A signal abnormality detection system and a method thereof are provided. The signal abnormality detection system includes a signal sensor and a computing device. The signal sensor generates a sample signal to be tested through sensing. The computing device is signal-connected to the signal sensor to receive the sample signal to be tested, perform a correction on the sample signal to be tested, and perform a time-frequency transform on a one-dimensional signal generated after the correction to generate a two-dimensional time-frequency signal. The computing device reconstructs the two-dimensional time-frequency signal by using an abnormality detection model to calculate a reconstructed difference value. The computing device performs comparison to determine whether the reconstructed difference value is greater than a detection threshold to determine whether the sample signal to be tested is an abnormal sample.

Techniques are disclosed for the use of Crest Factor Reduction (CFR) algorithm that performs oversampling of an input signal and a cancellation pulse, and detects a set of peak samples in the upsampled input signal that exceed a predetermined threshold value. The peak samples are clustered such that a subset of the oversampled signal peaks are used to compute gain factors for the generation of a scaled truncated upsampled cancellation pulse. Several scaled truncated upsampled cancellation pulses are applied in parallel to perform peak cancellation of the highest peak in each cluster as part of an initial peak cancellation process. Any remaining peaks are canceled by iterative gain factors computation process. A final cancellation pulse is then generated by multiplying a cancellation pulse by the computed gain factors.

Techniques are disclosed for the use of Crest Factor Reduction (CFR) algorithm that performs oversampling of an input signal and a cancellation pulse, and detects a set of peak samples in the upsampled input signal that exceed a predetermined threshold value. The peak samples are clustered such that a subset of the oversampled signal peaks are used to compute gain factors for the generation of a scaled truncated upsampled cancellation pulse. Several scaled truncated upsampled cancellation pulses are applied in parallel to perform peak cancellation of the highest peak in each cluster as part of an initial peak cancellation process. Any remaining peaks are canceled by iterative gain factors computation process. A final cancellation pulse is then generated by multiplying a cancellation pulse by the computed gain factors.

A method for identifying a cutscene in gameplay footage, the method comprising: receiving a first video signal and a second video signal each comprising a plurality of images; creating a first video fingerprint comprising a plurality of signatures, each signature of the plurality of signatures based on at least one image of the plurality of images in the first video signal; creating a second video fingerprint comprising a plurality of signatures, each signature of the plurality of signatures based on at least one image of the plurality of images in the second video signal; comparing the first video fingerprint with the second video fingerprint; and identifying a cutscene when at least a portion of the first video fingerprint has at least a threshold level of similarity with at least a portion of the second video fingerprint.

Two-dimensional face presentation attacks are one of most notorious and pervasive face spoofing types, causing security issues to facial authentication systems. To tackle these issues, a cost-effective face anti-spoofing (FAS) system based on acoustic modality, named as Echo-FAS, is devised, which employs a crafted acoustic signal to probe the presented face. First, a large-scale, high-diversity, acoustic-based FAS database, named as Echo-Spoof, is built. Based upon Echo-Spoof, we design a two-branch framework combining global and local frequency features of the presented face to distinguish live vs. spoofing faces. Echo-FAS has the following merits: (1) it only needs one speaker and one microphone; (2) it can capture three-dimensional geometrical information of the presented face and achieve a remarkable FAS performance; and (3) it can be handily allied with RGB-based FAS models to mitigate the overfitting problem in the RGB modality and make the FAS model more accurate and robust.

Two-dimensional face presentation attacks are one of most notorious and pervasive face spoofing types, causing security issues to facial authentication systems. To tackle these issues, a cost-effective face anti-spoofing (FAS) system based on acoustic modality, named as Echo-FAS, is devised, which employs a crafted acoustic signal to probe the presented face. First, a large-scale, high-diversity, acoustic-based FAS database, named as Echo-Spoof, is built. Based upon Echo-Spoof, we design a two-branch framework combining global and local frequency features of the presented face to distinguish live vs. spoofing faces. Echo-FAS has the following merits: (1) it only needs one speaker and one microphone; (2) it can capture three-dimensional geometrical information of the presented face and achieve a remarkable FAS performance; and (3) it can be handily allied with RGB-based FAS models to mitigate the overfitting problem in the RGB modality and make the FAS model more accurate and robust.

A fault signal detection method includes acquiring decomposition signals according to a frequency band, by decomposing an original signal in the time domain into a certain number of frequency band signals, combining the certain number of decomposition signals and calculating a classification result value for classifying the original signal into a normal signal or a fault signal using a classification model using at least one decomposition signal included in a combination method as input and determining a combination method for detecting a fault signal based on classification result values of the classification model.

A fault signal detection method includes acquiring decomposition signals according to a frequency band, by decomposing an original signal in the time domain into a certain number of frequency band signals, combining the certain number of decomposition signals and calculating a classification result value for classifying the original signal into a normal signal or a fault signal using a classification model using at least one decomposition signal included in a combination method as input and determining a combination method for detecting a fault signal based on classification result values of the classification model.

Disclosed are a signal filtering method and apparatus, a storage medium, and an electronic device. The signal filtering method may include: performing a Fourier transform on time-domain data acquired by M microphones to convert the time-domain data into frequency-domain signals of M channels of a current frame, M is a natural number greater than 1; separating the frequency-domain signals of the M channels to obtain a target signal and an interference signal of the current frame; determining a signal-to-interference ratio between a smooth energy of the target signal and a smooth energy of the interference signal; determining a threshold for the current frame based on the target signal and the interference signal; and filtering the target signal based on the threshold for the current frame and the signal-to-interference ratio between the smooth energies to obtain a pure target signal.

An image of at least a portion of a user during enrollment to a biometric identification system is acquired and processed with a first model to determine a first embedding that is representative of features in that image in a first embedding space. The first embedding may be stored for later comparison to identify the user, while the image is not stored. A second model that uses a second embedding space may be later developed. A transformer is trained to accept as input an embedding from the first model and produce as output an embedding consistent with the second embedding space. The previously stored first embedding may be converted to a second embedding in a second embedding space using the transformer. As a result, new embedding models may be implemented without requiring storage of user images for later reprocessing with the new models or requiring re-enrollment by users.