CN106782583A - Robust scale contour feature extraction algorithm based on nuclear norm - Google Patents
Robust scale contour feature extraction algorithm based on nuclear norm Download PDFInfo
- Publication number
- CN106782583A CN106782583A CN201611132721.3A CN201611132721A CN106782583A CN 106782583 A CN106782583 A CN 106782583A CN 201611132721 A CN201611132721 A CN 201611132721A CN 106782583 A CN106782583 A CN 106782583A
- Authority
- CN
- China
- Prior art keywords
- matrix
- frequency
- algorithm
- singular value
- music signal
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Processing of the speech or voice signal to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/003—Changing voice quality, e.g. pitch or formants
- G10L21/007—Changing voice quality, e.g. pitch or formants characterised by the process used
- G10L21/013—Adapting to target pitch
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Processing of the speech or voice signal to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
- G10L21/0208—Noise filtering
- G10L21/0216—Noise filtering characterised by the method used for estimating noise
- G10L21/0224—Processing in the time domain
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Processing of the speech or voice signal to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
- G10L21/0208—Noise filtering
- G10L21/0216—Noise filtering characterised by the method used for estimating noise
- G10L21/0232—Processing in the frequency domain
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/03—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of extracted parameters
- G10L25/18—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of extracted parameters the extracted parameters being spectral information of each sub-band
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/45—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of analysis window
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Processing of the speech or voice signal to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/003—Changing voice quality, e.g. pitch or formants
- G10L21/007—Changing voice quality, e.g. pitch or formants characterised by the process used
- G10L21/013—Adapting to target pitch
- G10L2021/0135—Voice conversion or morphing
Abstract
The invention discloses a kind of specially robust scale contour feature extraction algorithm based on nuclear norm, step 1, music signal to be entered conversion;Step 2, windowing process is carried out to music signal, and carry out Fourier transform, obtain the time-frequency matrix of music signal, it is determined that initial beat point;Step 3, using nuclear norm constraint frequency spectrum low-rank is carried out to time-frequency rank of matrix;Simultaneously with the noise spot in a norm constraint matrix, low-rank is carried out to signal spectrum with following convex optimization problem, and remove noise;Step 4, during iterative constrained, using the low-rank characteristic of frequency spectrum, realize threshold adaptive adjustment algorithm;Step 5, effective dimension-reduction treatment is carried out to time-frequency matrix, obtain the chord feature of 12 dimensions.Compared with prior art, the present invention has extracted the chord feature of robustness;Significantly reduce the time of algorithm;The scale contour feature of the music signal of different type and style can exactly be recovered.
Description
Technical field
The invention belongs to the audio signal analysis field in computer audio system, more particularly to a kind of scale profile is special
Levy extraction algorithm.
The method that the present invention is provided can exactly recover the scale contour feature of the music signal of different type and style.
Background technology
The harmonic components of music are the important elements of music, are the important topics in music information retrieval field.Audio signal
Different frequency fundamental frequency and its harmonic components be to constitute chord and influence the important component of music color.In addition, different frequency
The key factor that chord is advanced is constituted during composition extension in time.Intuitively say, the music in the chord duration is in frequency
Domain can present certain structural --- low-rank characteristic.The chord feature extraction of music belongs to the audio in computer audio system
A part in signal analysis, this field mainly processes the various information separated from voice signal.Meanwhile,
The chord feature of music is also to extract the basis of some advanced music information.
The mid-level features of music refer to be extracted from audio signal, and can represent the information of audio signal, most
Eventually can be used as a part for advanced features.In recent years, many scholars propose various mid-level features that can characterize music.
That be wherein most widely used is exactly sound level contour feature (Pitch Class Profiles, PCP).However, due to original sound
Include voice, drumbeat, plosive and Gaussian noise in music signal so that quality and the sound to be analyzed of PCP characteristic performances
The type of music signal has very big relation.There are many scholars to propose the improvement project based on PCP, for example, what Gomez was proposed
The EPCP (Enhanced PCP) that HPCP (Harmonic PCP), Lee are proposed.These schemes all from change frequency domain extraction composition
Hand, so the superior performance for obtaining being suitable for specific music type feature.
In addition, from for chord traveling, because each chord has certain duration, during this period of time PCP is special
The stability levied determines the accuracy rate of chord identification.There are many scholars to propose the improvement of-chromagram that advanced based on PCP
Scheme.Fujisjima assumes that chord persistently counts frame, using sliding window mean filter, so as to reduce the influence of noise, and avoids
Chord frequently changes;GeoffroyPeeters avoids the frequent change of chord using sliding window medium filtering;Bello is false
It is indeclinable to determine chord in a beat, avoids chord frequently to change using beat simultaneous techniques.
Most of Beat-Tracking Model is made up of note end-point detection, end points intensity curve periodicity extraction two parts.No matter
Which kind of model, the basic goal of end-point detection is all the peak value for choosing effective end points curve, its be substantially extreme point whether
It is the clustering problem of beat point.
It can be seen that, it is structural that most of chord feature extraction scheme does not all consider that music signal is showed on frequency spectrum,
It is known it is assumed that so as to optimize chord feature using some simple processing methods using some.
The content of the invention
Based on prior art, the present invention proposes a kind of robust scale contour feature extraction algorithm based on nuclear norm, will
Chord feature extraction problem is converted into convex optimization problem, using nuclear norm constraint and a norm constraint, meanwhile, using the frequency of chord
The showed low-rank characteristic of spectrum, realizes threshold adaptive algorithm.
A kind of specially robust scale contour feature extraction algorithm based on nuclear norm of the invention, the algorithm includes following step
Suddenly:
Step 1, music signal to be entered is converted to the single pass standard audios of sample rate 22050Hz/16bit/, as
The data points that audio signal x (n) being referenced, wherein n are included by audio signal after conversion;
Step 2, windowing process is carried out to music signal x (n), window function is W (k), wherein k is wide for the window of window function
Degree, so as to obtain signal time-domain matrix Xk×m, wherein X·,m=x (k m/2:K m/2+m) W (k), m is the frame obtained after framing
Number, then carries out Fourier transform (Fourier Transform), obtains the time-frequency matrix D=F X of music signal, and wherein F is
Fourier transform matrix;
The harmonic components and noise that step 3, hypothesis audio signal frequency spectrum are included are separate, i.e. D=A+E, its
Middle matrix A represents that the harmonic components included in spectral matrix do the matrix for constituting, and E represents making an uproar included in spectral matrix
The matrix that sound composition is constituted;According to assumed above, the recovery of harmonics matrix A can be attributed to following convex optimization problem:
S.t.A+E=D
Wherein | | | |*The nuclear norm (nuclearnorm) of representing matrix, i.e. the singular value sum of matrix;||||1Represent square
One norm of battle array, i.e., all nonzero element sums;The matrix A separated is exactly the frequency spectrum after low-rank, and matrix E is then wrapped
Containing sparse big noise and other nonharmonic components, D is then the frequency spectrum of original music signal;
Step 4, during iterative constrained, using the low-rank characteristic of frequency spectrum, realize threshold adaptive adjustment algorithm;Specifically
Step is as follows:Initialization singular value interceptive value parameter μ, parameter lambda, iteration index k=0, provisional matrix Y0=D, E0It is full zero moment
Battle array;Carry out singular value decompositionObtain singular value matrix Σ;Then, from μkTo 1.5 μkDeng
Choose 20 data points in intervalWherein 1≤i≤20, for eachCarry out singular value decomposition inverse operationOnly it is distributed on several Frequency points due to harmonic components, therefore calculating matrixIn a certain row variance,
And fromIt is middle to choose during so that variance is maximum, corresponding index i, and makeComplete threshold value adaptive
Answer selection algorithm;Calculate the matrix that this step is obtainedUpdate With k=k+1 until convergence.
Step 5, effective dimension-reduction treatment is carried out to time-frequency matrix, obtain the chord feature of 12 dimensions.Under normal circumstances, it is stipulated that
Note A0Frequency on the basis of the frequency 440Hz at place, and pass throughObtain the frequency values at other notes.Wherein b be note with
A0Between interval difference.Then, by mapping equationCome to humorous
Each frequency content of ripple matrix A is mapped, so as to obtain robust scale Outline Feature Vector.Wherein x homographies A is each
The corresponding frequency values of row, and frefThen pass throughObtain.
Compared with prior art, the present invention effectively removes voice while music frequency domain original structure is not destroyed
Destruction with other noises to structure of a chord, has extracted the chord feature of robustness;Significantly reduce the time of algorithm;Energy
Recover the scale contour feature of the music signal of different type and style exactly.
Brief description of the drawings
Fig. 1 is overall flow figure of the present invention;
Fig. 2 is different type chord traveling figure;
Fig. 3 is the present invention and the Comparative result schematic diagram of other algorithms, 1, original ALM algorithms;2nd, APS-ALM algorithms;3、
ASP algorithms.
Specific embodiment
The present invention is described in further detail below in conjunction with the accompanying drawings.
Step 1, music signal conversion:Music signal to be entered is converted into sample rate 22050Hz/16bit/ single pass
Standard audio, as the audio signal being referenced.
Step 2, windowing process is carried out to music signal x (n), window function is W (k), wherein k is wide for the window of window function
Degree, so as to obtain signal time-domain matrix Xk×m, wherein X·,m=x (k m/2:K m/2+m) W (k), m is the frame obtained after framing
Number, then carries out Fourier transform (Fourier Transform), obtains the time-frequency matrix D=F X of music signal, and wherein F is
Fourier transform matrix;
Step 3, frequency spectrum low-rank and noise remove:Be can see from frequency spectrum, music signal mainly includes two kinds of compositions:
Harmonic components and sparse big noise.Harmonic components show as with obvious low-rank structure in structure;And sparse big noise master
Show as openness.Therefore, low-rank is carried out to signal spectrum with following convex optimization problem, and removes noise:
S.t.A+E=D
Wherein | | | |*The nuclear norm (nuclear norm) of representing matrix, i.e. the singular value sum of matrix;||||1Represent square
One norm of battle array, i.e., all nonzero element sums.
The matrix A separated is exactly the frequency spectrum after low-rank, and matrix E is then comprising sparse big noise and some its
His nonharmonic component, D is then the frequency spectrum of original music signal.
Step 4, PCP characteristics extractions:
(4-1) defines a frequency spectrum to the mapping matrix of PCP features, and matrix form is as follows:
Wherein 2 π ωj, 0≤j≤N-1 represents the frequency values representated by each frequency content in frequency spectrum, and N represents frequency spectrum and obtains
The frequency number scope for obtaining;And fi, 1≤i≤12 item represent the frequency values corresponding to 12 scales.
Wherein,
It is mapping function, according to the function obtained by twelve-tone equal temperament, with universality;
(4-2) obtains the chord traveling feature under low-rank is constrained, i.e. RPCP features by C=PA.
Present invention employs the test database of the chord automatic identification of international music information retrieval evaluation and test match (MIREX)
(Practice Data), the different snatch of music of 20 style of song rhythm altogether, meanwhile, each snatch of music has 39 or 40
Position expert carries out manual markings to the chordal type of the fragment.
It is to verify the validity of inventive algorithm, the robustness scale profile based on nuclear norm proposed by the invention is special
The influence that algorithm advances chord is levied to be compared with existing epidemic algorithms.Using smooth journey of the main scale in chord traveling
Spend to be quantitatively described algorithms of different, so as to judge the influence that algorithms of different advances chord.Result is as shown in Figure 2.From
From the point of view of experimental result, compared with other algorithms, the chord on main scale can be advanced more preferable smoothing effect to this algorithm, makes
Chord keeps stabilization, the frequent degree reduction of change within a certain period of time, and this chord identification to entire song has guidance to make
With.
In addition, in order to verify inhibitory action of this algorithm to noise, while accurate to chord identification in order to verify this algorithm
The influence of rate, we use template matching algorithm, and with Harmonic PCP the most popular features as a comparison, to this calculation
The validity of method is illustrated.Experimental result is as shown in table 1.The chord obtained by this algorithm is can be seen that from experimental result
Recognition accuracy improves 9% than Harmonic PCP.
Table 1, the average chord discrimination based on robust PCP and HPCP compare
Generally, the method for solving the nuclear norm constraint convex optimization problem of low-rank is to use augmented vector approach (ALM),
This method is relatively broad as the application of input for sparse matrix.But, with the increase of matrix dimensionality, the time will be big
It is big to increase.
According to chord feature it is exclusive the characteristics of, the present invention propose a kind of threshold adaptive adjustment algorithm based on chord feature
ASP-ALM (adaptive selectingparameterArgument Lagrange Multiplier, ASP-ALM) algorithm.
The algorithm flow is as follows:Initialization singular value interceptive value parameter μ, parameter lambda, iteration index k=0, provisional matrix Y0=D, E0For
Full null matrix;Carry out singular value decompositionObtain singular value matrix Σ;Then, from μkArrive
1.5μk20 data points are chosen at equal intervalsWherein 1≤i≤20, for eachCarry out singular value decomposition inverse operationOnly it is distributed on several Frequency points due to harmonic components, therefore calculating matrixIn a certain row variance,
And fromIt is middle to choose during so that variance is maximum, corresponding index i, and makeComplete threshold value adaptive
Answer selection algorithm;Calculate the matrix that this step is obtainedUpdate With
Major chord | Ab | A | Bb | B | C | Db | D | Eb | E | F | Gb | G |
RPCP (%) | 76.1 | 80 | 76.6 | 69.0 | 76.1 | 71.8 | 80.4 | 72.9 | 79.6 | 77.6 | 73.3 | 63.6 |
HPCP (%) | 73 | 78 | 63.8 | 66.7 | 71.7 | 69.2 | 78.4 | 64.6 | 71.4 | 61.2 | 68.9 | 63.6 |
Minor triad | Abm | Am | Bbm | Bm | Cm | Dbm | Dm | Ebm | Em | Fm | Gbm | Gm |
RPCP (%) | 84.8 | 74 | 69 | 63.4 | 88.2 | 87 | 75 | 43.6 | 65.2 | 80.4 | 76.3 | 66.7 |
HPCP (%) | 72.7 | 73.7 | 67.9 | 58.5 | 74.5 | 85.7 | 73.5 | 41 | 65.2 | 67.9 | 63.2 | 56.4 |
K=k+1 is until convergence.
Adaptive algorithm algorithm flow is as shown in Figure 1.Wherein μ enters the degree that row matrix is recovered in representing ALM algorithms.ASP-
ALM algorithms can greatly reduce time loss of the ALM algorithms in chord characteristic extraction procedure.
Test result contrast is as shown in Figure 3:It will be clear that time loss has obtained big reduction from result.
Claims (2)
1. it is a kind of specially based on nuclear norm robust scale contour feature extraction algorithm, it is characterised in that the algorithm include following step
Suddenly:
Step (1), music signal to be entered is converted to the single pass standard audios of sample rate 22050Hz/16bit/, as quilt
The data points that audio signal x (n) of reference, wherein n are included by audio signal after conversion;
Step (2), windowing process is carried out to music signal x (n), window function is W (k), wherein k is the window width of window function,
So as to obtain signal time-domain matrix Xk×m, wherein X·,m=x (k m/2:K m/2+m) W (k), m is the frame number obtained after framing,
Then Fourier transform (Fourier Transform) is carried out, the time-frequency matrix D=F X of music signal are obtained, wherein F is Fu
In leaf transformation matrix;
The harmonic components and noise that step (3), hypothesis audio signal frequency spectrum are included are separate, i.e. D=A+E, wherein
Matrix A represents that the harmonic components included in spectral matrix do the matrix for constituting, and E represents the noise included in spectral matrix
The matrix that composition is constituted;According to assumed above, the recovery of harmonics matrix A can be attributed to following convex optimization problem:
S.t.A+E=D
Wherein | | | |*The nuclear norm (nuclear norm) of representing matrix, i.e. the singular value sum of matrix;|| ||1Representing matrix
A norm, i.e., all nonzero element sums;The matrix A separated is exactly the frequency spectrum after low-rank, and matrix E is then included
Sparse big noise and other nonharmonic components, D are then the frequency spectrums of original music signal;
Step (4), during iterative constrained, using the low-rank characteristic of frequency spectrum, realize threshold adaptive adjustment algorithm;Specific step
It is rapid as follows:Initialization singular value interceptive value parameter μ, parameter lambda, iteration index k=0, provisional matrix Y0=D, E0It is full zero moment
Battle array;Carry out singular value decompositionObtain singular value matrix Σ;Then, from μkTo 1.5 μkDeng
Choose 20 data points in intervalWherein 1≤i≤20, for eachCarry out singular value decomposition inverse operationOnly it is distributed on several Frequency points due to harmonic components, therefore calculating matrixIn a certain row variance, and
FromIt is middle to choose during so that variance is maximum, corresponding index i, and makeComplete threshold adaptive choosing
Select algorithm;Calculate the matrix that this step is obtainedUpdateYk+1=
Yk+μk(D-Ak+1-Ek+1) and k=k+1 until convergence.
Step (5), effective dimension-reduction treatment is carried out to time-frequency matrix, obtain the chord feature of 12 dimensions.Under normal circumstances, it is stipulated that sound
Symbol A0Frequency on the basis of the frequency 440Hz at place, and pass throughObtain the frequency values at other notes.Wherein b is note and A0
Between interval difference.Then, by mapping equationCome to harmonic wave square
Each frequency content of battle array A is mapped, so as to obtain robust scale Outline Feature Vector.Wherein x homographies A is per a line institute
Corresponding frequency values, and frefThen pass throughObtain.
2. a kind of robust scale contour feature extraction algorithm based on nuclear norm as claimed in claim 1, it is characterised in that institute
The threshold adaptive adjustment algorithm algorithm is stated to comprise the following steps:
Initialization singular value interceptive value parameter μ, parameter lambda, iteration index k=0, provisional matrix Y0=D, E0It is full null matrix;Enter
Row singular value decompositionObtain singular value matrix Σ;Then, from μkTo 1.5 μkAt equal intervals
Choose 20 data pointsWherein 1≤i≤20, for eachCarry out singular value decomposition inverse operation
Only it is distributed on several Frequency points due to harmonic components, therefore calculating matrixIn a certain row variance, and from
It is middle to choose during so that variance is maximum, corresponding index i, and makeComplete threshold adaptive selection algorithm;Calculate
The matrix that this step is obtainedUpdateYk+1=Yk+μk(D-
Ak+1-Ek+1) and k=k+1 until convergence.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611132721.3A CN106782583B (en) | 2016-12-09 | 2016-12-09 | Robust scale contour feature extraction algorithm based on nuclear norm |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611132721.3A CN106782583B (en) | 2016-12-09 | 2016-12-09 | Robust scale contour feature extraction algorithm based on nuclear norm |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106782583A true CN106782583A (en) | 2017-05-31 |
CN106782583B CN106782583B (en) | 2020-04-28 |
Family
ID=58879705
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201611132721.3A Expired - Fee Related CN106782583B (en) | 2016-12-09 | 2016-12-09 | Robust scale contour feature extraction algorithm based on nuclear norm |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106782583B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111753127A (en) * | 2019-03-29 | 2020-10-09 | 阿里巴巴集团控股有限公司 | Music information processing and recommending method and device |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102129456A (en) * | 2011-03-09 | 2011-07-20 | 天津大学 | Method for monitoring and automatically classifying music factions based on decorrelation sparse mapping |
CN103714806A (en) * | 2014-01-07 | 2014-04-09 | 天津大学 | Chord recognition method combining SVM with enhanced PCP |
CN104361611A (en) * | 2014-11-18 | 2015-02-18 | 南京信息工程大学 | Group sparsity robust PCA-based moving object detecting method |
CN104395953A (en) * | 2012-04-30 | 2015-03-04 | 诺基亚公司 | Evaluation of beats, chords and downbeats from a musical audio signal |
CN104867162A (en) * | 2015-05-26 | 2015-08-26 | 南京信息工程大学 | Motion object detection method based on multi-component robustness PCA |
CN104978582A (en) * | 2015-05-15 | 2015-10-14 | 苏州大学 | Contour chord angle feature based identification method for blocked target |
CN106056607A (en) * | 2016-05-30 | 2016-10-26 | 天津城建大学 | Monitoring image background modeling method based on robustness principal component analysis |
-
2016
- 2016-12-09 CN CN201611132721.3A patent/CN106782583B/en not_active Expired - Fee Related
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102129456A (en) * | 2011-03-09 | 2011-07-20 | 天津大学 | Method for monitoring and automatically classifying music factions based on decorrelation sparse mapping |
CN104395953A (en) * | 2012-04-30 | 2015-03-04 | 诺基亚公司 | Evaluation of beats, chords and downbeats from a musical audio signal |
CN103714806A (en) * | 2014-01-07 | 2014-04-09 | 天津大学 | Chord recognition method combining SVM with enhanced PCP |
CN104361611A (en) * | 2014-11-18 | 2015-02-18 | 南京信息工程大学 | Group sparsity robust PCA-based moving object detecting method |
CN104978582A (en) * | 2015-05-15 | 2015-10-14 | 苏州大学 | Contour chord angle feature based identification method for blocked target |
CN104867162A (en) * | 2015-05-26 | 2015-08-26 | 南京信息工程大学 | Motion object detection method based on multi-component robustness PCA |
CN106056607A (en) * | 2016-05-30 | 2016-10-26 | 天津城建大学 | Monitoring image background modeling method based on robustness principal component analysis |
Non-Patent Citations (9)
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111753127A (en) * | 2019-03-29 | 2020-10-09 | 阿里巴巴集团控股有限公司 | Music information processing and recommending method and device |
Also Published As
Publication number | Publication date |
---|---|
CN106782583B (en) | 2020-04-28 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103714806B (en) | A kind of combination SVM and the chord recognition methods of in-dash computer P feature | |
EP3440672B1 (en) | Estimating pitch of harmonic signals | |
CN106935248A (en) | A kind of voice similarity detection method and device | |
CN107146601A (en) | A kind of rear end i vector Enhancement Methods for Speaker Recognition System | |
CN108922541B (en) | Multi-dimensional characteristic parameter voiceprint recognition method based on DTW and GMM models | |
CN109256144B (en) | Speech enhancement method based on ensemble learning and noise perception training | |
US20110058685A1 (en) | Method of separating sound signal | |
CN102664010B (en) | Robust speaker distinguishing method based on multifactor frequency displacement invariant feature | |
CN103236260A (en) | Voice recognition system | |
CN110599987A (en) | Piano note recognition algorithm based on convolutional neural network | |
CN105261367A (en) | Identification method of speaker | |
CN104240720A (en) | Voice emotion recognition method based on multi-fractal and information fusion | |
CN101894548A (en) | Modeling method and modeling device for language identification | |
CN111128209A (en) | Speech enhancement method based on mixed masking learning target | |
CN104505100A (en) | Non-supervision speech enhancement method based robust non-negative matrix decomposition and data fusion | |
CN103559232A (en) | Music humming searching method conducting matching based on binary approach dynamic time warping | |
CN108520752A (en) | A kind of method for recognizing sound-groove and device | |
CN111508518A (en) | Single-channel speech enhancement method based on joint dictionary learning and sparse representation | |
CN110136730A (en) | A kind of automatic allocation system of piano harmony and method based on deep learning | |
CN111508504B (en) | Speaker recognition method based on auditory center perception mechanism | |
Rafii et al. | Combining Modeling Of Singing Voice And Background Music For Automatic Separation Of Musical Mixtures. | |
Brillinger | Some river wavelets | |
CN106782583A (en) | Robust scale contour feature extraction algorithm based on nuclear norm | |
CN107430850A (en) | Determine the feature of harmonic signal | |
CN108564967A (en) | Mel energy vocal print feature extracting methods towards crying detecting system |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20200428 Termination date: 20211209 |
|
CF01 | Termination of patent right due to non-payment of annual fee |