13 results about "Audio signal classification" patented technology

Disclosed are systems and methods including software processes executed by a server that detect audio-based synthetic speech (“deepfakes”). Embodiments implement a machine-learning architecture having a diffusion model that generates purified features that are fed to a deepfake detection model. The machine-learning architecture includes input layers that convert an audio signal into a Gaussian or frequency space representation (e.g., log spectrogram) to extract a set of initial features indicative of spoofing or deepfake attacks. The diffusion model identifies adversarial noise on the audio signal in the initial features and generates purified features or clean version of the input audio signal. A deepfake detector includes a neural network architecture and classifier programmed and trained to generate a deepfake detection score and classify the audio signal as genuine or fraudulent using the purified features.

Disclosed are systems and methods including software processes executed by a server that detect audio-based synthetic speech ("deepfakes"). Embodiments implement a machine-learning architecture having a diffusion model that generates purified features that are fed to a deepfake detection model. The machine-learning architecture includes input layers that convert an audio signal into a Gaussian or frequency space representation (e.g., log spectrogram) to extract a set of initial features indicative of spoofing or deepfake attacks. The diffusion model identifies adversarial noise on the audio signal in the initial features and generates purified features or clean version of the input audio signal. A deepfake detector includes a neural network architecture and classifier programmed and trained to generate a deepfake detection score and classify the audio signal as genuine or fraudulent using the purified features.

Methods and processors for detecting a synthetic audio signal is disclosed. The method comprising: acquiring a candidate audio signal; generating a spectrogram representation of the candidate audio signal, the spectrogram representation comprising values for respective frequency-time pairs; generating a frequency-based combined patch using the spectrogram, a given combined value in the frequency-based combined patch being a combination of values from the spectrogram sharing a same frequency coordinate and different time coordinates; classifying the candidate audio signal as the synthetic audio signal using the frequency-based combined patch.

The invention provides a method for dynamically adjusting subtitles and translation modes according to audio input types. The method comprises the following steps: acquiring audio data from multiple sources and preprocessing the audio data; constructing an audio classification model based on an intelligent audio signal classification algorithm, training the audio classification model, and classifying the audio data; inputting the current audio into the audio classification model, and determining the category of the current audio; constructing a mapping relationship among the audio category, the subtitles and the translation modes, matching the corresponding subtitles and the translation modes according to the category of the current audio, and displaying the subtitles and translation contents in real time; automatically hiding related option settings based on subtitles and translation modes matched with the current audio; according to the invention, the category of the current audio can be detected in real time, the subtitle and translation mode can be dynamically adjusted according to the audio category, related option settings can be automatically hidden or displayed according to the subtitle and translation mode, unnecessary setting options are reduced, and the user experience is optimized.

The present application discloses a method, device, terminal device and storage medium for reducing wind noise. By using a variety of preset signal analysis algorithms, signal analysis is performed on the audio signal collected by the microphone, and multiple wind noise recognition identifiers are output. The microphone includes a single microphone, and based on the multiple wind noise recognition identifiers, the wind noise category of the audio signal is determined; and based on the wind noise category of the audio signal, a target wind noise filter corresponding to the wind noise category is determined, and the target wind noise filter is used to reduce the noise of the audio signal, and the target audio signal after noise reduction is output, so as to classify and identify the wind noise category of the audio signal, so as to facilitate subsequent targeted noise reduction processing of audio signals of different wind noise categories. At the same time, signal analysis is performed using a variety of preset signal analysis algorithms, which can effectively avoid missed detection or misdetection, improve the accuracy of audio signal classification and recognition, and thus improve the noise reduction effect.

The application discloses a sleep breathing disorder identification system based on deep learning, comprising an audio signal feature extraction module, a feature processing module, a time-weighted attention mechanism module and a bidirectional long short-term memory network module connected in sequence. The audio signal feature extraction module extracts multi-dimensional audio features; the feature processing module processes the features through convolution and full connection layers; the time-weighted attention mechanism module highlights key moment information; the Bi-LSTM module models forward and backward time sequence dependency relationships and is classified through a full connection layer and a Softmax layer. The device comprises a snoring sound input module, a sound detection module, an embedded module and a data display module, can monitor snoring sound signals in real time, displays the detection results in a graphical interface after processing, realizes efficient and accurate identification of various sleep breathing disorders, overcomes the problems of incomplete feature extraction and low modeling efficiency in the prior art, and improves the accuracy and robustness of the audio signal classification task.

The invention relates to the field of hearing aid audio classification, in particular to a hearing aid audio classification method, which comprises a classification system, and the classification system comprises a data acquisition and preprocessing module, a neural network model building module, a training model module, a real-time audio input module, a scene recognition stage module, a personalized adjustment module and a feedback optimization module. The data acquisition and preprocessing module comprises data preprocessing, feature extraction and feature fusion, and the data acquisition and preprocessing module collects and preprocesses a large amount of audio data of different scenes, combines a deep learning technology and a traditional audio feature extraction method, and trains a neural network model to obtain an audio feature extraction result; the audio signals received by the hearing aid are automatically classified, the accuracy and the stability are high in the aspect of audio signal classification of the hearing aid, and therefore the stability of collected audio classification when the hearing aid is used is improved, and the overall receiving definition of the hearing aid is improved.

An audio signal classification method and apparatus, where the method includes determining, according to voice activity of a current audio frame, whether to obtain a frequency spectrum fluctuation of the current audio frame and store the frequency spectrum fluctuation in a frequency spectrum fluctuation memory, and updating, according to whether the audio frame is percussive music or activity of a historical audio frame, frequency spectrum fluctuations stored in the frequency spectrum fluctuation memory, and classifying the current audio frame as a speech frame or a music frame according to statistics of a part or all of effective data of the frequency spectrum fluctuations stored in the frequency spectrum fluctuation memory.

An audio signal classification method is provided, where the method includes: determining, according to voice activity of a current audio frame, whether to obtain a frequency spectrum fluctuation of the current audio frame and store the frequency spectrum fluctuation in a frequency spectrum fluctuation memory (101); updating, according to whether the audio frame is percussive music or activity of a historical audio frame, frequency spectrum fluctuations stored in the frequency spectrum fluctuation memory (102); and classifying the current audio frame as a speech frame or a music frame according to statistics of a part or all of effective data of the frequency spectrum fluctuations stored in the frequency spectrum fluctuation memory (103). An audio signal classification apparatus is further provided.

This utility model discloses a device for detecting thermal events in shipborne new energy vehicles, comprising: a headlight flicker detection sub-device, an alarm horn detection sub-device, and a new energy vehicle thermal event alarm sub-device; a camera group is deployed to ensure the acquisition of video images of all new energy vehicles; a headlight flicker detector detects headlight flicker events, avoiding misjudgments due to human negligence or fatigue; a microphone group is responsible for collecting sound from inside the vehicle compartment of the roll-on / roll-off ship; an audio signal classification and recognition device automatically identifies whether the audio data contains a car horn event; a signal superposition device superimposes the high-level signals detected by the headlight flicker detector and the audio signal classifier, and a thermal event judge determines whether a new energy vehicle thermal event has occurred based on the superimposed voltage signal, thereby significantly improving the objectivity and reliability of vehicle thermal event determination.

Methods for echo estimation or echo management (echo suppression or cancellation) on an input audio signal, with at least one of adaptation of a sparse prediction filter set, modification (for example, truncation) of adapted prediction filter impulse responses, generation of a composite impulse response from adapted prediction filter impulse responses, or use of echo estimation and / or echo management resources in a manner determined at least in part by classification of the input audio signal as being (or not being) echo free. Other aspects are systems configured to perform any embodiment of any of the methods.

Methods and processors for detecting a synthetic audio signal is disclosed. The method comprising: acquiring a candidate audio signal including a plurality of audio values; generating a set of patches using the plurality of audio values, a given one from the set of patches including a pre-determined number of audio values from the plurality of audio values, generating a combined patch using the set of patches, a given one value in the combined patch being a combination of values from respective ones from the set of patches; classifying the candidate audio signal as the synthetic audio signal using the combined patch.

Methods and systems for modification of electronic system operation based on acoustic ambience classification are presented. In an example method, at least one audio signal present in a physical environment of a user is detected. The at least one audio signal is analyzed to extract at least one audio feature from the audio signal. The audio signal is classified based on the audio feature to produce at least one classification of the audio signal. Operation of an electronic system interacting with the user in the physical environment is modified based on the classification of the audio signal.