CN107133643A - Note signal sorting technique based on multiple features fusion and feature selecting - Google Patents
Note signal sorting technique based on multiple features fusion and feature selecting Download PDFInfo
- Publication number
- CN107133643A CN107133643A CN201710299166.1A CN201710299166A CN107133643A CN 107133643 A CN107133643 A CN 107133643A CN 201710299166 A CN201710299166 A CN 201710299166A CN 107133643 A CN107133643 A CN 107133643A
- Authority
- CN
- China
- Prior art keywords
- feature
- classification
- classification accuracy
- mrow
- note 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
A kind of note signal sorting technique based on multiple features fusion and feature selecting, including：Song is divided, and a song is divided into many segments for being easy to extract feature；Feature extraction, extracts the feature of reflection note signal information in terms of time domain and frequency domain two；Feature selecting, removes the redundancy feature in feature；Different features, is fused into a fusion feature by multiple features fusion according to certain ratio；Classification and most ballots, are classified using extreme learning machine, and carrying out most ballots according to classification results obtains final classification results.Feature selecting is introduced into note signal classification by the present invention, and selection most simplifies maximally effective character subset and avoids dimension disaster, improves classification accuracy, so that the mode classification with faster pace of learning on the premise of ensureing that study accuracy rate does not decline.Classification accuracy compared to traditional multiple features fusion improves three percentage points.
Description
Technical field
The present invention relates to a kind of note signal sorting technique.It is more particularly to a kind of to be based on multiple features fusion and feature selecting
Note signal sorting technique.
Background technology
The process of note signal classification mainly includes feature extraction, grader and classified two parts.This is note signal classification
The most popular framework in field, and classification accuracy is also higher.How current most researchers are mainly around extracting
The feature of signal message and two aspects of selection of grader can more preferably be reflected and how to improve the accuracy rate of classification to study.
For reflecting that the feature of note signal information is broadly divided into three classes, including tamber characteristic, rhythm characteristic and pitch spy
Levy.In addition, there are some good features to reflect the feature of note signal, such as frequency spectrum divergence based on interval,
Modulation spectrum divergence based on interval etc..
Have many graders can be used to determine song type, for example gauss hybrid models, SVMs technology,
HMM etc..
In traditional feature selecting, might have Partial Feature is that redundancy is either incoherent.Redundancy feature meeting
So that the feature extracted has carried out excessive expression in a certain respect to note signal, and uncorrelated features in itself will not be to classification
As a result any influence is produced, that is to say, that done many idle works to analyze and process this feature.Although multiple features fusion is helped
In the characteristic of more fully expression note signal, but generate very high dimension.Higher dimension often causes dimension
The overlapping of part is might have between disaster, i.e. different characteristic, equivalent to having carried out in a certain respect too much to note signal
Expression, and in terms of have ignored other, this can reduce the accuracy rate of classification again on the contrary.
The content of the invention
The technical problems to be solved by the invention are to provide one kind to be had on the premise of ensureing that study accuracy rate does not decline
The faster note signal sorting technique based on multiple features fusion and feature selecting of pace of learning.
The technical solution adopted in the present invention is：A kind of note signal classification side based on multiple features fusion and feature selecting
Method, comprises the following steps：
1) song is divided, and a song is divided into many segments for being easy to extract feature；
2) feature extraction, extracts the feature of reflection note signal information in terms of time domain and frequency domain two；
3) feature selecting, removes the redundancy feature in feature；
4) different features, is fused into a fusion feature by multiple features fusion according to certain ratio；
5) classification and most ballots, are classified using extreme learning machine, and carrying out most ballots according to classification results obtains
Final classification results.
Step 1) described in song divide, be that one section of song is divided into many sections of textures windows mutually covered, then will be every
Section textures windows are divided into many sections of analysis windows mutually covered.
Step 2) described in terms of time domain and frequency domain two come extract reflection note signal information feature, be analysis
Window extracts the feature in short-term of reflection note signal in terms of time domain and frequency domain two；The textures windows are extracted
Reflect the feature of note signal change, specifically all averages and variance that feature is extracted in analysis window.
Step 3) include：
(1) feature of each information is classified by musical genre, and calculates classification accuracy, find out classification accurate
Rate highest feature is put into character subset；
(2) each remaining feature is calculated into classification accuracy together with the feature in character subset respectively, found out point
Class accuracy rate highest feature is put into character subset；
(3) repeat step 2), M feature and a classification accuracy are obtained in character subset；
(4) to calculating classification accuracy together per M-1 feature in character subset, M classification accuracy is obtained；
(5) highest point is found out in an existing classification accuracy in M classification accuracy and character subset
Class accuracy rate；
(6) by step 5) in highest classification accuracy and the correspondence highest classification accuracy characteristic set collectively as
New character subset；
(7) when highest classification accuracy is step 3) in existing classification accuracy when, return to step 3), otherwise return
Step 4), until the characteristic set in character subset and highest classification accuracy no longer change.
Classification described in (1) step is classified using extreme learning machine.
Described calculating classification accuracy just uses equation below：
In formula, N is the number of musical genre, and data is confusion matrix, datai,jThe value arranged for the i-th row of confusion matrix jth.
Step 4) include：Classified by musical genre and classification accuracy calculating is carried out to each feature, will be classified
Accuracy rate is set to from high to low：To the distribution of classification accuracy highest feature with 1
Weight, secondary high feature distribute withWeight, weight of all features in Fusion Features is obtained by that analogy, to institute
There is feature according to the weight merge obtained fusion feature.
Step 5) described in most ballots, be the school that each textures windows is predicted by extreme learning machine, occupy window
The most school of mouth is the bent school of this song, as classification results.
The note signal sorting technique based on multiple features fusion and feature selecting of the present invention, musical sound is introduced by feature selecting
In Modulation recognition, selection most simplifies maximally effective character subset and avoids dimension disaster, improves classification accuracy, so as to protect
Card study accuracy rate has the mode classification of faster pace of learning on the premise of not declining.Compared to point of traditional multiple features fusion
Class accuracy rate improves three percentage points.Therefore the multiple features fusion method proposed by the present invention based on classification accuracy can be
Classification accuracy is improved to a certain extent.
Brief description of the drawings
Fig. 1 is the flow chart of the note signal sorting technique of the invention based on multiple features fusion and feature selecting；
Fig. 2 is the classification accuracy comparison diagram of the multiple features fusion of single feature, multiple features fusion and the present invention；
In Fig. 2, feature 1 is modulation spectrum divergence；2 be modulation spectrum flatness and modulation spectrum peak value；3 be mel-frequency
Cepstrum coefficient；4 be the frequency spectrum divergence based on interval；5 be standardization sound spectrum envelope；6 be zero-crossing rate；7 be spectrum peak because
Son；8 be frequency spectrum rate of descent；9 be spectrum Pingdu；10 be frequency spectrum pitch chroma；11 be spectrum slope；12 be auto-correlation coefficient；13 are
Maximum auto-correlation；14 be peak envelope；15 be prediction rate；16 be standard deviation；17 be traditional multiple features fusion；After 18 are improvement
Multiple features fusion.
Fig. 3 is the optimal classification accuracy rate comparison diagram for the feature for selecting varying number.
Embodiment
The present invention is classified based on multiple features fusion and the note signal of feature selecting with reference to embodiment and accompanying drawing
Method is described in detail.
The note signal sorting technique based on multiple features fusion and feature selecting of the present invention, musical sound is introduced by feature selecting
In the automatic classification of signal, feature selecting is carried out using the sequence floating selection algorithm based on Wrapper frameworks, additionally carried
Go out a kind of new multiple features fusion mode based on classification accuracy, it is effective for the accuracy rate and efficiency of raising note signal classification
Ground provides help.
As shown in figure 1, the note signal sorting technique based on multiple features fusion and feature selecting of the present invention, including it is as follows
Step：
1) song is divided, and a song is divided into many segments for being easy to extract feature；
Described song is divided, and is that one section of song is divided into many sections of textures windows mutually covered, then by every section of texture
Window is divided into many sections of analysis windows mutually covered.
2) feature extraction, extracts the feature of reflection note signal information in terms of time domain and frequency domain two；
It is described to extract the feature of reflection note signal information in terms of time domain and frequency domain two, be analysis window from
Time domain and the aspect of frequency domain two extract the feature in short-term of reflection note signal, signal from time domain to frequency domain on conversion can lead to
Discrete Fourier transform is crossed to realize；It is characterized in the feature for reflecting note signal change that the textures windows, which are extracted, is specifically
All averages and variance that feature is extracted in analysis window.
3) feature selecting, removes the redundancy feature in feature；
Feature selecting algorithm is usually by the searching algorithm of generation different characteristic subset, some specific character subset is entered
The evaluation method and the part of stop condition three that row is evaluated are constituted.According to whether needing to divide character subset in feature selecting
Feature selecting, can be divided into two classes by class：Filter types and Wrapper types.The feature selecting of Wrapper types is direct
To obtain higher classification accuracy as target, so the result that this its usual feature selecting than Filter type is obtained is more
It is good.The feature selecting algorithm that the present invention is used is that the increasing l based on Wrapper types removes r selection algorithms.
The feature selecting of the present invention is to use sequence floating selection algorithm, and sequence floating selection algorithm also includes sequence and floated
Forward direction is selected and sequence is floated and selects damp two kinds of algorithms backward.Since empty set, then it is equally to selection algorithm that sequence is floated preceding
A feature is selected in non-selected feature every time so that the characteristic function value for adding the character subset after this feature reaches
It is optimal, the then reselection a subset in the character subset selected so that remove the character subset after this character subset
Characteristic function value be optimal.Therefore, if the characteristic function of character subset is unable to re-optimization at all, then need not just enter
Row trace-back process.Sequence is floated selects damp algorithm to be also same reason backward.Trace-back process is dynamic control in algorithmic procedure
, so any parameter need not be set at all.
The feature selecting of the present invention is specifically included：
(1) feature of each information is classified by musical genre, and calculates classification accuracy, find out classification accurate
Rate highest feature is put into character subset, and described classification is classified using extreme learning machine (ELM)；
(2) each remaining feature is calculated into classification accuracy together with the feature in character subset respectively, found out point
Class accuracy rate highest feature is put into character subset；
Described calculating classification accuracy just uses equation below：
In formula, N is the number of musical genre, and data is confusion matrix, datai,jThe value arranged for the i-th row of confusion matrix jth.
(3) repeat step 2), M feature and a classification accuracy are obtained in character subset；
(4) to calculating classification accuracy together per M-1 feature in character subset, M classification accuracy is obtained；
(5) highest point is found out in an existing classification accuracy in M classification accuracy and character subset
Class accuracy rate；
(6) by step 5) in highest classification accuracy and the correspondence highest classification accuracy characteristic set collectively as
New character subset；
(7) when highest classification accuracy is step 3) in existing classification accuracy when, return to step 3), otherwise return
Step 4), until the characteristic set in character subset and highest classification accuracy no longer change.
4) different features, is fused into a fusion feature by multiple features fusion according to certain ratio；Including：
Classified by musical genre and classification accuracy calculating is carried out to each feature, by classification accuracy by up to
It is low to be set to：To the distribution of classification accuracy highest feature with 1 weight, secondary high spy
Levy distribution withWeight, weight of all features in Fusion Features is obtained by that analogy, to all features according to institute
State weight and merge obtained fusion feature.
5) classification and most ballots, are classified using extreme learning machine (ELM), and most ballots are carried out according to classification results
Obtain final classification results.Described most ballots, are the schools that each textures windows is predicted by extreme learning machine, account for
It is the bent school of this song, as classification results to have the most school of window.
The size of analysis window is 23ms in the embodiment of the present invention, and the size of textures windows is 9s, from time domain and frequency domain two
Individual aspect reflects the feature of note signal information to extract, and 16 kinds of features are extracted altogether.So as to each section according to a first song
Textures windows feature determines the school of this first song.
As can be seen from Figure 2 the of the invention of the present invention is classified based on multiple features fusion and the note signal of feature selecting
Method can improve classification accuracy really.The classification accuracy classified to single feature is also list in figure, it can be seen that
Worst classification accuracy has 31%, and best classification accuracy has 81%.The method of the present invention has also obtained experimental result
Confirmation, it can be seen that the classification accuracy that traditional multiple features fusion is obtained only has 82%, and what the method for the present invention was obtained
Classification accuracy but has 85:As many as 6%, the classification accuracy compared to traditional multiple features fusion improves three percentage points.Cause
The note signal sorting technique based on multiple features fusion and feature selecting of this present invention can be improved to a certain extent really
Classification accuracy.
Feature selecting is carried out present invention employs sequence floating algorithm based on Wrapper types, i.e., often calculates one
Individual alternative character subset, can all classify to this character subset.The present invention calculates the spy for extracting varying number respectively
The classification accuracy obtained is as shown in Figure 3.
As can be seen from Figure 3 feature selection approach of the present invention can improve classification accuracy really.Work as fusion
The classification accuracy all obtained during 16 kinds of features only has 82%, and classification accuracy is up to after feature selecting is carried out
As many as 89.5%, improve seven percentage points compared to without the classification accuracy for carrying out feature selecting.This also illustrates to carry really
There are redundancy feature or uncorrelated features in the 16 kinds of features taken.It can see by observation, 4 kinds of features of selection and selection 11
89.5% classification accuracy can all be reached by planting feature, but according to the more few better principle of feature quantity, 4 kinds of present invention selection
Feature, because feature quantity is fewer, the classification time is shorter, and the classification time of 4 kinds of features of selection is 107s, selection 16
The classification time for planting feature is 348s, and the time of classifying in the case where ensureing that classification accuracy does not decline is more short better, so choosing
This 4 kinds of features have been selected to be classified.
Claims (8)
1. a kind of note signal sorting technique based on multiple features fusion and feature selecting, it is characterised in that comprise the following steps：
1) song is divided, and a song is divided into many segments for being easy to extract feature；
2) feature extraction, extracts the feature of reflection note signal information in terms of time domain and frequency domain two；
3) feature selecting, removes the redundancy feature in feature；
4) different features, is fused into a fusion feature by multiple features fusion according to certain ratio；
5) classification and most ballots, are classified using extreme learning machine, and carrying out most ballots according to classification results obtains final
Classification results.
2. the note signal sorting technique according to claim 1 based on multiple features fusion and feature selecting, its feature exists
In step 1) described in song divide, be that one section of song is divided into many sections of textures windows mutually covered, then by every section of texture
Window is divided into many sections of analysis windows mutually covered.
3. the note signal sorting technique according to claim 1 based on multiple features fusion and feature selecting, its feature exists
In step 2) described in terms of time domain and frequency domain two come extract reflection note signal information feature, be in analysis window
The feature in short-term of reflection note signal is extracted in terms of time domain and frequency domain two；It is characterized in reflection that the textures windows, which are extracted,
The feature of note signal change, specifically all averages and variance that feature is extracted in analysis window.
4. the note signal sorting technique according to claim 1 based on multiple features fusion and feature selecting, its feature exists
In step 3) include：
(1) feature of each information is classified by musical genre, and calculates classification accuracy, find out classification accuracy most
High feature is put into character subset；
(2) each remaining feature is calculated into classification accuracy together with the feature in character subset respectively, finds out classification accurate
True rate highest feature is put into character subset；
(3) repeat step 2), M feature and a classification accuracy are obtained in character subset；
(4) to calculating classification accuracy together per M-1 feature in character subset, M classification accuracy is obtained；
(5) a highest classification is found out in an existing classification accuracy in M classification accuracy and character subset accurate
True rate；
(6) by step 5) in highest classification accuracy and the correspondence highest classification accuracy characteristic set collectively as new
Character subset；
(7) when highest classification accuracy is step 3) in existing classification accuracy when, return to step 3), otherwise return to step
4), until the characteristic set in character subset and highest classification accuracy no longer change.
5. the note signal sorting technique according to claim 4 based on multiple features fusion and feature selecting, its feature exists
In the classification described in (1) step is classified using extreme learning machine.
6. the note signal sorting technique according to claim 4 based on multiple features fusion and feature selecting, its feature exists
In described calculating classification accuracy just uses equation below：
<mrow>
<mi>A</mi>
<mi>c</mi>
<mi>c</mi>
<mi>u</mi>
<mi>r</mi>
<mi>a</mi>
<mi>c</mi>
<mi>y</mi>
<mo>=</mo>
<mfrac>
<mrow>
<msubsup>
<mi>&Sigma;</mi>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>N</mi>
</msubsup>
<msub>
<mi>data</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>i</mi>
</mrow>
</msub>
</mrow>
<mrow>
<msubsup>
<mi>&Sigma;</mi>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>N</mi>
</msubsup>
<msubsup>
<mi>&Sigma;</mi>
<mrow>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>N</mi>
</msubsup>
<msub>
<mi>data</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
</msub>
</mrow>
</mfrac>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</mrow>
In formula, N is the number of musical genre, and data is confusion matrix, datai,jThe value arranged for the i-th row of confusion matrix jth.
7. the note signal sorting technique according to claim 1 based on multiple features fusion and feature selecting, its feature exists
In step 4) include：Classified by musical genre and classification accuracy calculating is carried out to each feature, by classification accuracy
It is set to from high to low：It is secondary to the distribution of classification accuracy highest feature with 1 weight
High feature distribute withWeight, weight of all features in Fusion Features is obtained by that analogy, to all features
According to the weight merge obtained fusion feature.
8. the note signal sorting technique according to claim 1 based on multiple features fusion and feature selecting, its feature exists
In step 5) described in most ballots, be the school that each textures windows is predicted by extreme learning machine, occupy window most
School be the bent school of this song, as classification results.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710299166.1A CN107133643A (en) | 2017-04-29 | 2017-04-29 | Note signal sorting technique based on multiple features fusion and feature selecting |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710299166.1A CN107133643A (en) | 2017-04-29 | 2017-04-29 | Note signal sorting technique based on multiple features fusion and feature selecting |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107133643A true CN107133643A (en) | 2017-09-05 |
Family
ID=59716507
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710299166.1A Pending CN107133643A (en) | 2017-04-29 | 2017-04-29 | Note signal sorting technique based on multiple features fusion and feature selecting |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107133643A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107832787A (en) * | 2017-10-31 | 2018-03-23 | 杭州电子科技大学 | Recognition Method of Radar Emitters based on bispectrum own coding feature |
| CN109299727A (en) * | 2018-08-04 | 2019-02-01 | 辽宁大学 | The improvement extreme learning machine method for diagnosing faults of signal reconstruct |
| CN111326164A (en) * | 2020-01-21 | 2020-06-23 | 大连海事大学 | Semi-supervised music melody extraction method |
| WO2024021882A1 (en) * | 2022-07-28 | 2024-02-01 | 腾讯科技（深圳）有限公司 | Audio data processing method and apparatus, and computer device and storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101840516A (en) * | 2010-04-27 | 2010-09-22 | 上海交通大学 | Feature selection method based on sparse fraction |
| CN104573114A (en) * | 2015-02-04 | 2015-04-29 | 苏州大学 | Music classification method and device |
| CN105139023A (en) * | 2015-07-24 | 2015-12-09 | 福州大学 | Seed identification method based on multi-scale feature fusion and extreme learning machine |
-
2017
- 2017-04-29 CN CN201710299166.1A patent/CN107133643A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101840516A (en) * | 2010-04-27 | 2010-09-22 | 上海交通大学 | Feature selection method based on sparse fraction |
| CN104573114A (en) * | 2015-02-04 | 2015-04-29 | 苏州大学 | Music classification method and device |
| CN105139023A (en) * | 2015-07-24 | 2015-12-09 | 福州大学 | Seed identification method based on multi-scale feature fusion and extreme learning machine |
Non-Patent Citations (4)
| Title |
|---|
| JIANRONG WANG ET AL: "Chinese opera genre classification based on multi-feature fusion and extreme learning machine", 《2015 ASIA-PACIFIC SIGNAL AND INFORMATION PROCESSING ASSOCIATION ANNUAL SUMMIT AND CONFERENCE (APSIPA)》 * |
| 张燕等: "基于MFCC和HMM的音乐分类方法研究", 《南京师范大学学报（工程技术版）》 * |
| 洪文学等: "《可视化模式识别》", 31 July 2014 * |
| 赵亮: "改进极限学习机的电子音乐分类模型", 《现代电子技术》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107832787A (en) * | 2017-10-31 | 2018-03-23 | 杭州电子科技大学 | Recognition Method of Radar Emitters based on bispectrum own coding feature |
| CN107832787B (en) * | 2017-10-31 | 2020-09-22 | 杭州电子科技大学 | Radar radiation source identification method based on bispectrum self-coding characteristics |
| CN109299727A (en) * | 2018-08-04 | 2019-02-01 | 辽宁大学 | The improvement extreme learning machine method for diagnosing faults of signal reconstruct |
| CN111326164A (en) * | 2020-01-21 | 2020-06-23 | 大连海事大学 | Semi-supervised music melody extraction method |
| CN111326164B (en) * | 2020-01-21 | 2023-03-21 | 大连海事大学 | Semi-supervised music theme extraction method |
| WO2024021882A1 (en) * | 2022-07-28 | 2024-02-01 | 腾讯科技（深圳）有限公司 | Audio data processing method and apparatus, and computer device and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Kim et al. | Sample-level CNN architectures for music auto-tagging using raw waveforms | |
| Richard et al. | Temporal action detection using a statistical language model | |
| CN109543740B (en) | Target detection method based on generation countermeasure network | |
| CN107133643A (en) | Note signal sorting technique based on multiple features fusion and feature selecting | |
| Paulus et al. | Music structure analysis by finding repeated parts | |
| CN109785850A (en) | A kind of noise detecting method, device and storage medium | |
| CN110108992B (en) | Cable partial discharge fault identification method and system based on improved random forest algorithm | |
| KR20140082157A (en) | Apparatus for speech recognition using multiple acoustic model and method thereof | |
| CN105760889A (en) | Efficient imbalanced data set classification method | |
| CN108985133B (en) | Age prediction method and device for face image | |
| CN110019779B (en) | Text classification method, model training method and device | |
| CN109461458B (en) | Audio anomaly detection method based on generation countermeasure network | |
| EP1932154B1 (en) | Method and apparatus for automatically generating a playlist by segmental feature comparison | |
| CN103761965B (en) | A kind of sorting technique of instrument signal | |
| JP2015176175A (en) | Information processing apparatus, information processing method and program | |
| CN104538036A (en) | Speaker recognition method based on semantic cell mixing model | |
| CN116524960A (en) | Speech emotion recognition system based on mixed entropy downsampling and integrated classifier | |
| Bruzzese et al. | DESPOTA: DEndrogram slicing through a pemutation test approach | |
| Virtanen et al. | Probabilistic model based similarity measures for audio query-by-example | |
| Yazdi et al. | Hierarchical tree clustering of fuzzy number | |
| JP6004014B2 (en) | Learning method, information conversion apparatus, and learning program | |
| CN104657473A (en) | Large-scale data mining method capable of guaranteeing quality monotony | |
| CN105956113B (en) | Video data digging High Dimensional Clustering Analysis method based on particle group optimizing | |
| KR101064833B1 (en) | Apparatus and method for mining of multi-dimensional association rules over interval data using clustering and characterization | |
| Tsau et al. | Content/context-adaptive feature selection for environmental sound recognition |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170905 |