28 results about "Sound separation" patented technology

A computer-implemented method of applying a trained neural network for sound separation based on distance estimation is provided. The method includes receiving, by an audio input component of a computing device, an audio mixture from one or more sources. The method includes predicting, by a trained distance estimation neural network and based on the audio mixture, respective distances of the one or more sources from the audio input component. The method includes determining one or more near sounds and one or more far sounds based on the respective distances. The near sounds correspond to sources that are located within a threshold distance of the audio input component, and the far sounds correspond to sources that are not located within the threshold distance of the audio input component. The method includes providing the predicted one or more near sounds.

The invention relates to the field of intelligent environment control, in particular to a method and system for controlling operation of air conditioning equipment and a storage medium. The method comprises the following steps: acquiring external sound; based on a voiceprint library of the air conditioning equipment, air conditioning equipment noise and environment noise are separated from the external sound in a frequency domain; setting, for each frequency in the spectrum of the external sound, a weight indicating a corresponding tolerance of noise at that frequency; on the basis of the set weight, whether the noise of the air conditioning equipment at the frequency affects the user experience or not is determined for each frequency; and when the noise of the air conditioning equipment at any frequency affects the user experience, operation of the air conditioning equipment is adjusted.

The present invention addresses the problem of providing an "audio signal processing device" that achieves a good sense of localization with respect to a sound having a narrow frequency band. [Solution] The present invention is provided with signal processing units provided so as to correspond to respective channels of audio signals of a plurality of channels, and each signal processing unit has: a target sound separation unit (411) that separates a target sound signal indicating a target sound, which is a sound to be subjected to localization enhancement, from a source sound signal, which is an audio signal of the corresponding channel; a sense-of-localization-enhanced sound generation unit that generates a sense-of-localization-enhanced sound signal in which the frequency of the target sound signal separated by the target sound separation unit (411) is multiplied by k (k is an integer of 2 or more); and a synthesis unit that synthesizes and outputs the localization-enhanced sound signal generated by the localization-enhanced sound generation unit and the source sound signal.

The present application relates to the field of audio processing, in particular to a sound separation method and system based on a vector microphone, the method comprising: constructing a fourth-order cumulant matrix of a sound signal received by the vector microphone, solving sound source direction angle estimation and sound source pitch angle estimation of each incident signal; constructing a spatial characteristic matrix of each incident signal at different frequencies; based on the spatial characteristic matrix, calculating a spatial correlation matrix estimation of the sound signal at different frequencies and different frame numbers, and constructing a spatial correlation matrix of the sound signal at different frequencies and different frame numbers; constructing an orthogonal MNMF model; adjusting the variables of the MNMF through the multiplication update method, so that the difference between the spatial correlation matrix estimation and the spatial correlation matrix is less than a preset target, to obtain an optimal spatial correlation matrix estimation; and separating the sound signal based on the optimal spatial correlation matrix. The present application effectively solves the sound separation problem when the distance between sound sources is close, and improves the accuracy of sound separation.

The invention belongs to the field of sound separation processing, and discloses an individual perception multi-bird sound separation method and system based on identity embedding, and the method comprises the steps: S1, carrying out the preprocessing of an audio segment containing bird sound, obtaining a preprocessed audio segment, and obtaining a time domain waveform corresponding to the audio segment; s2, carrying out separation processing on the time domain waveform to obtain a separated individual audio waveform; and S3, based on the separated individual audio waveform, carrying out classification identification on the bird sound. According to the method, under the conditions that voiceprint differences of individuals of the same type of birds are extremely small and sound sources are highly overlapped, stable individual consistency and separation precision can still be kept. According to the method, the technical bottleneck of identity drift and pseudo separation of a traditional model in long-time sequence monitoring is solved, refinement and automation of individual-level ecological acoustic analysis are also realized, and a feasible technical path is provided for field long-term ecological monitoring and acoustic big data processing.

The application discloses a mechanical arm deception attack side channel detection method based on acoustic characteristics and belongs to the technical field of deep learning, which comprises the following steps: constructing a training recognition model; receiving the collected target audio, performing a sound processing step operation to extract features, performing a training recognition step operation to output predicted motion data; comparing the predicted motion data with the obtained corresponding real-time motion data, and evaluating whether the difference between each pair of parameters exceeds a preset threshold value. The application uses sound separation technology to separate mixed multi-axis sound into different individual axis sound, constructs an acoustic motion information recognition model of each motion joint through a collaborative physical and data-driven method, identifies the target sound according to the recognition model, obtains the predicted motion information of each axis, compares the predicted motion information with the collected real-time motion command, combines a threshold value to determine whether a deception attack is received, and realizes mechanical arm deception attack side channel detection.

Systems and methods for audio source separation. A deep learning-based system uses azimuth angle position to separate an audio signal originating from a selected location from other sounds. Techniques for directing a virtual microphone towards a selected speaker are revealed. A deep learning-based audio regression method, which can be implemented as a neural network, learns to separate different speakers by effectively utilizing the spectral and spatial properties of all sources. The neural network can focus on multiple sources in multiple respective target directions and filter out other sounds. A user can select which source to hear. The network can use the time-domain signal and a frequency-domain signal to separate the target signal and produce separate audio outputs.The direction of the selected speaker relative to the microphone array can be entered into the system as a vector.

The application discloses a sound isolation method and device based on deep learning and a storage medium. The method comprises the following steps: obtaining an audio file used for constructing a DeepAudioSep model and preprocessing the audio file used for constructing the DeepAudioSep model; constructing the DeepAudioSep model and training the DeepAudioSep model, wherein the DeepAudioSep model comprises one mixed source input and ten isolated source outputs; and performing sound separation through the DeepAudioSep model. The application introduces the data driving and deep learning idea into sound separation and noise isolation processing, improves the sound separation and noise isolation processing capability in the environmental monitoring field, and therefore has wide noise processing prospects and practical value.

We provide an "audio signal processing device" that achieves good localization for sounds with a narrow frequency band. [Solution] The system has a signal processing unit corresponding to each channel of a multi-channel audio signal. Each signal processing unit includes a target sound separation unit 411 that separates a target sound signal, which is the sound to be targeted for localization enhancement, from a source sound signal, which is the audio signal of the corresponding channel; a localization enhancement sound generation unit that generates a localization enhancement sound signal by multiplying the frequency of the target sound signal separated by the target sound separation unit 411 by k (where k is an integer of 2 or more); and a synthesis unit that synthesizes the localization enhancement sound signal generated by the localization enhancement sound generation unit and the source sound signal and outputs the result.

The present application relates to a kind of adaptive band optimization chorus sound separation method and system based on harmonic maintenance, first, the signal is collected by microphone array and time delay compensation and multi-channel time-frequency transformation are carried out;Sound source spatial positioning and initial band division are carried out using MUSIC algorithm;Using the greedy optimization strategy, the band boundary is dynamically adjusted and optimized by maximizing the objective function including signal-to-noise ratio and harmonic integrity measure;Then, the fundamental frequency is obtained by harmonic analysis and detection, and the time-varying harmonic trajectory is tracked across frames by Viterbi algorithm;On this basis, combined with MVDR beam forming and the introduction of harmonic maintenance factor, enhanced time-frequency representation is obtained by soft time-frequency mask;Finally, inverse short-time Fourier transform time-domain reconstruction and synchronous harmonic phase are carried out.The present application realizes the accurate allocation of voice part spectrum, effectively reduces the overlap interference, protects the overtone column of human voice while accurately separating, significantly improves the timbre fidelity and naturalness of separated voice part.

The invention relates to the field of air conduction hearing aids, in particular to an air conduction hearing aid coincidence sound separation method, which comprises a hearing aid module, a coincidence sound processing module, a telescopic module, a sound insulation module and an auxiliary hearing aid module, and is characterized in that the hearing aid module comprises an external ear fixing assembly, a control assembly and a broadcaster assembly; the coincident sound processing module comprises a processing chip, a coincident sound separating module and a coincident sound playing module; the telescopic module comprises a storage assembly and a telescopic assembly, the sound transmitter assembly is controlled to extend into the ear canal, the storage assembly is opened, the inflation assembly is controlled to expand the inflation air bag to isolate external sound, the bone conduction assembly is driven to abut against the inner ear canal, and then coincident sound is separated through the processing chip; and then the positive sound part is played through the separated broadcaster assembly, and the bone conduction assembly plays the coincident sound part, so that the situation that the same player plays too disorderly is avoided, and then the situation that information in the coincident sound is missed under the normal use condition can be avoided.

The invention discloses a passive respiratory disease early warning device based on respiratory audio spectrum feature separation and a bidirectional long short-term memory network, which is characterized in that signal preprocessing is triggered when a posture is static by continuously collecting respiratory sound and synchronizing body position and motion state data; cardiopulmonary sounds are separated in real time by adopting an online non-negative matrix factorization technology, pure respiratory signals are extracted, and a dynamic signal environment is adapted through an incremental updating mechanism; key frequency band features are extracted in combination with wavelet transform time-frequency analysis, and breathing micro-variation features such as expiratory phase extension and wet rale are quantized by using a dynamic time bending algorithm; a two-way LSTM fused with an external attention mechanism is introduced to model breathing time sequence characteristics, key frequency domain information is focused, abnormal probabilities of pneumonia, asthma and chronic obstructive pulmonary disease are output, and when a preset early warning condition is met, a prompt is triggered; respiration feature extraction and desensitization are completed through local edge calculation, and only abnormal fragment ciphertexts are uploaded to guarantee data privacy. The method solves the problems that in the prior art, breathing sound separation precision is insufficient, single-mode analysis is limited, time sequence modeling adaptability is poor and the like, and zero-intervention and high-precision early passive monitoring of respiratory system diseases is achieved.

This invention discloses a multi-target sleep monitoring method and device based on snoring separation. The method includes: constructing independent respiratory wave time-stamped sequences corresponding to each monitored object based on the radio frequency signal corresponding to the target monitoring area; constructing a mixed snoring time-stamped sequence corresponding to all monitored objects based on the mixed environmental audio signal corresponding to the target monitoring area; performing cross-modal correlation matching between the mixed snoring time-stamped sequence and each independent respiratory wave time-stamped sequence to determine the independent snoring time-stamped sequence corresponding to each monitored object; and performing sleep monitoring analysis on the corresponding monitored objects based on the independent respiratory wave time-stamped sequence and / or the independent snoring time-stamped sequence of the same monitored object to obtain the sleep monitoring results corresponding to each monitored object. This achieves accurate sleep monitoring analysis of multiple monitored objects in the same space, resulting in precise sleep monitoring analysis results.

The invention relates to the field of hearing aids, in particular to an air conduction hearing aid far and near sound separation method, which comprises a hearing aid body module, a sound collection module, a control module, a broadcasting module, a vibration module and an isolation module, and is characterized in that the hearing aid body module comprises a body module and an ear hook module; the sound collection module comprises a stretching module and a sound receiver; the control module comprises a key module, a processing chip, a separation module and a transmission module; the hearing aid is worn, air is injected into the air bag, the air bag is expanded to isolate the upper broadcasting module and the lower broadcasting module and push the upper broadcasting module and the lower broadcasting module to abut against the ear canal, then the sound receiver extends out of the hearing aid body to enhance the receiving effect, and far and near sounds are separated and transmitted into the upper broadcasting module and the lower broadcasting module respectively. And the upper broadcasting module and the lower broadcasting module can synchronously start the upper vibration module and the lower vibration module during playing, so that a user can clearly hear the playing of far and near sounds and identify the far and near sounds.

According to an aspect of the present disclosure there is provided a method for adversarial training of a separator (30) for universal sound separation of an audio mixture m of arbitrary sound sources Sk=1, . . . ,K, the method comprising: training a context-based discriminator (34) configured to provide a context-based loss cue based on a consideration of an input set of separated sound sources; and training the separator (30) to minimize a loss based on the context-based loss cue provided by the context-based discriminator (34); wherein training the context-based discriminator (34) comprises maximizing a loss based on a set of ground-truth sound sources and a fake set of separated sound sources, wherein the fake set of separated sound sources is sorted to match an order of the set of ground-truth sound sources, and wherein the fake set of separated sound sources comprises sources corresponding to separated sound sources estimated by the separator (30) and further comprises one or more ground-truth sound sources of the set of ground-truth sound sources.

The invention discloses a dry and wet sound separation method and a related device, and relates to the technical field of audio processing, the dry and wet sound state is introduced, and the signal state is determined, so that the separation strategy better fits the dynamic change of the signal. On the basis, when dry and wet sound separation is realized, a spatio-temporal information decoupling technology is adopted, spectrum information which changes rapidly and relatively stable spatial information are respectively processed and cooperatively utilized, interference of spatial aliasing on spectrum separation is avoided, a separation result is verified and optimized through spatial correlation, the problem that stereo spatial information is not fully utilized is solved, and the separation efficiency is improved. Therefore, the spatial information of stereophonic sound is fully mined, and the overall separation effect is improved.

The invention provides a vehicle external sound processing method and system and a storage device. The method comprises the steps that original environment sound signals in multiple directions outside a vehicle are collected; based on a target sound pointing instruction, performing beam forming processing on the original environment sound signal, and extracting a first sound signal in a specific direction; performing noise suppression and target sound separation processing on the first sound signal to obtain a second sound signal; and performing high-fidelity audio reconstruction and amplification on the second sound signal to obtain an audio signal of the target sound, and outputting the audio signal to an in-vehicle loudspeaker. According to the method provided by the invention, directional screening and noise reduction of the environment sound outside the vehicle and high-fidelity playback in the cabin can be realized.

The present application relates to the technical field of sound processing, and provides a sound separation method, device, electronic equipment and storage medium, the method comprising: obtaining a first sound signal and a second sound signal; determining the short-time energy value of the ambient sound and the coherent sound respectively, and the difference factor of the coherent sound on different channels, based on the short-time energy value of the first sound signal and the second sound signal respectively, and the linear combination relationship between the ambient sound and the coherent sound contained in the first sound signal and the second sound signal respectively; mapping the short-time energy value of the ambient sound and the coherent sound respectively, and the difference factor, to the sound separation weight based on the weight mapping relationship, the weight mapping relationship being obtained by linear fitting based on the linear combination relationship; separating the ambient sound and the coherent sound from the first sound signal and the second sound signal based on the sound separation weight. The method, device, electronic equipment and storage medium provided by the present application ensure the effectiveness and reliability of the separation of the ambient sound and the coherent sound.

The application relates to a voice consultation device, a voice consultation method and a storage medium, and belongs to the technical field of computers.The device comprises a microphone module, an AEC circuit, a camera and a processor; the AEC circuit separates the sound of the loudspeaker of the voice consultation device, and does not send the sound to voice recognition; the voice consultation device can improve the clarity of the audio device collected from the source; meanwhile, the angle of the user facing the microphone is analyzed through a beamforming algorithm, so that the voice signals outside the angle are suppressed, the multichannel human voice signals inside the angle are integrated into single-channel audio first, and the single-channel audio is sent to voice recognition, the clarity of the audio data can be further improved, and the accuracy of voice recognition is improved. In addition, the distance between the user and the voice consultation device is determined through the camera, and the recognition mode corresponding to the distance is automatically switched, so that the accuracy of voice recognition is improved, and the response accuracy of the voice consultation device is improved.

The invention provides a stereophonic sound processing method and device, and relates to the technical field of sound processing, and the method comprises the steps: carrying out the dry and wet sound separation processing of a stereophonic sound audio signal comprising a left sound channel signal and a right sound channel signal, and obtaining a dry sound, a left channel wet sound, and a right channel wet sound; wherein the left channel wet sound is a signal component irrelevant to a right channel signal in the left channel signal, the right channel wet sound is a signal component irrelevant to the left channel signal in the right channel signal, and the dry sound is a signal component with the same amplitude and phase in the left channel signal and the right channel signal; performing filtering processing on the dry sound, the left channel wet sound and the right channel wet sound by using a preset first control filter, a preset second control filter and a preset third control filter respectively to obtain three paths of corresponding filtering signals; and superposing the three paths of filtering signals to generate a multi-channel output signal for driving the loudspeaker array.

The described aspects and implementations enable privacy-respecting detection, separation, and localization of sounds in vehicle environments. The techniques include obtaining, using audio detector(s) of a vehicle, a sound recording that includes a plurality of elemental sounds (ESs) in a driving environment of the vehicle, and processing, using a sound separation model, the sound recording to separate individual ESs of the plurality of ESs. The techniques further include identifying a content of individual ESs and causing a driving path of the vehicle to be modified in view of the identified content of the individual ESs. Further techniques include rendering speech imperceptibly by redacting temporal portions of the speech, using sound recognition models to identify and discard recordings of speech, and driving at speeds that exceed threshold speeds at which speech becomes imperceptible from noise masking.

The embodiment of the invention is applied to the field of artificial intelligence, and provides an audio processing method, a sound separation model training method and electronic equipment, and the method comprises the steps: carrying out the fusion of M collected audios collected by M microphones in a microphone array, and obtaining a fused audio, M being an integer greater than 1; performing multiple denoising processing on the fused audio by using the sound separation model to obtain a target audio, the target audio representing a sound of which a sound source is located in a target area, the input of the sound separation model for performing the first denoising processing being the fused audio, the input of the sound separation model for other denoising processing is the output audio obtained by the last denoising processing, and the output audio obtained by the sound separation model for the last denoising processing comprises the target audio. Based on the technical method provided by the invention, the sound of which the sound source is located in the target area can be accurately separated based on the audio collected by the plurality of microphones.

The invention discloses an audio reverberation space clue parameter estimation method and device and a storage medium, and relates to the technical field of audio processing. The method comprises the following steps: acquiring a target wet sound signal, and carrying out dry and wet sound separation on the target wet sound signal to obtain a dry sound signal and a pure reverberation signal; obtaining parameter estimation rules corresponding to a plurality of target audio space cue parameters used for representing reverberation space characteristics; the parameter estimation rule is pre-constructed based on the correlation between the signal statistical characteristic of the audio signal and the audio space clue parameter; and according to the parameter estimation rule, analyzing the target wet sound signal, the dry sound signal and the pure reverberation signal to obtain a target audio space clue parameter of the target wet sound signal. Automatic audio reverberation space clue parameter estimation can be realized, the audio space clue parameter estimation efficiency and accuracy are improved, and the method is suitable for linear and nonlinear reverberation effects.

Systems and methods can utilize a conformer model to process a data set for various data processing tasks, including, but not limited to, speech recognition, sound separation, protein synthesis determination, video or other image set analysis, and natural language processing. The conformer model can use feed-forward blocks, a self-attention block, and a convolution block to process data to learn global interactions and relative-offset-based local correlations of the input data.

Provided is an acoustic signal analysis system and the like that can accurately separate a sound into its individual components even if the components of the constituent units overlap in time. [Solution] An approximation calculation unit 31 performs nonnegative matrix factorization on a spectrogram matrix obtained by short-time Fourier transform of an acoustic signal of a speech sound to estimate a basis matrix indicating a spectral pattern and an activation matrix indicating time changes in signal strength. A normalization unit 32 normalizes the basis matrix using the Euclidean norm of the basis matrix and scales the activation matrix using the Euclidean norm. A separation unit 33 separates the acoustic signal into basis components based on the normalized basis matrix and the scaled activation matrix.

The described aspects and implementations enable privacy-respecting detection, separation, and localization of sounds in vehicle environments. The techniques include obtaining, using audio detector(s) of a vehicle, a sound recording that includes a plurality of elemental sounds (ESs) in a driving environment of the vehicle, and processing, using a sound separation model, the sound recording to separate individual ESs of the plurality of ESs. The techniques further include identifying a content of individual ESs and causing a driving path of the vehicle to be modified in view of the identified content of the individual ESs. Further techniques include rendering speech imperceptibly by redacting temporal portions of the speech, using sound recognition models to identify and discard recordings of speech, and driving at speeds that exceed threshold speeds at which speech becomes imperceptible from noise masking.