134 results about "Short time fourier transformation" patented technology

The invention discloses a multi-speaker speech separation method and system. The method comprises the following steps : acquiring mixed voice signals, obtaining multichannel multi-speaker mixed voicesignals, and scanning the multi-channel multi-speaker mixed voice signals to obtain MUSIC energy spectra; obtaining S peak values from the MUSIC energy spectra, wherein each peak value corresponds toa beam direction; respectively enhancing S beams to obtain mixed voice in S directions; performing short-time Fourier transform on the mixed voice corresponding to each direction to obtain short-timeFourier amplitude spectra of the S target speaker voice, and respectively inputting the short-time Fourier amplitude spectra into a depth neural network to estimate a phase sensing mask correspondingto each target speaker; and performing element-by-element multiplication between the phase sensing mask of each target speaker and amplitude spectra of the corresponding mixed speech to obtain the amplitude spectrum of the target speaker, and the time domain signal of the target speaker is recovered by inverse short-time Fourier transform using the phase spectrum of the corresponding mixed speech.

The invention provides a passive positioning method of a single-vector hydrophone. The passive positioning method includes: (1) performing short-time Fourier transformation on acoustic pressure signals to obtain time frequency distribution of noise signals; (2) extracting time history of strong line spectrum signals, evaluating the recent passing time of a target according to the time history of a line spectrum frequency, and evaluating the moving speed of the target according to Doppler frequency offset of the line spectrum; (3) finding the direction of the target corresponding to the line spectrum according to the acoustic pressure signals and horizontal vibration velocity signals received by the single-vector hydrophone, and obtaining the direction time history of the target during the point measuring process; and (4) evaluating motion tracks of the target according to the direction time history and motion speed of the target. The passive positioning method of the single-vector hydrophone has the advantages of being capable of positioning the target only by continuously measuring the acoustic pressure and the horizontal vibration speed information without knowing the hydrology parameters of the ocean, and is suitable for deep oceans and shallow oceans.

The invention discloses an inhibiting method of background noise based on multiple statistical models and minimum mean square deviation, which comprises the following steps: proceeding short-time Fuliye transmission for voice signal corresponding to present input frame; utilizing the frame to retain the pure voice amplitude variance estimation and noise amplitude variance estimation on each frequency; calculating the estimation of real part and imaginary part corresponding to each frequent component of voice signal in the present frame; counting the probability of non-existed prior voice of each component frequency of the input frame; modifying the estimation result of real part and imaginary part of each frequency component; obtaining the true distribution of voice and noise in the real applying precisely; improving the inhibiting effect with stable estimating course; reducing the calculating complexity; fitting for kinds of voice communicating systems.

An inversion-based algorithm for computing the time frequency analysis of reflection seismograms using constrained least-squares spectral analysis is formulated and applied to modeled seismic waveforms and real seismic data. The Fourier series coefficients are computed as a function of time directly by inverting a basis of truncated sinusoidal kernels for a moving time window. Spectra may be provided that have reduced window smearing for a given window length relative to the discrete Fourier transform irrespective of window shape, and a time-frequency analysis with a combination of time and frequency resolution that is superior to the short time Fourier transform and the continuous wavelet transform. The reduction in spectral smoothing enables enhanced determination of spectral characteristics of interfering reflections within a short window. The degree of resolution improvement relative to the short time Fourier transform increases as window length decreases.

The invention discloses an epilepsy seizure prediction method based on a semi-supervised deep generative adversarial network. The method comprises the following steps: carrying out segmentation windowing processing on each EEG(electroencephalography) data of N original EEG samples through a sliding window in specified length, and during segmentation, carrying out superposition expansion processingon adjacent sub-windows to obtain S sections of sub-signals; carrying out short-time Fourier transform on the S sections of sub-signals to obtain spectrum signals, and converting the obtained spectrum data into an image for expression; inputting the processed picture information to a discriminator in the semi-supervised deep generative adversarial network for training and classification accordingto labeled training samples, training sample labels and unlabeled training samples; and inputting Z-dimensional noise data into a generator, and generating a pseudo-data image similar to distributionof the unlabeled data through the generator. The method gradually releases dependence of epilepsy seizure prediction on the labeled data through a semi-supervised learning mode, and obtains better classification accuracy.

The invention discloses a deep convolutional neural network (DCNN) based fine grain electrocardiogram classification method fusing with online decisions. Unlike a conventional method of using manual features or learning features from an original signal domain, the DCNN based method provided by the invention learns the features and classifications from a time frequency domain in an end-to-end manner. Firstly, electrocardiogram waveform signals are transformed into the time frequency domain by using short time Fourier transform; secondly, concrete DCNN network models are trained by training samples of a specific length; and finally, an online fusion method is provided to integrate previous and present decisions from the different models into a more accurate decision. Experimental results of20 kinds of ECG data sets show the effectiveness of the provided method.

The invention provides a speaker abnormal sound detecting method based on short-time Fourier transformation. A to-be-detected speaker is stimulated by swept-frequency signals, sound response signals of the to-be-detected speaker are collected, and the short-time Fourier transformation is performed on the sound response signals to obtain the time frequency picture corresponding to the sound response signals. The time frequency picture is segmented according to the features of speaker abnormal sound signals, the difference degree of each segmented area and the area corresponding to the standard time frequency is quantified, and display mode conversion is performed on the obtained difference degree to construct the abnormal sound curve. The abnormal sound curve is compared with the set threshold so as to judge whether abnormal sounds occur in the speaker. The threshold is the abnormal sound curve constructed by M times of the largest difference degree of the good time frequency picture and the standard time frequency picture.

The invention discloses a speech language classifying method based on a CNN and GRU fused deep neural network. The method comprises the following steps that S1, source audio data of a server is obtained, audio preprocessing is conducted, and the source audio data is cut; S2, audio data file information is read, and an audio data inventory CSV file is generated; S3, an audio data file is subjectedto short-time Fourier transformation, and two-dimensional speech spectrums associated with time and frequency domains of expansion of a series of frequency spectrum functions obtained after speech signal time domains are analyzed are obtained; S4, a model is built; S5, two-dimensional speech spectrum image data is input into the CNN and GRU fused speech language classifying deep neural network model, and language classification data is classified and output; S6, the language classification data and source audio data file information are stored. By means of the method, the problem about speechlanguage classification is solved, the method has the advantages of being automatic, high in identification rate, high in robustness, low in cost, high in portability and the like, and the business connection with a third-party system can be facilitated.

There is provided a sound signal processing device, in which an observation signal analysis unit receives multi-channels of sound-signals acquired by a sound-signal input unit and estimates a sound direction and a sound segment of a target sound to be extracted and a sound source extraction unit receives the sound direction and the sound segment of the target sound and extracts a sound-signal of the target sound. By applying short-time Fourier transform to the incoming multi-channel sound-signals this device generates an observation signal in the time-frequency domain and detects the sound direction and the sound segment of the target sound. Further, based on the sound direction and the sound segment of the target sound, this device generates a reference signal corresponding to a time envelope indicating changes of the target's sound volume in the time direction, and extracts the signal of the target sound, utilizing the reference signal.

The invention discloses a sound signal time delay estimation method and device and belongs to the technical field of audio processing. The sound signal time delay estimation method comprises the steps of obtaining two sound signals, conducting coherence matching on the two sound signals through short-time Fourier transformation of the two sound signals to obtain a first matching result, wherein the first matching result includes first matching positions and first matching degrees of the two sound signals; conducting coherence matching on the two sound signals according to spectral correlation of power spectrums of the two sound signals to obtain a second matching result, wherein the second matching result includes second matching positions and second matching degrees of the two sound signals; calculating the time delay of the two sound signals according to the first matching result and the second matching result. The problem that a relevant time delay estimation method is low in accuracy is solved. The sound signals are matched from two aspects of frequency domain distribution and power spectrums, the two matching results are integrated to determine a final matching result, and matching accuracy and time delay estimation accuracy are improved.

The invention provides a communication signal modulating and identifying method based on generalized S transformation. The communication signal modulating and identifying method comprises the following steps of: (1) carrying out analytic signal configuration on an input modulating signal according to the characteristic of an input signal to noise ratio modulating signal to obtain an analytic signal to be used as a signal for carrying out generalized S transformation; (2) configuring a Gaussian window function needed by the generalized S transformation; (3) determining a Gaussian window width factor sigma according to an expression formula of the generalized S transformation; carrying out the generalized S transformation on an input modulation signal by combining short-time Fourier transformation and the Gaussian window function to obtain time-frequency energy distribution images of the modulation signal; and (4) comparing energy images of all modulation signals according to the time-frequency energy distribution images obtained by the step (3) to find out a difference between the time-frequency energy distribution images of all modulation signals which are subjected to the generalized S transformation; selecting a frequency strip quantity with concentrated energy, a maximum value ratio of high-frequency component energy to low-frequency component energy, energy time domain distribution, high-frequency and low-frequency component extreme value time domain distribution and high-frequency component extreme value distribution so as to identify all modulation signals. According to the invention, a high identification rate is achieved under the condition of a low signal to noise ratio, and the method is suitable for modulating and identifying communication signals under a heavy clutter environment.

The invention belongs to emotional state recognition technology, and provides an electroencephalogram emotional state feature extraction method based on adaptive tracking in different frequency bands, which is a more objective emotional state recognition method and is a more objective evaluation method for therapy evaluation to mental diseases. The technical scheme includes that the electroencephalogram emotional state feature extraction method based on adaptive tracking in different frequency bands comprises following steps: data acquisition and preprocessing, time frequency feature extraction, adaptive frequency band selection and emotional state recognition; the specific data acquisition and preprocessing step comprises emotionally inducing a testee by the aid of emotional images and preprocessing acquired original electroencephalogram signals, and preprocessing comprises changing reference potential, reducing sampling, realizing bandpass filtering and removing electrooculogram; time-frequency change is realized for processed electroencephalogram signals by the aid of short time Fourier transform; and adaptive frequency band selection is realized by a frequency band variable adaptive tracking method, and an SVM (support vector machine) is used for realizing classification recognition for feature frequency bands. The electroencephalogram emotional state feature extraction method is mainly applied to recognizing emotional states.

Problems of permutation can be solved with high accuracy without utilizing knowledge about original signals or information concerning positions of microphones and the like when each one of plural signals mixed in an audio signal is separated using independent component analysis. A short-time Fourier transformation section generates spectrograms of observation signals from observation signals in time domain. A signal separation section separates the spectrograms of the observation signals into spectrograms of respective signals, to generate spectrograms of separate signals. A permutation problem solution section calculates a scale corresponding to the degree of permutation, e.g., a Kullback-Leiblar information amount calculated by use of a multidimensional probability density function or multidimensional kurtosis, from substantial whole of the spectrograms of the separate signals. Based on the scale, signals at each of frequencies bin of the spectrograms of the separate signals are exchanged between channels, to solve the permutation problem.

The invention discloses a brain electrical signal independent component extraction method based on convolution blind source separation. The brain electrical signal independent component extraction method based on the convolution blind source separation includes concrete steps: building a brain electrical signal independent component extraction system based on the convolution blind source separation, which comprises an AD (analog to digital) sampling module, a short time Fourier transformation module, a frequency domain instantaneous blind source separation module, a sequence adjustment module and a short time inverse Fourier transformation module; using the AD sampling module to sample brain electrical signals; using the short time Fourier transformation module to transform the brain electrical signals from a time domain to a frequency domain; using the frequency domain instant blind source separation module to separate instantaneous mixing signals in the frequency domain; using the sequence adjustment module to perform sequence adjustment on independent components in a vector on each frequency domain segment; using the short time inverse Fourier transformation module to transform a frequency domain separation result into an independent component on the time domain. The brain electrical signal independent component extraction method based on the convolution blind source separation extracts the independent components of brain electrical signals based on a true convolution mixing model, uses a convolution blind source separation frequency domain algorithm, and is simple to achieve, good in separation effect, and low in calculation complexity.

The invention discloses a historical voice frequency noise detection and elimination method. The existing historical voice frequency noise elimination method has high complexity, depends on the veracity of predicting model coefficients, has a complex algorithm and low efficiency and needs to estimate the statistical property of the noise in advance. The method of the invention comprises: (1) sound modeling: a sound signal can be described with the following method: y(k)=x(k)+j(k)*d(k); (2) short time discrete Fourier transform and spectrogram: an analysis window sliding along time is adopted for windowing and cutting off a non-stable signal so as to decompose the non-stable signal into a series of approximatively stable short signal, and Fourier transform is adopted to analyze the frequency spectrum of each stable short signal; (3) pulse noise detection; (4) pulse noise detection performance analysis; (5) signal repair and reconstruction; and (6) experiment result and analysis. The invention is used for removing the pulse noise in the audio data.

The invention discloses a method and a system for positioning coal rock dynamic disaster through infrasonic wave monitoring. The system comprises a monitoring main station and a plurality of monitoring substations, and the monitoring substations are connected with the monitoring main station through a fiber optic network; each monitoring substation is composed of three infrasonic wave sensors arranged in a triangular form, and the infrasonic wave sensors are connected with an infrasonic wave monitor; and the monitoring main station is provided with a computing server. The positioning method isthat the main station computing server filters collected data, the time-frequency analysis is performed on the data by the short-time Fourier transform (STFT), the frequency energy density is compared, signals of the main frequency band are extracted, and the calculation result is positioned through statistical analysis and time delay estimation theory. The system monitors the infrasonic wave signal generated by the fracture of the loaded coal rock. Through this method, the non-contact and regional tests are carried out on the stress distribution state of the surrounding rock of the tunnel ormines to determine the area of coal rock fracture and stress anomaly. The method is simple in installation and operation, has little impact on production, and has a large detection range and a long detection distance.

The invention discloses a method for vehicle flow statistics and vehicle type identification based on continuous wave radar, belonging to the field of traffic radar target detection. When a vehicle target is in the radiation area of radar, the radar obtains a Doppler intermediate-frequency signal. Through short-time Fourier transform on the collected signal, a two-dimensional domain of time and frequency is obtained to represent the change characteristic of the Doppler frequency of the vehicle target with time. The time-frequency image is binarized through an Otsu method, and the time-frequency characteristics of the vehicle target are separated and extracted. The change of the time-frequency characteristics of the target at the time of crossing the radiation area of the radar is represented in a list method. Interference targets smaller than a threshold in a connected region are removed according to the energy dominance criterion. Traffic flow statistics is completed by counting characteristic lists. Then, each characteristic list is traversed, and the vehicle type is identified according to the spectrum ridge broadening width. Compared with the traditional detection line method, the method of the invention has the characteristics of visual judgment and accurate measurement.

The invention discloses a blind estimation method for underdetermined hybrid frequency hopping parameters based on time frequency diagram correction in order to mainly solve the problem of large estimation errors caused by a poor signal to noise ratio in the prior art. The method comprises the following implementation steps: (1) performing short-time Fourier transformation on a signal to obtain a time frequency matrix of an observed signal; (2) calculating an energy matrix of the time frequency matrix; (3) correcting the time frequency diagram; (4) estimating frequency hopping signal parameters according to the corrected time frequency diagram. The method disclosed by the invention can be used for obtaining a clear time frequency diagram at a low signal to noise ratio, has the advantage of improving the estimation precision of the frequency hopping signal parameters, and can be applied to various frequency hopping communication scenes.

The invention discloses a target human motion state recognition method suitable for a micro-motion interference scene. The target human motion state recognition method comprises the following steps that (1) a continuous wave radar transceiver is constructed to acquire a Doppler signal formed by human body motion in a target space; (2) an empirical modal decomposition algorithm is used for removingmicro-Doppler signals generated by other human body micro-motion in the target space, and the signals generated by a target human body are extracted; (3) the short-time Fourier transform and the Hermite multi-window are used for performing time-frequency energy spectrum analysis on the signals of the target human body, and the signals are subjected to energy accumulation; (4) the time-frequency energy spectrum obtained after energy accumulation is used for extracting the characteristics of human trunk motion and swing arm motion; and (5) integrated learning ideas are used for combining Bagging with a decision-making tree to form a motion state classifier, and the recognition of six motion states of still, running, crawling, one-arm motion, dual-arm motion and armless motion. The target human motion state recognition method realizes the removal of other human body micro-motion interference signals.