41 results about "Gammatone filter" patented technology

A method and apparatus for identifying running vehicles in an area to be monitored using acoustic signature recognition. The apparatus includes an input sensor for capturing an acoustic waveform produced by a vehicle source, and a processing system. The waveform is digitized and divided into frames. Each frame is filtered into a plurality of gammatone filtered signals. At least one spectral feature vector is computed for each frame. The vectors are integrated across a plurality of frames to create a spectro-temporal representation of the vehicle waveform. In a training mode, values from the spectro-temporal representation are used as inputs to a Nonlinear Hebbian learning function to extract acoustic signatures and synaptic weights. In an active mode, the synaptic weights and acoustic signatures are used as patterns in a supervised associative network to identify whether a vehicle is present in the area to be monitored. In response to a vehicle being present, the class of vehicle is identified. Results may be provided to a central computer.

The invention discloses a voice enhancement method for fusing phase estimation and human ear hearing characteristics in a digital hearing aid, comprising steps of obtaining a frequency domain expression mode containing noise through Fourier transformation, adopting a minimum value control recursive average method to obtain a noise power spectrum, obtaining initial enhancement voice and a noise amplitude spectrum, obtaining the initial enhancement voice and noise through correcting the phase of the voice and the noise through improving the phase estimation of the voice distortion under the low signal-to-noise ratio environment, performing filtering processing on the initial enhancement voice and the noise through a gammatone filter assembly which simulates the working mechanism of the artificial cochlea, performing analysis on the time frequency of the gammatone filter assembly to obtain the time frequency expression form consisting of time frequency units, using the hearing characteristics of the human ear to calculate binary mast containing noise in the time frequency domain, and using the mask value to obtain the enhanced voice after synthesis.

The invention belongs to the technical field of underwater acoustic signal processing, and particularly relates to an underwater acoustic signal target classification and recognition method based on deep learning. The method comprises the following steps: (1) carrying out feature extraction on an original underwater acoustic signal through a Gammatone filtering cepstrum coefficient (GFCC) algorithm; (2) extracting instantaneous energy and instantaneous frequency by utilizing an improved empirical mode decomposition (MEMD) algorithm, fusing the instantaneous energy and the instantaneous frequency with characteristic values extracted by a GFCC algorithm, and constructing a characteristic matrix; (3) establishing a Gaussian mixture model GMM, and keeping the individual characteristics of theunderwater acoustic signal target; And (4) finishing underwater target classification and recognition by using a deep neural network (DNN). According to the underwater acoustic signal target classification and recognition method, the problems that a traditional underwater acoustic signal target classification and recognition method is single in feature extraction and poor in noise resistance can be solved, the underwater acoustic signal target classification and recognition accuracy can be effectively improved, and certain adaptability is still achieved under the conditions of weak target acoustic signals, long distance and the like.

The invention provides a bi-ear time delay estimating method based on frequency division and improved generalized cross correlation in reverberation environment, and relates to the field of sound source positioning. A Gammatone filter is used to effectively simulate characteristics of a basal membrane of a human ear, voice signals are subjected to frequency division processing, and two-ear cross-correlation delay is estimated under a reverberation environment. Compared with a generalized cross correlation delay estimating method, the method can estimate time delay more accurately. The sound source positioning system has better robustness. A Gammatone filter is used to conduct frequency dividing processing for bi-ear signals, and each sub-band signal is subjected to inverse transformation to a time domain after reverberation processing of cepstrum and pre-filtering. Each sub-band signal of left and right ears are subjected to generalized cross correlation operation, an improved phase transformation weight function is employed in a generalized cross correlation algorithm to obtain cross correlation value of each sub-band for summing operation, and the bi-ear time difference corresponding to maximal cross correlation value is obtained.

The invention discloses an auditory perception characteristic-based speech quality objective evaluating method. The method is characterized in that a frequency spectrum is mapped into a Buck spectrum module, and the Buck spectrum module is added into a Gammatone filter bank to filter. The method comprises the following specific steps of (1) processing a reference signal and a degraded signal through POLQA (Perceptual Objective Listening Quality Analysis), and adding the reference signal and the degraded signal into a core model; (2) mapping the frequency spectrum in the core model into the Buck spectrum module, adding the Buck spectrum module into the Gammatone filter bank to filter, and performing acoustic conversion to enable the extracted auditory spectrum to be more approximate to the auditory perception of ears of people; (3) after performing acoustic conversion, performing interference analysis to analyze the distortion of the degraded signal relative to the reference signal so as to obtain an objective evaluation MOS score. Compared with the other methods, the method has the advantage that the relevancy between an objective evaluation result and a subjective evaluation result is effectively improved.

The invention discloses a self-adaptive hearing compensation method. The self-adaptive hearing compensation method comprises the following steps of firstly performing multi-channel decomposition for input signals by utilizing a gammatone filter group; then determining a compensation method based on a dynamic range of the signals in the channels and a hearing range of a hearing loss patient; if channel signals after being subjected to linear gain treatment are still within the hearing range of the patient, performing hearing compensation by using linear amplification so as to reduce distortion; otherwise, compensating by using dynamic range compression so as to add audibility. In addition, in order to reduce signal distortion caused by dynamic range compression and increase SNR (Signal to Noise Ratio) of an output signal in a noise environment, the self-adaptive compression method is adopted for hearing compensation so as to enable a compression ratio to be close to 1 to the greatest extent. Compared with existing hearing compensation methods, the self-adaptive hearing compensation method has higher intelligibility of compensated voice and obtains very strong practicability.

The invention discloses a deep learning-based binaural sound source positioning method in a digital hearing aid. The method comprises the following steps: firstly, decomposing a binaural sound sourcesignal into a plurality of channels through a gammatone filter, extracting a high-energy channel through a weighting coefficient, then extracting a first type of features by using a head-related-transform function (HRTF), namely an Interaural Time Difference (ITD) and an Interaural Intensity Difference (IID) which are used as inputs of deep learning, and dividing a horizontal plane into four quadrants to narrow a positioning range; secondly, extracting a second type of features of head-related transform, namely an Interaural Level Difference (ILD) and an Interaural Phase Difference (IPD); andfinally, in order to realize more precise positioning, taking the first and second types of four features as inputs of next deep learning, thereby obtaining an azimuth angle of sound source positioning. Precise positioning of 72 azimuth angles is realized on the horizontal plane from 0 degree to 360 degrees at a step length of 5 degrees.

The invention discloses a sound scene classification method based on network model fusion. According to the method, multiple different input characteristics are created through a sound track separation mode, an audio cutting mode and other modes, gamma pass filter cepstrum coefficients, Mel spectral characteristics and first-order and second-order differences of audio signals are extracted to serve as the input characteristics, multiple different corresponding convolutional neural network models are trained respectively, and lastly a support vector machine stacking method is adopted to realizea final fusion model. Through the method, the audio input characteristics with strong recognition are extracted by the adoption of the sound track separation mode, the audio cutting mode and other modes, a single-channel convolutional neural network and a two-channel convolutional neural network are constructed, a unique model fusion structure is generated finally, richer and more stereoscopic information can be obtained, the recognition rate and robustness of different sound scene classifications are effectively improved, and the method has a good application prospect.

An embodiment of the invention discloses a voiceprint identifying method and device, a server and a storage medium; the method comprises the following steps: voice signals to be identified are collected; according to at least one improved Gammatone wave filter, the voice signals are subjected to frequency band-based voiceprint feature extraction operation, and voice signal identifying results areconfirmed based on extracted voiceprint features. The voiceprint identifying method and device, the server and the storage medium disclosed in the embodiment of the invention can be used for solving aproblem that poor identifying effects are caused due to information loss of a high frequency part of voice based on technologies of the prior art, a resolution ratio of the wave filter for the high frequency part of the voice can be improved, accuracy of the voiceprint feature extraction can be improved, identifying effects on the high frequency part of the voice can be improved, and computational complexity and response time involved in voiceprint identification can be reduced.

The invention relates to a signal processing technology, and in particular, relates to a speaker recognition self-adaption method in a complex environment based on a GMM model. The method comprises aGMM-based speaker recognition model construction stage, that is to say, after low-pass filtering, pre-emphasis, windowing, framing and other preprocessing are carried out on a voice signal, filteringand denoising are carried out through a Gammatone filter, and GMFCC combined characteristic parameters are extracted. The method further comprises a speaker recognition and self-adaptation stage, thatis to say, speaker recognition is completed by extracting speech feature parameters of a speaker to be recognized and conducting self-adaptation adjustment on the original model. According to the method, the defect that the speaker recognition accuracy is reduced due to illness or complex environment is overcome, a new feature parameter combination method is provided, different features can be analyzed in a combined mode, errors caused by voice changes due to different conditions of speakers are effectively compensated, and thus the recognition accuracy is improved.

The invention puts forward a Gammatone filter bank chip system supporting voice real-time decomposition/synthesis, and belongs to the field of digital circuit design. The system comprises five parts, namely, an input module, a parameter module, a control module, a calculation module, and an output module. The input module activates the control module after receiving a frame of voice data, adjusts the delay of each channel according to the delay of human ear basilar membranes on different sub-bands, and sends the voice data to the control module. The control module makes the parameter module read the parameters of the corresponding channels and transmit the parameters to the calculation module. The calculation module completes the Gammatone filtering algorithm of each channel, and saves the result to the output module. After the calculation module completes calculation of all the channels of the frame of voice data, the output module allows the stored data to be read externally. With the system, the number of clocks consumed for calculation of channels is reduced, and less power is consumed. A parameter configurable function is realized, and the parameters of the system can be adjusted flexibly according to the need. Voice decomposition and synthesis is realized.

The invention discloses a Parkinson's disease speech detection method based on characteristics of power normalized cepstrum coefficients. The Parkinson's disease speech detection method solves the problem that Parkinson's disease speech detection is prone to being interfered by noise. The robustness of the extracted characteristics is enhanced through methods such as a Gammatone filter, noise removal and power normalization, and the detection method comprises the following steps that (1) a Parkinson's disease speech library and a healthy speech library are established; (2) characteristics of the power normalized cepstrum coefficients are extracted on a speech signal, specifically, firstly a speech signal is preprocessed, then the Gammatone filter is used for filtering to obtain a speech short-time power spectrum, then the speech short-time power spectrum is weighted and smoothed, and finally the characteristics of the power normalized cepstrum coefficients are calculated; (3) the outerproduct is used for obtaining characteristic vectors; (4) the characteristic vectors are subjected to power and l<2> norm normalization; (5) an SVM is used for training a Parkinson's disease speech model and a healthy speech model; and (6) an SVM classification method is used for classifying, and Parkinson's disease speech detection is realized.

The invention discloses a multi-sound-source positioning method based on Gammatone filter and a histogram. A microphone array is used for collecting a sound source signal, and a sub-band signal is obtained through a Gammatone filter bank, and framing and windowing processing are carried out for converting into a frequency domain; next, a controllable response power value is calculated, a histogramis drawn, the number of the main peak orientation and the secondary peak orientation are counted, and the orientation of the primary and secondary sound sources are estimated. According to the methoddisclosed by the invention, the frequency domains are overlapped with each other and are not separated, so that the phase winding is avoided; the side lobes are inhibited by the average effect of thespatial spectrums of the multiple frequency components, so that the main lobe is protruded, the space between the array elements is not strictly limited to a half wavelength, multi-frame informationis not needed, and the fact that the sound source does not need to be kept static in the continuous multi-frame is not needed to be presumed, and the real-time multi-sound source positioning is realized; and all the sub-band information in the same frame is fused with the histogram to serve as the judgment amount of the azimuth estimation, so that the method is simple and easy to operate, the calculation amount is low, the positioning success rate of the main sound source and the secondary sound sources is remarkably improved, and particularly the positioning success rate of the secondary sound sources is more obvious.

The invention discloses a fault location method of a capsule type floor heating device. The fault location method is applied into a floor heating track. The fault location method comprises the steps that fault speech signals are acquired and unit impulse response is subjected to convolution to obtain binaural signals; and after the binaural signals pass through a Gammatone filter, sub-band signalsafter frequency division are obtained, dereverbration processing of minimum phase decomposition is carried out in each sub-band signal, and a generalized cross-correlation function is obtained by calculating the cross-correlation of various sub-bands after the inverse transformation from a cepstrum domain to a time domain. The fault location method takes binaural speech localization as a multi-classification problem, takes GCCF (Generalized Cross-Correlation Function) and an interaural level difference as localization characteristics to input into DNNs of a softmax regression structure on a top floor, outputs the probability of which a sound source is located in each direction, and takes a maximum probability azimuth as the position of the sound source. The fault location method is simpleand convenient in installation, and the capsule type floor heating device which can locate the fault is provided in an innovation mode.

The invention relates to the field of sonar signal processing, in particular to a signal detection method and system based on decision statistic design. The method comprises the steps of employing a Gammatone filter group to carry out the frequency band processing of sensor pickup data, and obtaining the data of each frequency band, according to the difference between the signal and the background noise in intensity and stability, conducting variance normalization processing on each frequency band data, and establishing a judgment statistic amount, and performing judgment processing on each frequency band data according to the judgment statistics to realize signal enhancement and detection. According to the method, under the condition that a single filtering frequency band does not need to be searched, the signal-to-noise ratio of synthetic data is improved, and signal enhancement and detection are achieved.

The invention discloses a voice separation method based on a parameterized multi-phase gammatone filter bank. The method comprises the following steps of: firstly, constructing the parameterized multi-phase gammatone filter bank on the basis of a gammatone filter, then replacing an encoder of a Conv-Tasnet network with the parameterized multi-phase gammatone filter bank, keeping the encoder unchanged or adopting inverse transformation of the parameterized multi-phase gammatone filter bank to form a new Conv-Tasnet network, and training the new Conv-Tasnet network to obtain a final voice separation network. According to the method provided by the invention, under the condition that the decoder has learnable features, competitive performance is obtained; and under the condition that a decoder is inverse transformation of an encoder, the feature is superior to the features of manual design such as STFT and MPGTF.