75 results about "Noise classification" patented technology

A communication device includes a noise flattening filter having a filter response that dynamically adjusts based on the current noise spectrum in a wireless channel. The noise spectrum of the wireless channel is estimated and used to determine a noise classification for the channel. A noise flattening filter response is then selected based upon the noise classification for use in filtering signals received from the channel. The filtered signals are then delivered to an equalizer for further processing.

The invention discloses a noise classification method of Gaussian Mixture Model based on a neural network. The recognition rate of noise classification can be improved by using the method. When the method is training, the extracted eigenvector acts as the input of auto-associative neural network (AANN) after being delayed, the structure of the eigenvector is studied by AANN, and the temporal information of the eigenvector sequence is extracted. Then the study result is provided to Gaussian Mixture Models (GMM) in the form of residual eigenvector. The GMM training is carried out by using Expectation Maximization (EM). The weight coefficient of AANN is updated using the method of backward inversion with inertia. The noise classification method makes full use of the advantages of AANN and GMM and greatly improves the recognition rate of the whole noise classification system.

The invention relates to an environment noise identification classification method based on a convolutional neural network. The method comprises the following steps of: S1, extracting natural environment noise, and editing the natural environment noise into noise segments with duration of 300ms to 30s and a converted frequency of 44.1kHz; S2, carrying out short time Fourier transformation on the noise segments, and converting a one-dimensional time-domain signal into a two-dimensional time-domain signal to obtain a sonagraph; S3, extracting a MFCC (Mel Frequency Cepstrum Coefficient) of the signal; S4, forming a training set with 80% of all the noise segments and forming a testing set with the residual 20% of all the noise segments; S5, carrying out noise classification by a convolutionalneural network model; and S6, training a classification model by the training set, and verifying accuracy of the model by the testing set so as to complete environment noise identification classification based on the convolutional neural network. According to the invention, the sound segments are input, sound feature information is extracted, an output is a classification result, and automatic extraction on the sound feature information can be implemented.

The invention relates to the technical field of intelligent voice signal processing, mode recognition and artificial intelligence and in particular relates to a record replay attack detection method and system in a speaker recognition system based on a channel mode noise. The invention discloses a simpler and more efficient record replay attack detection method in a speaker recognition system. The method comprises the following steps: (1) inputting a to-be-recognized voice signal; (2) pre-processing the voice signal; (3) extracting the channel mode noise in the pre-processed voice signal; (4) extracting a long time statistic feature based on the channel mode noise; and (5) classifying the long time statistic feature according to a channel noise classifying judging model. By using the channel mode noise to perform the record replay attack detection, the extracted feature dimension is low, the computation complexity is low, and the recognition error rate is low, therefore, the safety performance of the speaker recognition system is greatly improved, and the method and system provided by the invention can be used in the reality more easily.

The invention discloses a digital image noise reduction method and device. The noise reduction method includes: performing noise analysis on an input image, calculating the difference frame / field image between input image frames or between two fields separated by one field, and based on the difference frame Image analysis of the frame / field noise statistical characteristics; perform noise analysis, calculate noise power, determine noise signs and determine noise classification; adaptively select frame / intra-field noise reduction or / and frame / inter-field noise reduction methods. The device for digital image noise reduction includes: frame memory, noise analysis unit and frame / field noise reduction unit. Since the present invention selects and adopts frame / intra-field and / or frame / inter-field noise reduction methods by analyzing the noise characteristics of difference frame / field images, and selects appropriate weights according to the characteristics of noise and image information, the noise reduction Finally, images with better visual effects are obtained.

The claimed invention is directed to a real-time noise classification and tuning system for cochlear implant and other hearing device applications. The system is capable of automatically selecting the optimized parameters of a noise suppression algorithm in response to different noisy environments. The feature vector and the classifier deployed in the system to automatically identify the background noise environment are selected so that the computation burden is kept low to achieve a real-time throughput. The results reported herein indicate improvement in speech enhancement when using this intelligent real-time hearing device system.

Sound discrimination techniques are disclosed that can be employed, for example, in the task of discriminating speech and non-speech in a noisy environment and other noise classification applications. In one particular embodiment, a classifier system is provided that includes a linear Regularized Least Squares classifier used directly on a high-dimensional cortical representation of the target sound. The regularization constant lambda (λ) can be selected automatically, yielding a parameter-free learning system. In addition, the high-dimensional hyperplane can be viewed directly in the cortical space, leading to greater interpretability of the classifier results.

Loudness control systems or methods may normalize audio signals to a predetermined loudness level. If the audio signal includes moderate background noise, then the background noise may also be normalized to the target loudness level. Noise signals may be detected using content-versus-noise classification, and a loudness control system or method may be adjusted based on the detection of noise. Noise signals may be detected by signal analysis in the frequency domain or in the time domain. Loudness control systems may also produce undesirable audio effects when content shifts from a high overall loudness level to a lower overall loudness level. Such loudness drops may be detected, and the loudness control system may be adjusted to minimize the undesirable effects during the transition between loudness levels.

A method for evaluating the opto-coupler storage life based on low-frequency noise classification comprises the following steps that firstly, a low-frequency noise adapter is arranged according to the inner structure of each opto-coupler; secondly, original low-frequency noise of each opto-coupler is measured; thirdly, samples are classified according to the low-frequency noise; fourthly, the temperature of a accelerated life test is determined and the accelerated life test is conducted on the samples in a classified mode; fifthly, a function of a degeneration trend is determined; sixth, the opto-coupler storage life at the normal temperature is extrapolated through a high-temperature storage result. According to the method for evaluating the opto-coupler storage life based on low-frequency noise classification, the opto-couplers of the same batch are classified, service life evaluation is respectively conducted on the opto-couplers, a thought allowing the service lives of the opto-couplers to be ranked and classified before the accelerated life test is conducted is provided, an effective way is offered for the current research in the aspect of storage life evaluation of a semiconductor circuit, the opto-couplers with the similar service life are in the same classification, and therefore the accuracy of opto-coupler service life evaluation is improved.

The invention discloses an audio ground electromagnetic signal de-noising method based on signal-noise classification. The method comprises the following steps: extracting audio ground electromagneticsignal samples; respectively computing unevenness and irregularity of multifractal spectrum of each audio ground electromagnetic signal sample; training a preset classification model by utilizing theunevenness and irregularity of multifractal spectrum of each audio ground electromagnetic signal sample and a classification value of each audio ground electromagnetic signal sample; performing classification on a to-be-prepared measured audio ground electromagnetic signal according to a signal-noise classification mathematical model to obtain a signal segment not in strong interference and the signal segment in strong interference; performing matching tracing de-noising processing on the electromagnetic signal segment in strong interference; and combining the signal segment after de-noisingprocessing and the signal segment not in strong interference to obtain a reconstructed audio ground electromagnetic useful signal. Through the above method disclosed by the invention, the audio groundelectromagnetic signal with higher quality can be acquired, the condition that the signal not in the strong interference is filtered can be effectively avoided, and the de-noising precision is improved.

The invention provides a noise classifying method based on a convolutional neural network. The noise classifying method comprises the steps that input noise signals are subjected to framing and windowing, and the window length is 10 ms-30 ms; frequency domain characteristics and time domain characteristics are extracted from each frame of noise signals subjected to framing and windowing, and thusa two-dimensional matrix with the size being 12*24 is constituted; the convolutional neural network is established and constituted by an input layer, a hidden layer, a full-connection layer and an output layer, wherein the hidden layer is constituted by one or more convolution layers and one or more pooling layers which are arranged at intervals; the convolutional neural network is trained; and the frequency domain characteristics and time domain characteristics of each frame of noise signals are input into the trained convolutional neural network to obtain a classifying result. According to the noise classifying method based on the convolutional neural network, the accuracy rate of noise classifying can be effectively increased, the size of the input characteristics is only 12*24, and thecomputation complexity of the convolutional neural network is effectively lowered.

The present invention discloses an intelligent hearing aid method and system, and relates to the technical field of computers. One embodiment of the method comprises the steps of: obtaining collectedvoice data, and employing frequency domain to eliminate echo; according to the eliminated echo voice data, extracting the noise data to perform noise classification learning to obtain a noise classification result; and according to the noise classification result, performing noise inhibition processing for the eliminated echo voice data to obtain the voice data after noise reduction and play the voice data after noise reduction. The problems are solved that the hearing aids are poor in use experience and high in cost in the prior art.

A voice identification method and an apparatus thereof are disclosed. The voice identification method comprises the following steps of processing collected input sound signals and acquiring input sound data, wherein the input sound data includes a noise portion and a voice portion; carrying out characteristic extraction on the noise portion and the voice portion so as to acquire a noise characteristic vector corresponding to the noise portion and a voice characteristic vector corresponding to the voice portion; inputting the extracted noise characteristic vector into a preset noise classification model and determining a noise type which the noise portion belongs to; selecting a voice identification model associated to the determined noise type; calculating a similarity value between the voice characteristic vector and the voice identification model associated to the determined noise type; and when the similarity value is determined to be greater than a preset value threshold, identifying input sound data. By using the above scheme, a speed and real-time performance of voice identification can be increased.

A method (200) of cancelling echo in a duplex communication device (100). The method can include detecting a level of noise present on an uplink signal path (104), generating a noise classifier (194) based on the detected level of noise, detecting whether uplink audio is present on the uplink signal path (104) and detecting whether downlink audio is present on a downlink signal path (102). The method further can include generating a double talk flag (136) based at least on the noise classifier, whether uplink audio is present on the uplink signal path, and whether downlink audio is present on the downlink signal path. In addition, the double talk flag, the noise classifier and an uplink signal can be processed to generate an output signal (120) having reduced echo.

The invention discloses the identification method and system of an impulsive noise in a power line communication system. The method comprises the following steps of establishing a signal sequence; establishing a signal change sequence; calculating the average power and the mean square error of the active power of a signal in the signal change sequence; calculating an impulse noise trigger threshold, a first noise classification threshold, and a second noise classification threshold; comparing the active power of the signal in the signal change sequence with the impulse noise trigger threshold;comparing a first noise classification determination value with the first noise classification threshold; comparing a second noise classification determination value with the second noise classification threshold; and outputting a moment when a first impulse noise is generated and a moment when a second impulse noise is generated. By using the method and the system of the invention, a long-pulsenoise and a short-pulse noise can be directly and accurately identified and extracted from power line communication modulation signals, and mutual influences among the signals can be well avoided.

The invention provides a voice enhancement method based on DNN noise classification, and relates to the technical field of voice recognition. The method comprises the following steps: carrying out preprocessing on noise signals, and determining signals of the non-voice segment; for each frame of voice signals, carrying out fast Fourier transform, and calculating the spectral line energy; enablingthe frequency spectrum of the voice signals to pass through a Mel filter group, multiplying the frequency spectrum energy with the frequency response of the Mel filters, thus obtaining the Mel filtering energy; carrying out discrete cosine transform on the logarithm of the Mel filtering energy of each frame of signals, thus obtaining a Mel cepstrum parameter which serves as the feature vector of each frame of Mel filtering of the voice; taking the feature vector of each frame of Mel filtering as a 24-dimensional vector, and also as the input of the deep neural network; and carrying out training and classifying on the noises by utilizing the deep neural network. For the voice enhancement method based on DNN noise classification, through the classification for noises, the subsequent voice enhancement quality is obviously improved in the subjective / object testing.

The present technology provides robust, high quality expansion of the speech within a narrow bandwidth acoustic signal which can overcome or substantially alleviate problems associated with expanding the bandwidth of the noise within the acoustic signal. The present technology carries out a multi-faceted analysis to accurately identify noise within the narrow bandwidth acoustic signal. Noise classification information regarding the noise within the narrow bandwidth acoustic signal is used to determine whether to expand the bandwidth of the narrow bandwidth acoustic signal. By expanding the bandwidth based on the noise classification information, the present technology can expand the speech bandwidth of the narrow bandwidth acoustic signal and prevent or limit the bandwidth expansion of the noise.

The invention discloses an active voice detection method based on noise scene recognition, and the method comprises the steps: extracting an optimal feature facing a noise classification task from anaudio signal, inputting a feature value into a noise type classifier, so as to recognize a noise type in the audio signal, according to the noise type, determining a preferred feature and a classifiersuitable for the voice and noise classification task, extracting optimal features oriented to the voice and noise classification task from the audio signals, inputting the optimal features into a voice and noise classifier, and judging whether the voice signals exist in the audio signals or not. According to activity voice detection system based on noise scene recognition, before binary classification of noisy voice and noise signals is carried out, according to the method, the current noise type is detected and identified, the most distinctive feature combination is preferably selected for the specific noise type, model parameters can be designed for the specific noise type, and the effectiveness and stability of the performance of the whole detection process under different noise typesare ensured.

The invention discloses a noise classification method based on a BP network. The method comprises the following steps that an input noise signal is preprocessed; each frame of preprocessed noise signal is subjected to Fourier transform to obtain a noise signal power spectrum; a Mel-frequency cepstral coefficient of each frame of noise signal and first-order difference of the Mel-frequency cepstralcoefficient are calculated by utilizing the power spectrum of each frame of noise signal; a Gammatone frequency cepstral coefficient of each frame of noise signal is calculated; the Mel-frequency cepstral coefficient, the first-order difference of the Mel-frequency cepstral coefficient and the Gammatone frequency cepstral coefficient of each frame of noise signal are combined to serve as a jointfeature of the frame of noise signal, a part of joint features of all frames of noise signal is used as training data, and the other part is used as test data; a primary BP network and a secondary BPnetwork are trained; and the primary BP network and the secondary BP network are jointly tested to obtain a final noise signal classification result. The method has the higher noise classification accuracy.

The invention discloses an iris image noise classification detection method, which comprises steps: S1, light source image point noise in an iris area is detected; S2, eyelid noise in the iris area is detected; S3, eyelash noise in the iris area is detected; and S4, based on the light source image point noise information, the eyelid noise information and the eyelash noise information detected and obtained in steps from the first to the third, all noise points in the iris area are marked, and the pixel point is eliminated. The technical scheme of the invention can effectively and accurately eliminate interference on iris positioning and feature extraction by the noise.

The invention discloses a voice enhancement method based on a noise classification optimization IMCRA algorithm. The method comprises the following steps of: searching an optimal alpha s, alpha d andalpha parameter combination for different noise types, alpha s and alpha d being two smooth parameters of the IMCRA algorithm during the estimation of a noise power spectrum, and alpha being a weightparameter of priori signal-to-noise ratio estimation; classifying noise types of input signals by using a convolutional neural network CNN; according to different noise types, selecting different parameter groups alpha s, alpha d and alpha according to the optimization result in the step 1; and according to the parameter group selected in the step 3, carrying out voice enhancement on a test set signal by using an IMCRA method to obtain a finally enhanced voice. The method has a better voice enhancement effect, and the quality and intelligibility of the enhanced voice are improved.

The invention discloses a point cloud data Poisson curved surface reconstruction method based on noise classification and MLS, and the method comprises the steps: classifying noise points and introducing the noise points to vector field estimation of a point cloud data isosurface for corresponding smoothing, forming a new sample point, and precisely calculating and correcting a point cloud data normal vector by using moving least squares (MLS), carrying out the surface reconstruction to form a three-dimensional surface with the abundant details, wherein the three-dimensional surface visually approaches to an actual model in a better way. The method effectively achieves the smoothing processing and hole restoration through the more accurate normal information assistance under the conditionthat the original reconstruction precision is not reduced, improves the reconstruction quality of a curved surface, and solves a problem of a non-closed curved surface in the conventional three-dimensional reconstruction technology.

A system for classifying impulsive noise on a communications signal comprises an impulse signal generator, an integrator, a first comparator, and an impulse peak detector. The impulse signal generator receives a communications signal that includes impulsive noise and is configured to provide an impulse signal that includes just the impulsive noise. The integrator receives the impulse signal and integrates the impulse signal to determine the power of the impulse signal. The first comparator receives the impulse signal and is configured to compare the impulse signal to a first reference signal and indicate the time during which the value of the impulse signal is greater than the value of the first reference signal. The impulse peak detector receives the impulse signal and is configured to process the impulse signal, compare the processed signal to a second reference signal, and detect the peak value of the impulse signal.

The invention discloses a cage type asynchronous motor stator and rotor fault joint diagnosis method based on stack type self-encoding and a light gradient elevator algorithm. The method comprises the following steps: firstly carrying out the fast Fourier decomposition of a stator current instantaneous signal collected according to a certain frequency, and extracting the fault-related 20-dimensional feature quantity of each sample; then encoding the data through a stack type auto-encoder to realize automatic feature extraction; and inputting the coded data into a light gradient elevator classifier to carry out motor state multi-classification so as to diagnose fault types and severity. Manual feature extraction and manual intervention are not needed in the signal processing process, and meanwhile part of noise can be restrained through noise reduction self-coding. The classifier selects an optimization algorithm light gradient elevator algorithm based on a decision tree, the performance of structured data is superior to that of a deep neural network, the training time of the unit cycle index is the shortest on the premise that the highest precision is guaranteed, and the whole process code can be packaged and stored for subsequent training and practical application.

The invention discloses a noise classifying method based on an FPGA accelerated PCNN algorithm. The noise classifying method comprises the following steps: 1, acquiring a noise sample by using recording equipment, and editing into an audio file; 2, performing time-frequency conversion on the audio file; 3, performing feature extraction: converting a noise spectrum map into a grayscale map, using agray value as input of a PCNN model, accelerating the iterative process of the PCNN algorithm through an FPGA, and outputting time sequences as extracted features of different classes of noises; 4, iterating each class of noises in the noise sample for 50-200 times through processing in the step 3, outputting the time sequences, and then dividing into a training set and a testing set; 5, averaging the time sequences during each iteration of the training set, using the training set as a reference template, calculating the Euclidean distance between the time sequence of the testing set and thetime sequence of the reference template, determining as the same class of noises when the Euclidean distance is smaller than the threshold of a noise class, and outputting a recognition result. By thenoise classifying method, the feature extracting time is shortened and the time cost is saved.

The invention discloses an electromagnetic signal extraction and processing method based on a deep convolutional neural network. The electromagnetic signal extraction and processing method based on the deep convolutional neural network comprises the steps of performing primary denoising on original data through wavelet filtering, performing measurement point data leveling before channel extraction, performing measurement point superposition and channel extraction processing, designing a structure of a deep convolutional neural network (DCNN) model, performing calculation and identification by using the model and the like. According to the method, a noise and signal feature extraction mechanism is established, a signal-noise classification and recognition model is established through continuous iterative learning of a large amount of measured data, the signal and noise separation degree of underground large-depth, full-coverage and blind-area-free detection data is effectively improved, and data information used for inversion calculation and interpretation analysis is obtained to the maximum extent.