CN105810213A - Typical abnormal sound detection method and device - Google Patents
Typical abnormal sound detection method and device Download PDFInfo
- Publication number
- CN105810213A CN105810213A CN201410850883.5A CN201410850883A CN105810213A CN 105810213 A CN105810213 A CN 105810213A CN 201410850883 A CN201410850883 A CN 201410850883A CN 105810213 A CN105810213 A CN 105810213A
- Authority
- CN
- China
- Prior art keywords
- eigenmatrix
- sound
- identification
- tested
- abnormal sound
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Landscapes
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
The invention discloses a typical abnormal sound detection method and a device, relates to the voice signal processing field and particularly provides an abnormal sound detection method based on spectrograms. According to the method, a sound signal is acquired, the sound signal is pre-processed, and effective signal segments of the sound signal are determined; the spectrograms of the effective signal segments are acquired, a to-be-tested identification characteristic matrix is determined, and the to-be-tested identification characteristic matrix is used for representing sound intensity distribution of the sound signal on the time-frequency domain; similarities between the to-be-tested identification characteristic matrix and each standard identification characteristic matrix in an abnormal sound model database are calculated, and abnormal sound types of the effective signal segments are determined according to calculation results; computational complexity in an abnormal sound identification process is simplified, and detection accuracy is improved.
Description
Technical field
The present invention relates to field of voice signal, particularly relate to a kind of typical case's abnormal sound detection method and device.
Background technology
Abnormal sound refers to lower sound that should not occur of home, is often referred in the normal productive life of the mankind, the sound relevant to people's security of the lives and property occurred suddenly.Abnormal sound detection refers to the end points (beginning and end) finding out abnormal sound from one section of continuous print acoustical signal, and is extracted from which by effective abnormal sound signal.The aspect ratio of glass breaking sound is more apparent, detection is got up more convenient in actual applications, and general all along with glass breaking sound when blasting in some occasions such as market, bus, street etc., therefore carry out detecting certain practical value as a kind of typical abnormal sound using glass breaking sound.
Technology in the outer speech recognition technology being both along about the characteristic parameter adopted in the Study of recognition of abnormal sound of Present Domestic and parameter, the correlational study about abnormal sound detection mainly has:
First kind of way: carry out abnormal sound detection based on cepstral analysis, namely mainly adopts cepstral domain feature parameter to carry out the feature description of sound.Generally the research of glass breaking sound is continued to use mostly the technology of speech recognition, utilize mel cepstrum coefficients, short-time energy etc. are as the eigenvalue of this abnormal sound, to the eigenvalue extracted, utilize some models, such as HMM, neural network model, gauss hybrid models etc., are trained eigenvalue identifying, detect glass breaking sound.But only when the output of each passband is added up impartial, mel cepstrum coefficients could express maximum quantity of information, and in fact, abnormal sound distribution in frequency band is not necessarily equally distributed, and utilize the mode of training to identify, complexity is higher, can be subject to certain restrictions in actual applications, and glass breaking sound can not be well identified by the method for therefore traditional speech recognition.
The second way: voice data is divided into 50ms mono-frame, using ZCR (zero-crossing rate), LPC (linear prediction), LPCC (linear prediction residue error) etc. as feature set, gauss hybrid models is as Multilayer Classifier, its core manifestation is in multi-level categorizing process, ground floor distinguishes background sound, the second layer distinguishes voice and non-voice, third layer identification abnormal sound.Although this method is it can be considered that the impact of background noise, but it is extremely complex to realize process, it is not easy to be applied to video and the Voice Surveillance field of reality.
Summary of the invention
The embodiment of the present invention provides a kind of typical case's abnormal sound detection method and device, it is possible to the feature based on sound spectrogram detects abnormal sound, simplifies the computation complexity in exceptional sound recognition process, improves the accuracy rate of detection.
The embodiment of the present invention provides a kind of typical case's abnormal sound detection method, and the method includes:
Collected sound signal, carries out pretreatment to described acoustical signal, it is determined that the useful signal fragment of described acoustical signal；
Obtaining the sound spectrogram of described useful signal fragment and determine identification eigenmatrix to be tested according to described sound spectrogram, described identification eigenmatrix to be tested is for representing the distribution situation of acoustical signal intensity of sound on time-frequency domain；
Calculate the similarity of described identification eigenmatrix to be tested and each standard identification eigenmatrix in abnormal sound model library, determine the abnormal sound type in described useful signal fragment according to result of calculation.
The embodiment of the present invention also provides for a kind of typical case's abnormal sound detecting device, and this device includes:
Collecting unit, for collected sound signal, carries out pretreatment to described acoustical signal, it is determined that the useful signal fragment of described acoustical signal；Obtain the sound spectrogram of described useful signal fragment；
Determining unit, for determining the identification eigenmatrix of abnormal sound sound spectrogram, described identification eigenmatrix to be tested is for representing the distribution situation of acoustical signal intensity of sound on time-frequency domain；
Computing unit, for calculating the similarity of described identification eigenmatrix to be tested and each standard identification eigenmatrix in abnormal sound model library, determines the abnormal sound type in described useful signal fragment according to result of calculation.
Visible, the embodiment of the present invention does not use the method for traditional speech recognition to detect abnormal sound, but uses sound spectrogram that abnormal sound is detected.Avoid in characteristic extraction procedure owing to using the extraction such as mel cepstrum coefficients of the speech characteristic value of routine, cause because of abnormal sound uneven distribution in frequency band, cause that the output statistics of each passband is unequal, so that mel cepstrum coefficients can not well characterize the feature of sound, and the problem that the recognition accuracy caused is not high.It addition, the method that the embodiment of the present invention adopts sound spectrogram detects abnormal sound, computation complexity is relatively low, makes it possible to well be applied in actual field of video monitoring, expands the scope used.
Accompanying drawing explanation
In order to be illustrated more clearly that the technical scheme in the embodiment of the present invention, below the accompanying drawing used required during embodiment is described is briefly introduced, apparently, accompanying drawing in the following describes is only some embodiments of the present invention, for those of ordinary skill in the art, under the premise not paying creative work, it is also possible to obtain other accompanying drawing according to these accompanying drawings.
The schematic flow sheet of a kind of typical case's abnormal sound detection method that Fig. 1 provides for the embodiment of the present invention；
The schematic flow sheet of a kind of abnormal sound detection method based on sound spectrogram that Fig. 2 provides for the embodiment of the present invention；
Fig. 3 extracts the schematic flow sheet of process for the identification eigenmatrix a kind of to be tested that the embodiment of the present invention provides；
A kind of schematic flow sheet calculating similarity that Fig. 4 provides for the embodiment of the present invention；
The structural representation of a kind of typical case's abnormal sound detecting device that Fig. 5 provides for the embodiment of the present invention.
Detailed description of the invention
In order to make the object, technical solutions and advantages of the present invention clearly, below in conjunction with accompanying drawing, the present invention is described in further detail, it is clear that described embodiment is only a part of embodiment of the present invention, rather than whole embodiments.Based on the embodiment in the present invention, all other embodiments that those of ordinary skill in the art obtain under not making creative work premise, broadly fall into the scope of protection of the invention.
The embodiment of the present invention proposes a kind of method that feature based on sound spectrogram detects abnormal sound.For acoustic information, people rely on tone color, pitch, three features of volume to make a distinction, and these features all can by the parameter characterization in sound spectrogram time-frequency structure.Sound spectrogram is a kind of three-dimensional perception figure describing acoustical signal, is made up of frequency, time, three dimensional information of intensity of sound.And the time-frequency structure of different abnormal sound sound spectrograms towards with fine degree in have significant difference, different abnormal sound signals has the announcement of essence simultaneously possess again good distinctive.Using sound spectrogram as primitive character source, according to certain rule, extract the eigenmatrix of sound spectrogram, by calculating the distance identified between eigenmatrix in identification eigenmatrix to be tested and abnormal sound model library, detect abnormal sound.The embodiment of the present invention can better characterize the feature of sound, improves the accuracy rate of detection, and identifies that process computation is fairly simple, and complexity is relatively low, it is possible to is well applied in actual video monitoring.
Fig. 1 illustrates the schematic flow sheet of a kind of typical abnormal sound detection method that the embodiment of the present invention provides, as it is shown in figure 1, the method may include that
Step 11: collected sound signal, carries out pretreatment to acoustical signal, extracts the useful signal fragment of acoustical signal.
Step 12: obtaining the sound spectrogram of this useful signal fragment and extract identification eigenmatrix to be tested according to this sound spectrogram, this identification eigenmatrix to be tested is for representing the distribution situation of acoustical signal intensity of sound on time-frequency domain.
Step 13: calculate the similarity of identification eigenmatrix to be tested and each standard identification eigenmatrix in abnormal sound model library, determine the abnormal sound type in useful signal fragment according to result of calculation.
Optionally, in above-mentioned steps 12, the oscillogram of useful signal fragment is converted to the sound spectrogram being made up of the frequency of useful signal fragment, time and three dimensional information of intensity of sound；Extract the identification eigenmatrix to be tested being used for characterizing the intensity of sound distribution situation of sound spectrogram.
Optionally, in above-mentioned steps 13, calculate the similarity of identification eigenmatrix to be tested and each standard identification eigenmatrix in abnormal sound model library, by the noise type corresponding to the standard identification eigenmatrix in the abnormal sound model library maximum with identifying eigenmatrix similarity, it is determined that for the abnormal sound type of useful signal fragment.
Optionally, in above-mentioned steps, useful signal fragment windowing is divided into several frames；Each frame being carried out Short Time Fourier Transform, obtains the spectrum information of this frame, this spectrum information is for representing the relation between the frequency of this frame and intensity of sound；Connecting the spectrum information of all frames, obtain the sound spectrogram of useful signal fragment, sound spectrogram is made up of several points, and (x, y) is used for representing that this point is in the x moment coordinate of any point, intensity of sound corresponding in y frequency.
Optionally, in above-mentioned steps, frame at each point in sound spectrogram, frequency and intensity of sound, extract sound spectrogram identification eigenmatrix；For each frame, extract intensity of sound value front K frequency values from big to small in sound spectrogram matrix；According to the relative position in time domain of each frequency values in K frequency values, generate the identification eigenmatrix of intensity of sound distribution situation for characterizing sound spectrogram.
Optionally, in above-mentioned steps 13, for the either standard identification eigenmatrix in described abnormal sound model library, element by the last string of the first row of the element of the first row first row of described standard identification eigenmatrix to described standard identification eigenmatrix, determine respectively and in described identification eigenmatrix to be tested difference between the value of the first row first row less than each element in the described standard identification eigenmatrix of predetermined threshold value, carry out each element present position respectively as starting point setting window operation, respectively obtain the window matrix that this window covers, this window is identical with the dimension of identification eigenmatrix to be tested；Calculate business's matrix of each window matrix and identification eigenmatrix to be tested, each row element in described business's matrix is multiplied by default weighted value respectively, obtain the similarity of each window matrix and described identification eigenmatrix to be tested；Whole similarity according to each standard identification eigenmatrix in calculated described identification eigenmatrix to be tested and abnormal sound model library, is defined as the similarity of described standard identification eigenmatrix and described identification eigenmatrix to be tested by similarity maximum for the value in whole similarities.
Optionally, in above-mentioned steps 11, acoustical signal is carried out preemphasis process, obtain the acoustical signal after preemphasis；Acoustical signal after preemphasis is carried out framing windowing process, useful signal fragment will be defined as in short time period self-energy value more than one section of acoustical signal of predetermined threshold value.
It should be noted that, the abnormal sound data base that the noise model used in this method can be through investigating the related data arranging the research of domestic and international abnormal sound and obtains includes various unlike material, different size, glass breaking sound when different-thickness, in order to close to actual use occasion, be also added into the background noise of typical public place: railway station and large-scale square.The embodiment of the present invention only carries out citing with above-mentioned noise model and describes, but the embodiment of the present invention is not limited only to detection glass breaking sound, it is also possible to be preferably applicable to detection and the identification of other typical case's abnormal sounds.
Below the embodiment of the present invention is described in detail.
Fig. 2 illustrates the schematic flow sheet of a kind of abnormal sound detection method based on sound spectrogram that the embodiment of the present invention provides, as in figure 2 it is shown, the method may include that
Step 21: collected sound signal, carries out pretreatment to the acoustical signal collected, and extracts useful signal fragment.
Concrete, acoustical signal is carried out preemphasis process, obtains the acoustical signal after preemphasis；Acoustical signal after preemphasis is carried out framing windowing process, useful signal fragment will be defined as in short time period self-energy value more than one section of acoustical signal of predetermined threshold value.
When implementing, in embodiments of the present invention the acoustical signal collected is carried out pretreatment and can include preemphasis, end-point detection and framing windowing etc..In the embodiment of the present invention, first, the acoustical signal collected is carried out preemphasis, utilize the HFS of the specific high pass filter acoustical signal to collecting to compensate；Then, acoustical signal after preemphasis is carried out end-point detection, determine the starting point of useful signal in this acoustical signal, in general, would be likely to occur the quiet or blank of a period of time due in the acoustical signal that collects, in order to improve the detection efficiency of abnormal sound, (namely the embodiment of the present invention can be determined by the effective acoustic information in acoustical signal, useful signal) starting point, then useful signal is carried out abnormal sound coupling or detection again；Subsequently, to having determined that the useful signal of starting point characteristic parameter within the specific limits carries out framing windowing process so that it is meet statistical property steady.
Step 22: the oscillogram of useful signal fragment is converted to the sound spectrogram being made up of the frequency of useful signal fragment, time and three dimensional information of intensity of sound.
When implementing, sound spectrogram is widely used in analysis and the process of audio signal.It is X-Y scheme, although each corresponding three-dimensional value on it.Its transverse axis is the time, and the longitudinal axis is frequency.(x, y) corresponding point represents that, at moment x, the intensity of sound on frequency y, this is showed by different colors with coordinate.From the sound spectrogram of acoustical signal, it is possible to investigate out distribution and the situation of change of intensity of sound within the scope of whole T/F；And this is cannot to present in oscillogram.For obtaining sound spectrogram, acoustical signal is divided into very short frame, and consecutive frame has certain overlap.Then each frame is done Short Time Fourier Transform and obtains the spectrum information of correspondence, owing to sound spectrogram is made up of frequency, time, three dimensional information of intensity of sound, it is therefore desirable to the value of intensity of sound is calculated.Spectrum information connects into complete sound spectrogram the most at last.
For example, useful signal fragment windowing is divided into several frames by the embodiment of the present invention；Each frame being carried out Short Time Fourier Transform, obtains the spectrum information of this frame, spectrum information is for representing the relation between the frequency of this frame and intensity of sound；Connecting the spectrum information of all frames, obtain the sound spectrogram of useful signal fragment, sound spectrogram is made up of several points, and (x, y) is used for representing that this point is in the x moment coordinate of any point, intensity of sound corresponding in y frequency.In embodiments of the present invention, the clock signal of sound being carried out Fourier's change in short-term, the length of Fourier transformation is 2N point, and the signal of so each frame can obtain the frequency spectrum that length is N, and the sound pressure level of every bit is expressed as: P=20*log10|x(1/N)|
Wherein, P is the sound pressure level of this point, and x is the spectrum value of this frame signal.
Step 23: distinguish abnormal sound and background noise according to sound spectrogram.
When implementing, it is found through experiments, when glass breaking occurs, owing to different tensions produces some overtones, therefore in sound spectrogram except including the harmonics of fundamental frequency integral multiple, also have non-integral multiple harmonics, the time-frequency structure of sound spectrogram shows as each frequency range the band of irregular ripples grain pattern occurs.And the background noise energy of public place is evenly distributed, its sound spectrogram does not have obvious time-frequency result, and sound spectrogram therefore can be utilized to remove the noise impact on voice recognition.It should be noted that the embodiment of the present invention is not limited only to produced abnormal sound during detection glass breaking, it is also with various abnormal sound model and other various types of abnormal sounds are detected and identify.
Step 24: frame at each point in sound spectrogram, frequency and intensity of sound, generates sound spectrogram matrix.
When implementing, each point in sound spectrogram actually corresponds to the intensity of sound value in preset time, given frequency, therefore selects sound spectrogram can better show the time-frequency structure of acoustical signal as primitive character source.
For example, frequency spectrum is pressed predefined parameter discretization by the process of Fourier transformation, frequency spectrum array corresponding for every frame is coupled together, is the formation of sound spectrogram matrix.Wherein, the row in sound spectrogram matrix are corresponding to the frame on the time period, and the row in sound spectrogram matrix is corresponding to the frequency on frequency band.
Step 25: extract the identification eigenmatrix to be tested being used for characterizing the intensity of sound distribution situation of sound spectrogram from sound spectrogram matrix.
When implementing, identification eigenmatrix to be tested is extracted from sound spectrogram matrix, because the frequency that intensity is less is little on the impact of sound effect, therefore for each frame data, it is ranked up by the intensity level according to each frequency, it is dynamically K frequency values of every frame preservation maximum intensity, ignore other frequency, minimum space so can be used to preserve the most significant information, greatly reduce the memory space of feature, so it is possible not only to reduce computation complexity, and useful information can be preserved preferably, do not lose the representational of content.
Due to when generating sound spectrogram, the frequency domain set is unified, discrete is uniform, what therefore identical in sound spectrogram matrix line identifier was corresponding is identical frequency, the time dependent situation of the frequency being identical that identical row is corresponding, therefore, line identifier is equivalent to the strongest frequency relative position on frequency domain.Fig. 3 illustrates that the identification eigenmatrix a kind of to be tested that the embodiment of the present invention provides extracts the schematic flow sheet of process, as it is shown on figure 3, this process can include but not limited to following steps:
First, sound spectrogram matrix finds the line identifier that in each column, maximum is corresponding (such as, what the first row was corresponding is designated A, what the second row was corresponding is designated B, what the third line was corresponding is designated C, ... by that analogy, repeat no more here), these line identifiers are stored in the first row identifying eigenmatrix.Arranging corresponding to the frame on the time period, row is corresponding to the frequency on frequency band, and therefore line identifier can regard the conversion expression-form of frequency values as, say, that line identifier is the positional information that peak frequency is corresponding；Different line identifiers is stored in identify eigenmatrix the first row can be understood as identification eigenmatrix in store the different frequency value that diverse location is corresponding.If it should be noted that there is multiple maximum, then preserve minimum line identifier (that is, line identifier from small to large is followed successively by A～Z).Secondly, find out the line identifier that Second Largest Value in each column is corresponding, be stored in the second row identifying eigenmatrix, by that analogy, repeat K time altogether, constitute the eigenmatrix of K row, and its columns is equal to frame number.Owing to experiment finds, when the value of K is more than 5, the impact of experimental result is little, therefore the value of K can be defined as 5 by the embodiment of the present invention, so, is possible not only to reduce computation complexity, moreover it is possible to well represent the feature of sound.Further, what last column of identification eigenmatrix to be tested was deposited is the short-time average magnitude value that front 5 maximums are corresponding, constitutes the identification eigenmatrix to be tested of (K+1) * frame number.And the line identifier deposited in identification eigenmatrix to be tested represents the strongest frequency relative position in frequency domain.
Step 26: calculate the similarity of identification eigenmatrix to be tested and each standard identification eigenmatrix in abnormal sound model library.
When implementing, owing to the data in identification eigenmatrix to be tested are not continuous print characteristic vector values, Euclidean distance can not well represent the characteristic of two matrixes, and therefore the embodiment of the present invention calculates the similarity of two matrixes in the following ways.
Preferably, abnormal sound model library storage in embodiments of the present invention has the standard identification eigenmatrix corresponding to various types of typical abnormal sound template types.
It should be noted that in embodiments of the present invention, only it is only used as, with following steps S41～step S47, the citing calculating similarity and is described in detail, the method being not limited to step S41～step S47 when practical application.In abnormal sound detection process, after performing step 26, continue executing with step 27.
Fig. 4 illustrates a kind of schematic flow sheet calculating similarity that the embodiment of the present invention provides, and as shown in Figure 4, this process may include that
Step S41: judge that in identification eigenmatrix to be tested, whether the value of the first row first row element is closest with the value of the first row i-th column element of the standard identification eigenmatrix in abnormal sound model library, if so, then performs step S43；Otherwise, step S42 is performed.
Step S42: sliding window is moved rearwards by 1 row by the i-th row in the standard identification eigenmatrix in abnormal sound model library, makes i=i+1.And perform S41 (that is, what now compare is the value that in identification eigenmatrix to be tested, the first row i-th (i=i+1) of the standard identification eigenmatrix in the value of the first row first row element and abnormal sound model library arranges), until the standard identification eigenmatrix in abnormal sound model library finding the value of string and identifying that in eigenmatrix, the value of the first row first row element is closest.
Step S43: i-th in the standard identification eigenmatrix in abnormal sound model library is arranged the first row as sliding window, using the dimension of the identification eigenmatrix to be tested dimension as sliding window, it is determined that go out the matrix of and identification eigenmatrix same dimension to be tested.
Step S44: calculate business's matrix of sliding window place matrix and identification eigenmatrix to be tested.
For example, sliding window the data point of element in a matrix divided by the value of the element being in same position in identification eigenmatrix to be tested, element each in two matrixes is sequentially carried out a division operation, thus obtaining business's matrix of sliding window place matrix and identification eigenmatrix to be tested, if the element value being in two matrix same positions is equal, then it is designated as 1 in the value of business's matrix corresponding position.
Step S45: the similarity according to business's matrix calculus sliding window place matrix Yu identification eigenmatrix to be tested.
Step S46: judge whether this window moves to the last string of the first row of standard identification eigenmatrix, if so, then continues executing with step S47, otherwise, performs step S42.
When implementing, until this sliding window moves to the final position (the last string position of the first row) of standard identification eigenmatrix, obtain each similarity of this window place matrix and described identification eigenmatrix to be tested.
It should be noted that, in abnormal sound model library, storage has various typical case's abnormal sound model, each corresponding standard identification eigenmatrix corresponding with the characteristic of this typical case's abnormal sound of typical case's abnormal sound model, only any one the standard identification eigenmatrix in abnormal sound model library is mated by above-mentioned steps S41～step S46, in order to mate with the typical abnormal sound model of every kind in abnormal sound model library, the embodiment of the present invention can also for other typical case's abnormal sound model of typical case's abnormal sound model, perform above-mentioned steps S41～step S46 successively, until to the whole typical case's abnormal sound models in abnormal sound model library (namely, whole standard identification eigenmatrixes) it is sequentially carried out coupling, so that it is determined that go out the abnormal sound type of useful signal fragment.After to standard identification eigenmatrix standard identification eigenmatrix, continue executing with step S47.
Step S47: similarity maximum for the value in each similarity is defined as the similarity of described standard identification eigenmatrix and described identification eigenmatrix to be tested.
In embodiments of the present invention, to ensure that the dimension of two matrixes is equal when calculating business's matrix, and the duration of generally corresponding than each standard identification eigenmatrix in the abnormal sound model library sound of the duration of identification eigenmatrix correspondence sound to be tested is short, it is thus desirable to find out the dimension identical with identification eigenmatrix to be tested, one can be utilized always to start to slide into the last string of the first row with the first row first row of the sliding window standard identification eigenmatrix from abnormal sound model library identifying eigenmatrix same dimension, and compare one by one.In order to improve computational efficiency, first find with the first row the first train value close proximity in identification eigenmatrix to be tested as original position, mark the window matrix of a piece and identification eigenmatrix formed objects to be tested, calculate business's matrix of identification eigenmatrix to be tested and window matrix.Due to the first row the strongest frequency values of correspondence, corresponding strong frequency values of the second row in identification eigenmatrix to be tested, and so on.As a rule, intensity is more weak, is more easily subject to external influence, and error is more big.Treat it is thus desirable to each row to be regarded as different significance levels, it is possible to being multiplied by weights for each row, intensity is more little, weights are more little, through many experiments, the embodiment of the present invention can be set to secondary from first row to last leu for the weight (that is, each row element in business's matrix being multiplied by default weighted value respectively) of the business's matrix calculated: 1,1,0.8,0.6,0.5,0.6, thus calculating the similarity of two matrixes.Followed by moving window, if first position is not right in window, avoid the need for continuing to compare backward, direct moving window, until finding new original position, therefore calculated similarity is likely to more than one, select similarity maximum as such, it is desirable to continue executing with step 27.It should be noted that default weighted value in embodiments of the present invention only illustrates for above-mentioned more preferably value, this default weighted value can also be other value, no longer repeats one by one here.
For example, standard identification eigenmatrix in described abnormal sound model library finds and the immediate numerical value present position of value of the first row first row in described identification eigenmatrix to be tested, and this position is set window as starting point, this window is identical with the dimension of identification eigenmatrix to be tested, calculate business's matrix of this window and identification eigenmatrix to be tested, by the business's matrix calculated and default weighted value, obtain the similarity of window and identification eigenmatrix to be tested.Then continuing to moving window, if the first of new window position is kept off with recognition matrix to be tested, then continuing to be moved rearwards by window without comparing, until finding new original position.So can obtain more than one similarity, using wherein maximum as the Similarity value identifying eigenmatrix and identification eigenmatrix to be tested in model library.Further, each the standard identification eigenmatrix in abnormal sound model library is performed aforesaid operations, and obtains several similarities.
Step 27: by the abnormal sound model corresponding to each standard identification eigenmatrix in the abnormal sound model library maximum with identification eigenmatrix similarity to be tested, it is determined that for the abnormal sound type of useful signal fragment.
Visible, the embodiment of the present invention does not use the method for traditional speech recognition to detect abnormal sound, but uses sound spectrogram that abnormal sound is detected.Avoid in characteristic extraction procedure owing to using the extraction such as mel cepstrum coefficients of the speech characteristic value of routine, cause because of abnormal sound uneven distribution in frequency band, cause that the output statistics of each passband is unequal, so that mel cepstrum coefficients can not well characterize the feature of sound, and the problem that the recognition accuracy caused is not high.It addition, the method that the embodiment of the present invention adopts sound spectrogram detects abnormal sound, computation complexity is relatively low, makes it possible to well be applied in actual field of video monitoring, expands the scope used.
Based on identical technology design, the embodiment of the present invention also provides for a kind of device that can be used for performing above-mentioned abnormal sound detection method, and Fig. 5 illustrates a kind of typical case's abnormal sound detecting device that the embodiment of the present invention provides, as it is shown in figure 5, this device may include that
Collecting unit 51, for collected sound signal, carries out pretreatment to described acoustical signal, it is determined that the useful signal fragment of described acoustical signal；Obtain the sound spectrogram of described useful signal fragment；
Determining unit 52, for determining the identification eigenmatrix of abnormal sound sound spectrogram, described identification eigenmatrix to be tested is for representing the distribution situation of acoustical signal intensity of sound on time-frequency domain；
Computing unit 53, for calculating the similarity of described identification eigenmatrix to be tested and each standard identification eigenmatrix in abnormal sound model library, determines the abnormal sound type in described useful signal fragment according to result of calculation.
Optionally, described collecting unit 51 specifically for: the oscillogram of described useful signal fragment is converted to the sound spectrogram being made up of the frequency of described useful signal fragment, time and three dimensional information of intensity of sound；Described determine that unit 52 is specifically for obtaining the identification eigenmatrix of intensity of sound distribution situation for characterizing described sound spectrogram；Described computing unit 53 specifically for: calculate the similarity of each standard identification eigenmatrix in described identification eigenmatrix to be tested and abnormal sound model library, by the abnormal sound model maximum with described identification eigenmatrix similarity to be tested, it is determined that for the abnormal sound type of described useful signal fragment.
Optionally, described collecting unit 51 specifically for: described useful signal fragment is divided into several frames according to default sample rate；Any frame being carried out Short Time Fourier Transform, obtains the spectrum information of this frame, described spectrum information is for representing the relation between the frequency of this frame and intensity of sound；Connecting the spectrum information of all frames, obtain the sound spectrogram of described useful signal fragment, described sound spectrogram is made up of several points, and (x, y) is used for representing that this point is in the x moment coordinate of any point, intensity of sound corresponding in y frequency.
Optionally, described determine that unit 52 is specifically for frame at each point in described sound spectrogram, frequency and intensity of sound, it is determined that sound spectrogram identification eigenmatrix；For each frame, it is determined that intensity of sound value front K frequency values from big to small in described sound spectrogram identification eigenmatrix；According to the relative position in time domain of each frequency values in described K frequency values, generate the identification eigenmatrix to be tested of intensity of sound distribution situation for characterizing described sound spectrogram.
Optionally, described computing unit 53 specifically for: for the either standard identification eigenmatrix in described abnormal sound model library, element by the last string of the first row of the element of the first row first row of described standard identification eigenmatrix to described standard identification eigenmatrix, determine respectively and in described identification eigenmatrix to be tested difference between the value of the first row first row less than each element in the described standard identification eigenmatrix of predetermined threshold value, carry out each element present position respectively as starting point setting window operation, respectively obtain the window matrix that this window covers, this window is identical with the dimension of identification eigenmatrix to be tested；Calculate business's matrix of each window matrix and identification eigenmatrix to be tested, each row element in described business's matrix is multiplied by default weighted value respectively, obtain the similarity of each window matrix and described identification eigenmatrix to be tested；Whole similarity according to each standard identification eigenmatrix in calculated described identification eigenmatrix to be tested and abnormal sound model library, is defined as the similarity of described standard identification eigenmatrix and described identification eigenmatrix to be tested by similarity maximum for the value in whole similarities.
Optionally, described collecting unit 51 specifically for: described acoustical signal is carried out preemphasis process, obtains the acoustical signal after preemphasis；Acoustical signal after preemphasis is carried out framing windowing process, useful signal fragment will be defined as in short time period self-energy value more than one section of acoustical signal of predetermined threshold value.
The present invention is that flow chart and/or block diagram with reference to method according to embodiments of the present invention, equipment (system) and computer program describe.It should be understood that can by the combination of the flow process in each flow process in computer program instructions flowchart and/or block diagram and/or square frame and flow chart and/or block diagram and/or square frame.These computer program instructions can be provided to the processor of general purpose computer, special-purpose computer, Embedded Processor or other programmable data processing device so that the function that the instruction performed by the processor of this computer or other programmable data processing device can be specified in a flow process in flowchart or multiple flow process and/or one square frame of block diagram or multiple square frame.
These computer program instructions may be alternatively stored in and can guide in the computer-readable memory that computer or other programmable data processing device work in a specific way, the instruction making to be stored in this computer-readable memory produces to include the manufacture of command device, and this command device realizes the function specified in one flow process of flow chart or multiple flow process and/or one square frame of block diagram or multiple square frame.
These computer program instructions also can be loaded in computer or other programmable data processing device, make on computer or other programmable devices, to perform sequence of operations step to produce computer implemented process, thus the instruction performed on computer or other programmable devices provides for realizing the step of function specified in a flow process of flow chart or a square frame of multiple flow process and/or block diagram or multiple square frame.
Although preferred embodiments of the present invention have been described, but those skilled in the art are once know basic creative concept, then these embodiments can be made other change and amendment.So, claims are intended to be construed to include preferred embodiment and fall into all changes and the amendment of the scope of the invention.
Obviously, the present invention can be carried out various change and modification without deviating from the spirit and scope of the present invention by those skilled in the art.So, if these amendments of the present invention and modification belong within the scope of the claims in the present invention and equivalent technologies thereof, then the present invention is also intended to comprise these change and modification.
Claims (12)
1. a typical abnormal sound detection method, it is characterised in that the method includes:
Collected sound signal, carries out pretreatment to described acoustical signal, it is determined that the useful signal fragment of described acoustical signal；
Obtaining the sound spectrogram of described useful signal fragment and determine identification eigenmatrix to be tested according to described sound spectrogram, described identification eigenmatrix to be tested is for representing the distribution situation of acoustical signal intensity of sound on time-frequency domain；
Calculate the similarity of described identification eigenmatrix to be tested and each standard identification eigenmatrix in abnormal sound model library, determine the abnormal sound type in described useful signal fragment according to result of calculation.
2. the method for claim 1, it is characterised in that the sound spectrogram of described acquisition described useful signal fragment also determines identification eigenmatrix to be tested according to described sound spectrogram, specifically includes:
The oscillogram of described useful signal fragment is converted to the sound spectrogram being made up of the frequency of described useful signal fragment, time and three dimensional information of intensity of sound；
Determine the identification eigenmatrix to be tested of intensity of sound distribution situation for characterizing described sound spectrogram；
The described identification eigenmatrix to be tested of described calculating and the similarity of each standard identification eigenmatrix in abnormal sound model library, determine the abnormal sound type in described useful signal fragment according to result of calculation, specifically include:
Calculate the similarity of described identification eigenmatrix to be tested and each standard identification eigenmatrix in abnormal sound model library, by the abnormal sound model maximum with described identification eigenmatrix similarity to be tested, it is determined that for the abnormal sound type of described useful signal fragment.
3. method as claimed in claim 2, it is characterised in that the described oscillogram by described useful signal fragment is converted to sound spectrogram, specifically includes:
Described useful signal fragment windowing is divided into several frames；
Each frame being carried out Short Time Fourier Transform, obtains the spectrum information of this frame, described spectrum information is for representing the relation between the frequency of this frame and intensity of sound；
Connecting the spectrum information of all frames, obtain the sound spectrogram of described useful signal fragment, described sound spectrogram is made up of several points, and (x, y) is used for representing that this point is in the x moment coordinate of any point, intensity of sound corresponding in y frequency.
4. method as claimed in claim 3, it is characterised in that described determine identification eigenmatrix to be tested according to described sound spectrogram, specifically includes:
Frame at each point in described sound spectrogram, frequency and intensity of sound, it is determined that sound spectrogram matrix；
For each frame, it is determined that intensity of sound value front K frequency values from big to small in described sound spectrogram matrix；
According to the relative position on frequency domain of each frequency values in described K frequency values, generate the identification eigenmatrix to be tested of intensity of sound distribution situation for characterizing described sound spectrogram.
5. the method as according to any one of claim 1-4, it is characterised in that the described identification eigenmatrix to be tested of described calculating and the similarity of each standard identification eigenmatrix in abnormal sound model library, specifically includes:
For the either standard identification eigenmatrix in described abnormal sound model library, element by the last string of the first row of the element of the first row first row of described standard identification eigenmatrix to described standard identification eigenmatrix, determine respectively and in described identification eigenmatrix to be tested difference between the value of the first row first row less than each element in the described standard identification eigenmatrix of predetermined threshold value, carry out each element present position respectively as starting point setting window operation, respectively obtain the window matrix that this window covers, this window is identical with the dimension of identification eigenmatrix to be tested；Calculate business's matrix of each window matrix and identification eigenmatrix to be tested, each row element in described business's matrix is multiplied by default weighted value respectively, obtain the similarity of each window matrix and described identification eigenmatrix to be tested；
Whole similarity according to each standard identification eigenmatrix in calculated described identification eigenmatrix to be tested and abnormal sound model library, is defined as the similarity of described standard identification eigenmatrix and described identification eigenmatrix to be tested by similarity maximum for the value in whole similarities.
6. the method as according to any one of claim 1-5, it is characterised in that described described acoustical signal is carried out pretreatment, it is determined that the useful signal fragment of described acoustical signal, specifically includes:
Described acoustical signal is carried out preemphasis process, obtains the acoustical signal after preemphasis；
Acoustical signal after preemphasis is carried out framing windowing process, useful signal fragment will be defined as in short time period self-energy value more than one section of acoustical signal of predetermined threshold value.
7. a typical abnormal sound detecting device, it is characterised in that this device includes:
Collecting unit, for collected sound signal, carries out pretreatment to described acoustical signal, it is determined that the useful signal fragment of described acoustical signal；Obtain the sound spectrogram of described useful signal fragment；
Determining unit, for determining the identification eigenmatrix to be tested of abnormal sound sound spectrogram, described identification eigenmatrix to be tested is for representing the distribution situation of acoustical signal intensity of sound on time-frequency domain；
Computing unit, for calculating the described to be tested similarity identifying eigenmatrix and each standard identification eigenmatrix in abnormal sound model library, determines the abnormal sound type in described useful signal fragment according to result of calculation.
8. device as claimed in claim 7, it is characterised in that described collecting unit specifically for: the oscillogram of described useful signal fragment is converted to the sound spectrogram being made up of the frequency of described useful signal fragment, time and three dimensional information of intensity of sound；
Described determine unit specifically for: determine the identification eigenmatrix of intensity of sound distribution situation for characterizing described sound spectrogram；
Described computing unit specifically for: calculate the similarity of each standard identification eigenmatrix in described identification eigenmatrix to be tested and abnormal sound model library, by the abnormal sound model maximum with described identification eigenmatrix similarity to be tested, it is determined that for the abnormal sound type of described useful signal fragment.
9. device as claimed in claim 8, it is characterised in that described collecting unit specifically for:
Described useful signal fragment windowing is divided into several frames；
Each frame being carried out Short Time Fourier Transform, obtains the spectrum information of this frame, described spectrum information is for representing the relation between the frequency of this frame and intensity of sound；Connecting the spectrum information of all frames, obtain the sound spectrogram of described useful signal fragment, described sound spectrogram is made up of several points, and (x, y) is used for representing that this point is in the x moment coordinate of any point, intensity of sound corresponding in y frequency.
10. device as claimed in claim 9, it is characterised in that described determine unit specifically for:
Frame at each point in described sound spectrogram, frequency and intensity of sound, it is determined that sound spectrogram matrix；
For each frame, it is determined that intensity of sound value front K frequency values from big to small in described sound spectrogram matrix；According to the relative position on frequency domain of each frequency values in described K frequency values, generate the identification eigenmatrix to be tested of intensity of sound distribution situation for characterizing described sound spectrogram.
11. the device as according to any one of claim 7-10, it is characterised in that described computing unit specifically for:
For the either standard identification eigenmatrix in described abnormal sound model library, element by the last string of the first row of the element of the first row first row of described standard identification eigenmatrix to described standard identification eigenmatrix, determine respectively and in described identification eigenmatrix to be tested difference between the value of the first row first row less than each element in the described standard identification eigenmatrix of predetermined threshold value, carry out each element present position respectively as starting point setting window operation, respectively obtain the window matrix that this window covers, this window is identical with the dimension of identification eigenmatrix to be tested；Calculate business's matrix of each window matrix and identification eigenmatrix to be tested, each row element in described business's matrix is multiplied by default weighted value respectively, obtain the similarity of each window matrix and described identification eigenmatrix to be tested；
Whole similarity according to each standard identification eigenmatrix in calculated described identification eigenmatrix to be tested and abnormal sound model library, is defined as the similarity of described standard identification eigenmatrix and described identification eigenmatrix to be tested by similarity maximum for the value in whole similarities.
12. the device as according to any one of claim 7-11, it is characterised in that described collecting unit specifically for:
Described acoustical signal is carried out preemphasis process, obtains the acoustical signal after preemphasis；Acoustical signal after preemphasis is carried out framing windowing process, useful signal fragment will be defined as in short time period self-energy value more than one section of acoustical signal of predetermined threshold value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410850883.5A CN105810213A (en) | 2014-12-30 | 2014-12-30 | Typical abnormal sound detection method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410850883.5A CN105810213A (en) | 2014-12-30 | 2014-12-30 | Typical abnormal sound detection method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105810213A true CN105810213A (en) | 2016-07-27 |
Family
ID=56421528
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410850883.5A Pending CN105810213A (en) | 2014-12-30 | 2014-12-30 | Typical abnormal sound detection method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105810213A (en) |
Cited By (32)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106531193A (en) * | 2016-10-31 | 2017-03-22 | 济南中维世纪科技有限公司 | Abnormal sound detection method and system of background noise adaption |
| CN106531195A (en) * | 2016-11-08 | 2017-03-22 | 北京理工大学 | Dialogue conflict detection method and device |
| CN106919662A (en) * | 2017-02-14 | 2017-07-04 | 复旦大学 | A kind of music recognition methods and system |
| CN106971499A (en) * | 2017-04-14 | 2017-07-21 | 北京克路德人工智能科技有限公司 | Intelligent monitor system based on auditory localization |
| CN107516534A (en) * | 2017-08-31 | 2017-12-26 | 广东小天才科技有限公司 | A kind of comparison method of voice messaging, device and terminal device |
| CN107622797A (en) * | 2017-09-26 | 2018-01-23 | 李涵之 | A kind of health based on sound determines system and method |
| CN108021635A (en) * | 2017-11-27 | 2018-05-11 | 腾讯科技（深圳）有限公司 | The definite method, apparatus and storage medium of a kind of audio similarity |
| CN108053842A (en) * | 2017-12-13 | 2018-05-18 | 电子科技大学 | Shortwave sound end detecting method based on image identification |
| CN108073856A (en) * | 2016-11-14 | 2018-05-25 | 华为技术有限公司 | The recognition methods of noise signal and device |
| WO2018133247A1 (en) * | 2017-01-20 | 2018-07-26 | 华为技术有限公司 | Abnormal sound detection method and apparatus |
| CN108376486A (en) * | 2018-01-17 | 2018-08-07 | 上海交通大学医学院附属瑞金医院 | A kind of method and system of trained sounding |
| CN108462932A (en) * | 2018-02-09 | 2018-08-28 | 北京云狐时代科技有限公司 | A kind of audio testing method of equipment, apparatus and system |
| CN108511002A (en) * | 2018-01-23 | 2018-09-07 | 努比亚技术有限公司 | The recognition methods of hazard event voice signal, terminal and computer readable storage medium |
| CN108712671A (en) * | 2018-04-26 | 2018-10-26 | 国政通科技股份有限公司 | Monitor video document play-back method, device and terminal device |
| CN108769572A (en) * | 2018-04-26 | 2018-11-06 | 国政通科技股份有限公司 | Monitor video file generated, device and terminal device |
| CN109087655A (en) * | 2018-07-30 | 2018-12-25 | 桂林电子科技大学 | A kind of monitoring of traffic route sound and exceptional sound recognition system |
| CN110425710A (en) * | 2019-08-30 | 2019-11-08 | 盈盛智创科技(广州)有限公司 | A kind of fault detection method of air-conditioning, device, equipment and storage medium |
| CN110677718A (en) * | 2019-09-27 | 2020-01-10 | 腾讯科技（深圳）有限公司 | Video identification method and device |
| CN110988137A (en) * | 2019-12-31 | 2020-04-10 | 四川长虹电器股份有限公司 | Abnormal sound detection system and method based on time-frequency domain characteristics |
| CN111326162A (en) * | 2020-04-15 | 2020-06-23 | 厦门快商通科技股份有限公司 | Voiceprint feature acquisition method, device and equipment |
| CN111699368A (en) * | 2019-05-22 | 2020-09-22 | 深圳市大疆创新科技有限公司 | Strike detection method, device, movable platform and computer readable storage medium |
| CN111768799A (en) * | 2019-03-14 | 2020-10-13 | 富泰华工业（深圳）有限公司 | Voice recognition method, voice recognition apparatus, computer apparatus, and storage medium |
| CN112164390A (en) * | 2020-09-16 | 2021-01-01 | 珠海格力电器股份有限公司 | Equipment fault identification method and system |
| CN112331208A (en) * | 2020-09-30 | 2021-02-05 | 音数汇元(上海)智能科技有限公司 | Personal safety monitoring method and device, electronic equipment and storage medium |
| CN112466290A (en) * | 2021-02-02 | 2021-03-09 | 鹏城实验室 | Abnormal sound detection model training method and device and computer storage medium |
| CN112710486A (en) * | 2019-10-24 | 2021-04-27 | 广东美的白色家电技术创新中心有限公司 | Equipment fault detection method, equipment fault detection device and computer storage medium |
| CN113432856A (en) * | 2021-06-28 | 2021-09-24 | 西门子电机(中国)有限公司 | Motor testing method, device, electronic equipment and storage medium |
| CN113514147A (en) * | 2021-05-18 | 2021-10-19 | 浙江吉利控股集团有限公司 | Vehicle noise identification method, system, device and computer readable storage medium |
| CN113689888A (en) * | 2021-07-30 | 2021-11-23 | 浙江大华技术股份有限公司 | Abnormal sound classification method, system, device and storage medium |
| CN114007132A (en) * | 2020-07-28 | 2022-02-01 | 中国电信股份有限公司 | Anomaly detection method, device and computer-readable storage medium |
| CN114131202A (en) * | 2021-12-30 | 2022-03-04 | 华中科技大学 | Method, system and device for extracting abnormal sound signal in laser-arc hybrid welding |
| CN112164390B (en) * | 2020-09-16 | 2024-07-05 | 珠海格力电器股份有限公司 | Equipment fault recognition method and system |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102522082A (en) * | 2011-12-27 | 2012-06-27 | 重庆大学 | Recognizing and locating method for abnormal sound in public places |
| US20130064379A1 (en) * | 2011-09-13 | 2013-03-14 | Northwestern University | Audio separation system and method |
| CN103295584A (en) * | 2012-02-24 | 2013-09-11 | 索尼公司 | Voice data detecting device, voice monitoring system and method thereof |
-
2014
- 2014-12-30 CN CN201410850883.5A patent/CN105810213A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130064379A1 (en) * | 2011-09-13 | 2013-03-14 | Northwestern University | Audio separation system and method |
| CN102522082A (en) * | 2011-12-27 | 2012-06-27 | 重庆大学 | Recognizing and locating method for abnormal sound in public places |
| CN103295584A (en) * | 2012-02-24 | 2013-09-11 | 索尼公司 | Voice data detecting device, voice monitoring system and method thereof |
Non-Patent Citations (2)
| Title |
|---|
| 周炯槃 等: "《通信原理（合订本）》", 30 November 2005 * |
| 孔旭等: "以声谱图相似度为度量的波形音乐检索", 《计算机工程与应用》 * |
Cited By (43)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106531193A (en) * | 2016-10-31 | 2017-03-22 | 济南中维世纪科技有限公司 | Abnormal sound detection method and system of background noise adaption |
| CN106531195A (en) * | 2016-11-08 | 2017-03-22 | 北京理工大学 | Dialogue conflict detection method and device |
| CN106531195B (en) * | 2016-11-08 | 2019-09-27 | 北京理工大学 | A kind of dialogue collision detection method and device |
| CN108073856A (en) * | 2016-11-14 | 2018-05-25 | 华为技术有限公司 | The recognition methods of noise signal and device |
| WO2018133247A1 (en) * | 2017-01-20 | 2018-07-26 | 华为技术有限公司 | Abnormal sound detection method and apparatus |
| CN106919662B (en) * | 2017-02-14 | 2021-08-31 | 复旦大学 | Music identification method and system |
| CN106919662A (en) * | 2017-02-14 | 2017-07-04 | 复旦大学 | A kind of music recognition methods and system |
| CN106971499A (en) * | 2017-04-14 | 2017-07-21 | 北京克路德人工智能科技有限公司 | Intelligent monitor system based on auditory localization |
| CN107516534B (en) * | 2017-08-31 | 2020-11-03 | 广东小天才科技有限公司 | Voice information comparison method and device and terminal equipment |
| CN107516534A (en) * | 2017-08-31 | 2017-12-26 | 广东小天才科技有限公司 | A kind of comparison method of voice messaging, device and terminal device |
| CN107622797A (en) * | 2017-09-26 | 2018-01-23 | 李涵之 | A kind of health based on sound determines system and method |
| CN107622797B (en) * | 2017-09-26 | 2020-07-28 | 李涵之 | Body condition determining system and method based on sound |
| CN108021635A (en) * | 2017-11-27 | 2018-05-11 | 腾讯科技（深圳）有限公司 | The definite method, apparatus and storage medium of a kind of audio similarity |
| CN108053842A (en) * | 2017-12-13 | 2018-05-18 | 电子科技大学 | Shortwave sound end detecting method based on image identification |
| CN108053842B (en) * | 2017-12-13 | 2021-09-14 | 电子科技大学 | Short wave voice endpoint detection method based on image recognition |
| CN108376486A (en) * | 2018-01-17 | 2018-08-07 | 上海交通大学医学院附属瑞金医院 | A kind of method and system of trained sounding |
| CN108511002B (en) * | 2018-01-23 | 2020-12-01 | 太仓鸿羽智能科技有限公司 | Method for recognizing sound signal of dangerous event, terminal and computer readable storage medium |
| CN108511002A (en) * | 2018-01-23 | 2018-09-07 | 努比亚技术有限公司 | The recognition methods of hazard event voice signal, terminal and computer readable storage medium |
| CN108462932A (en) * | 2018-02-09 | 2018-08-28 | 北京云狐时代科技有限公司 | A kind of audio testing method of equipment, apparatus and system |
| CN108712671A (en) * | 2018-04-26 | 2018-10-26 | 国政通科技股份有限公司 | Monitor video document play-back method, device and terminal device |
| CN108769572A (en) * | 2018-04-26 | 2018-11-06 | 国政通科技股份有限公司 | Monitor video file generated, device and terminal device |
| CN109087655A (en) * | 2018-07-30 | 2018-12-25 | 桂林电子科技大学 | A kind of monitoring of traffic route sound and exceptional sound recognition system |
| CN111768799A (en) * | 2019-03-14 | 2020-10-13 | 富泰华工业（深圳）有限公司 | Voice recognition method, voice recognition apparatus, computer apparatus, and storage medium |
| CN111699368A (en) * | 2019-05-22 | 2020-09-22 | 深圳市大疆创新科技有限公司 | Strike detection method, device, movable platform and computer readable storage medium |
| CN110425710A (en) * | 2019-08-30 | 2019-11-08 | 盈盛智创科技(广州)有限公司 | A kind of fault detection method of air-conditioning, device, equipment and storage medium |
| CN110677718A (en) * | 2019-09-27 | 2020-01-10 | 腾讯科技（深圳）有限公司 | Video identification method and device |
| CN110677718B (en) * | 2019-09-27 | 2021-07-23 | 腾讯科技（深圳）有限公司 | Video identification method and device |
| CN112710486B (en) * | 2019-10-24 | 2022-01-25 | 广东美的白色家电技术创新中心有限公司 | Equipment fault detection method, equipment fault detection device and computer storage medium |
| CN112710486A (en) * | 2019-10-24 | 2021-04-27 | 广东美的白色家电技术创新中心有限公司 | Equipment fault detection method, equipment fault detection device and computer storage medium |
| CN110988137A (en) * | 2019-12-31 | 2020-04-10 | 四川长虹电器股份有限公司 | Abnormal sound detection system and method based on time-frequency domain characteristics |
| CN111326162A (en) * | 2020-04-15 | 2020-06-23 | 厦门快商通科技股份有限公司 | Voiceprint feature acquisition method, device and equipment |
| CN114007132A (en) * | 2020-07-28 | 2022-02-01 | 中国电信股份有限公司 | Anomaly detection method, device and computer-readable storage medium |
| CN112164390A (en) * | 2020-09-16 | 2021-01-01 | 珠海格力电器股份有限公司 | Equipment fault identification method and system |
| CN112164390B (en) * | 2020-09-16 | 2024-07-05 | 珠海格力电器股份有限公司 | Equipment fault recognition method and system |
| CN112331208A (en) * | 2020-09-30 | 2021-02-05 | 音数汇元(上海)智能科技有限公司 | Personal safety monitoring method and device, electronic equipment and storage medium |
| CN112331208B (en) * | 2020-09-30 | 2024-07-05 | 音数汇元(上海)智能科技有限公司 | Personal safety monitoring method, device, electronic equipment and storage medium |
| CN112466290B (en) * | 2021-02-02 | 2021-05-28 | 鹏城实验室 | Abnormal sound detection model training method and device and computer storage medium |
| CN112466290A (en) * | 2021-02-02 | 2021-03-09 | 鹏城实验室 | Abnormal sound detection model training method and device and computer storage medium |
| CN113514147A (en) * | 2021-05-18 | 2021-10-19 | 浙江吉利控股集团有限公司 | Vehicle noise identification method, system, device and computer readable storage medium |
| CN113514147B (en) * | 2021-05-18 | 2023-10-31 | 浙江吉利控股集团有限公司 | Vehicle noise identification method, system, equipment and computer readable storage medium |
| CN113432856A (en) * | 2021-06-28 | 2021-09-24 | 西门子电机(中国)有限公司 | Motor testing method, device, electronic equipment and storage medium |
| CN113689888A (en) * | 2021-07-30 | 2021-11-23 | 浙江大华技术股份有限公司 | Abnormal sound classification method, system, device and storage medium |
| CN114131202A (en) * | 2021-12-30 | 2022-03-04 | 华中科技大学 | Method, system and device for extracting abnormal sound signal in laser-arc hybrid welding |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105810213A (en) | Typical abnormal sound detection method and device | |
| KR102339594B1 (en) | Object recognition method, computer device, and computer-readable storage medium | |
| Wang et al. | Deep extractor network for target speaker recovery from single channel speech mixtures | |
| CN108597496B (en) | Voice generation method and device based on generation type countermeasure network | |
| CN103646649A (en) | High-efficiency voice detecting method | |
| CN105118502A (en) | End point detection method and system of voice identification system | |
| CN104091603B (en) | Endpoint detection system based on fundamental frequency and calculation method thereof | |
| KR101863097B1 (en) | Apparatus and method for keyword recognition | |
| Pang | Spectrum energy based voice activity detection | |
| CN106023986B (en) | A kind of audio recognition method based on sound effect mode detection | |
| CN109887489A (en) | Speech dereverberation method based on the depth characteristic for generating confrontation network | |
| Torres-Boza et al. | Hierarchical sparse coding framework for speech emotion recognition | |
| CN106531195B (en) | A kind of dialogue collision detection method and device | |
| CN104103280B (en) | The method and apparatus of the offline speech terminals detection based on dynamic time consolidation algorithm | |
| CN109215634A (en) | A kind of method and its system of more word voice control on-off systems | |
| Zhang et al. | Speech emotion recognition using combination of features | |
| CN105679312A (en) | Phonetic feature processing method of voiceprint identification in noise environment | |
| Chuangsuwanich et al. | Robust Voice Activity Detector for Real World Applications Using Harmonicity and Modulation Frequency. | |
| CN108091340B (en) | Voiceprint recognition method, voiceprint recognition system, and computer-readable storage medium | |
| CN106548786A (en) | A kind of detection method and system of voice data | |
| CN110970020A (en) | Method for extracting effective voice signal by using voiceprint | |
| Fan et al. | Utterance-level permutation invariant training with discriminative learning for single channel speech separation | |
| CN108847218B (en) | Self-adaptive threshold setting voice endpoint detection method, equipment and readable storage medium | |
| Kaminski et al. | Automatic speaker recognition using a unique personal feature vector and Gaussian Mixture Models | |
| Chang et al. | Application of abnormal sound recognition system for indoor environment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20160727 |