that efficient concurrent writing of the 3D stacked array is realized, interlayer crosstalk is effectively inhibited, and the data stability is improved.

The application provides a generative self-supervised hyperspectral image target detection method based on a spectral mask, features of hyperspectral remote sensing data are encoded and decoded through a generative model and a spectral mask strategy, a general feature representation is learned in the absence of labeled samples, a transfer learning model based on a lightweight transformer encoder, a full connection layer and a Sigmoid layer is constructed, a detector composed of a full connection layer and a Sigmoid layer is fine-tuned in combination with a small amount of training samples and a weighted binary cross-entropy loss, rapid training of the transfer learning model is realized, and the pre-training model is effectively transferred to different target detection tasks. The application has the beneficial effects that: the hyperspectral remote sensing image itself features are fully utilized, the general feature representation is learned from the unlabeled data according to the self-supervised learning theory by using the generation task, and the pre-training model is transferred to various detection tasks, so that the detection precision and time efficiency are improved.

This terminal is provided with an authentication information holding unit and a wireless signal processing unit. The authentication information holding unit holds authentication information relating to authentication of a use region with respect to a spectrum mask when preamble puncturing is used. The wireless signal processing unit transmits authentication information relating to the use region of the terminal to an access point.

The invention relates to the technical field of vehicle-mounted communication, in particular to a vehicle-mounted wireless communication module real-time interference avoidance method based on edge computing, which comprises the following steps of: capturing burst pulse interference of the Internet of Vehicles and a cellular frequency band through marginalized spectrum sensing, extracting a time domain duty ratio and a spectrum mask feature, and forming a dynamic interference feature set; performing multi-level classification on interference sources by using a lightweight convolutional neural network, calculating confidence coefficients in combination with signal entropy values, and generating interference situation data; fusing the vehicle position, the road topology and the surrounding node state to construct a risk thermodynamic diagram, driving an edge engine to dynamically redistribute spectrum resources, and outputting a channel switching instruction; controlling a radio frequency front end to adjust power and modulate a coding scheme through an instruction, and synchronously acquiring link quality feedback; by comparing a feedback value with a safety threshold value in real time, a redundancy mechanism is triggered to realize seamless switching of a main link and a standby link, an uploaded log is encrypted, and high reliability and safety of communication are ensured.

The invention belongs to the technical field of gas pipeline unmanned aerial vehicle leak detection, and discloses an unmanned aerial vehicle inspection control system and method based on acoustic imaging gas pipeline leak detection, an optical imaging auxiliary unit collects environment images in real time, and a scene adaptive switching unit identifies scene types through image features and loads corresponding interference feature templates; the interference characteristic acquisition subunit synchronously acquires three-dimensional data of pipeline vibration, machine body vibration and environmental noise, modeling is performed through the CNN-LSTM interference characteristic learning unit to generate an interference type, an intensity grade and a frequency spectrum mask, the interference type, the intensity grade and the frequency spectrum mask are transmitted to the composite noise reduction preprocessing unit to support a three-stage noise reduction process, a pure leakage point signal is finally generated, residual interference is effectively eliminated, and the noise reduction efficiency is improved. The omission ratio is reduced; the collaboration module establishes a control instruction channel, a detection data channel and an optimization parameter channel, a collaboration instruction priority channel unit ensures that a control instruction is transmitted preferentially, and a block chain data evidence storage unit generates hash values for all transmission data to ensure that full-link data is credible.

The present application relates to the field of vehicle-mounted communication technology, and more particularly to a real-time interference avoidance method for a vehicle-mounted wireless communication module based on edge computing, which captures burst pulse interference in the vehicle networking and cellular frequency bands through edge spectrum sensing, extracts time domain duty cycle and spectrum mask features to form a dynamic interference feature set, classifies interference sources at multiple levels using a lightweight convolutional neural network, calculates confidence levels in combination with signal entropy values to generate interference situation data, constructs a risk heat map by fusing vehicle position, road topology and surrounding node state, drives the edge engine to dynamically redistribute spectrum resources, and outputs channel switching instructions; the instructions control the power and modulation coding scheme of the radio frequency front end, and synchronously collect link quality feedback; by comparing the feedback value with the safety threshold in real time, the standby link is seamlessly switched through the trigger of the redundancy mechanism, and logs are uploaded after encryption, ensuring high reliability and security of communication.

This invention discloses a speaker recognition method, system, storage medium, and device based on ordinary pronunciation. The method includes: acquiring real-time audio data and extracting spectral features based on the real-time audio data to obtain spectral features corresponding to the real-time audio data; inputting the spectral features corresponding to the real-time audio data into a trained UNET network to generate a spectral mask corresponding to the real-time audio data, and detecting whether the real-time audio data is an ordinary pronunciation based on the spectral mask; if the real-time audio data is an ordinary pronunciation, fusing the spectral mask and spectral features to obtain an enhanced spectrum corresponding to the real-time audio data; inputting the enhanced spectrum corresponding to the real-time audio data into a trained speaker embedding layer network to obtain a real-time speaker embedding layer corresponding to the real-time audio data; and comparing the real-time speaker embedding layer with the registered speaker embedding layer to identify the speaker corresponding to the real-time audio data.

The invention discloses an acoustic side channel attack detection and identification method, which comprises the following steps of: acquiring keystroke audio data when a user inputs a keyboard, and acquiring an acoustic signal generated in a keystroke process; the collected keystroke audio is detected and separated, and a target keystroke fragment is extracted; performing noise reduction processing on the keystroke fragment; extracting Mel time-frequency features from the denoised keystroke segments; self-supervised learning is adopted on the keystroke audio data without labels, spectrum masks are applied to keystroke features, a feature reconstruction task is designed, and a model is guided to obtain discriminative keystroke features in combination with discrimination task constraints; and carrying out model fine tuning by using a tagged keystroke sample, optimizing classification performance, and carrying out classification prediction and user input decoding on a keystroke fragment to obtain a final keystroke content restoration result. According to the method, the modeling capability of keystroke signals is remarkably improved, the generalization performance of downstream classification tasks is enhanced, the accuracy and robustness of keystroke classification are improved, and relatively high environmental robustness is shown.

The present application belongs to the technical field of unmanned aerial vehicle leak detection of gas pipeline, and discloses an unmanned aerial vehicle inspection control system and method for leak detection of gas pipeline based on acoustic imaging, wherein an optical imaging auxiliary unit collects environmental images in real time, a scene self-adaptive switching unit identifies the scene type through image feature recognition and loads a corresponding interference feature template; an interference feature acquisition subunit synchronously collects three-dimensional data of pipeline vibration, body vibration and environmental noise, interference types, intensity levels and spectrum masks are generated through modeling of a CNN-LSTM interference feature learning unit, and are transmitted to a composite noise reduction preprocessing unit to support a three-level noise reduction process, finally generating a pure leak point signal, effectively eliminating residual interference and reducing the missed detection rate; a collaboration module establishes three types of channels of control instructions, detection data and optimization parameters, a collaborative instruction priority channel unit ensures priority transmission of control instructions, and a blockchain data notarization unit generates hash values for all transmission data, ensuring that the data of the whole link is reliable.

The invention discloses a multi-frame multispectral mask design method for a real system. The method comprises the following steps: obtaining an imaging response model of the multispectral imaging system under multiple wavelengths, wherein the model is obtained by actual measurement; on the basis of the imaging response model, constructing a continuous and derivable forward imaging model, and then performing position-related convolution superposition on the scene under an incoherent imaging condition to obtain a prediction observation result; a forward imaging model is embedded into a machine learning framework, end-to-end joint optimization is carried out on multiple frames of mask patterns by utilizing a prediction observation result, continuous code value interpolation is adopted in a training stage, and a mask pattern of a corresponding quantized gray scale is derived in a deployment stage. The method is adaptive to a multi-frame compressed sampling system with time sequence modulation capability, and closed-loop collaborative optimization of mask patterns and system response is realized on the premise of not depending on specific optical structure modification.

This invention discloses an emotion monitoring system and method based on voice perception of smart safety helmets, belonging to the technical field of speech processing. The system acquires video streams from a construction site and extracts audio and visual features. These features are then fused to obtain multimodal features, which are input into an improved model to obtain a binary spectral mask. The binary spectral mask is combined with the audio features and subjected to time-frequency domain transformation to obtain clean speech features. A text sequence is generated based on these clean speech features. The clean speech features, text sequence, and video stream are input into a target model to obtain emotion tags. Work logs are retrieved, and emotion tags are added to the work logs to obtain digital archives. By fusing audio and visual features from the construction video stream, the improved model achieves speech separation under complex noise conditions, extracts work content, generates text, and accurately identifies the emotions of mask-wearing workers by combining multi-dimensional information, generating digital archives with emotion tags.

Embodiments of this application provide a sound source separation method and apparatus, and relates to the technical field of data processing. The method includes: transforming an audio signal to be separated from a time-domain signal into a time-frequency domain signal; performing frequency band segmentation on the time-frequency domain signal to segment the time-frequency domain signal into a plurality of sub band signals, where frequency bands of the plurality of sub band signals do not overlap; acquiring spectrum features of the plurality of sub band signals respectively; acquiring a spectral mask of at least one sound source of the audio signal to be separated according to the spectrum features of the plurality of sub band signals; and acquiring an audio signal of the at least one sound source according to the spectral mask of the at least one sound source and the time-frequency domain signal.