Intelligent Music Classification

1
Intelligent Music Classification

• This presentation will probably involve audience
  discussion, which will create action items. Use
  PowerPoint to keep track of these action items
  during your presentation
• In Slide Show, click on the right mouse button
• Select Meeting Minder
• Select the Action Items tab
• Type in action items as they come up
• Click OK to dismiss this box
• This will automatically create an Action Item
  slide at the end of your presentation with your
  points entered.

• with
• Support Vector Machines
• Chris Felton
• C Wang

2
Summary

• Comments
• Overview (Again)
• Implementation
• Results
• Difficulties
• Future Work

3
Terminology

• Samples
• Patterns, Cases, Inputs, Instances, Observations,
  Exemplar
• Labels
• Domain, Targets, Outputs, Observations

4
2D Binary Classification Plots
5
Random Signal Classification
6
Music Classification

• Extract Features
• Classify using Support Vector Machines

7
Feature Extraction

• Mostly Spectral Information
• Analyze
• FFT
• MFCC
• Spectrogram
• LPC
• Pwelch, Power Spectrum Density Estimate
• Room for Improvement

8
(No Transcript)
9
Fourier Analysis (FFT)
Train 2 Rock, 2 Classical Test 2 Rock, 2
Classical
10
Mel Frequency Cepstrum Coefficients (MFCC)
Train 2 Rock, 2 Classical Test 2 Rock, 2
Classical
11
More Fourier, Spectrogram
Train 2 Rock, 2 Classical Test 2 Rock, 2
Classical
12
Current Input Vector

• Combine the Previous 3
• Good Test Confusion Matrix

13
Linear Prediction Coding (LPC)
Train 2 Rock, 2 Classical Test 2 Rock, 2
Classical
14
Pwelch Spectral Density
Train 2 Rock, 2 Classical Test 2 Rock, 2
Classical
15
Support Vector Machines

• Definition
• Multi Class Architecture
• RBF Kernel

16
Definition
17
One vs. Rest
18
RBF Kernel
19
Implementation

• Data Sets
• Vector Formats
• Matlab

20
Data

• 20 Samples from each Genre
• 10 Types/Genres
• Alternative, Blues, Classical, Country,
  Electronica, Hip-Hop/Rap, Jazz, RB/Soul,
  Rock/Pop, World/Latin
• 10 Know Test Samples
• 10k Songs to Classify (In Progress)

21
Vector Formats
22
Results

• 4 Classes
• 10 Classes
• Train Best Sigma and C Algorithm

23
Matlab Software

• Determine Optimal Sigma and C.
• Train SVM with the Discovered Sigma and C
• Test on Know Test Samples
• Run on more Samples, Survey Results

24
4 SVM Classifierson Testing Data
25
10 SVM Classifiers on Testing Data
Testing Jazz Class with Jazz SVM Classifier
1 0 1 0 Testing RB-Soul Class with
RB-Soul SVM Classifier 1 0 1
0 Testing Rap-HipHop Class with Rap-HipHop SVM
Classifier 1 0 1 0 Testing
Rock-Pop Class with Rock-Pop SVM Classifier
1 0 1 0 Testing World-Folk Class
with World-Folk SVM Classifier 1 0
1 0

• Testing Alternative Class with Alternative SVM
  Classifier
• 0.2000 0.8000
• 0.0556 0.9444
• Testing Blues Class with Blues SVM Classifier
• 1 0
• 1 0
• Testing Classical Class with Classical SVM
  Classifier
• 0.3000 0.7000
• 0.9028 0.0972
• Testing Country Class with Country SVM Classifier
• 0.8000 0.2000
• 0.9583 0.0417
• Testing Electronica Class with Electronica SVM
  Classifier
• 0.7000 0.3000
• 0.9583 0.0417

26
(No Transcript)
27
Difficulties

• Getting the sigma and C Algorithm to Work!
• Collecting Samples
• Time to Train, Test, etc.

28
Future Work

• More Feature Extraction
• Cascaded SVM Multi Class Architecture

29
Software

• Matlab
• OSU SVM Matlab Toolbox
• Auditory Matlab Toolbox
• JavaLayer MP3 converter (Interfaced into Matlab)
• JVorbis Ogg Vorbis Audio Compression Decoder

30
References

• Learning with Kernels, Scholkoppf and Smola, 2002
• Neural Networks a Comprehensive Foundation, Simon
  Haykin
• An Introduction to Support Vector Machines and
  other kernel-based learning methods, Nello
  Cristianini and John Shawe-Taylor
• Content-Based Audio Segmentation using Support
  Vector Machines
• Neural and Adaptive Systems, Jose C. Principe,
  Neil R. Euliano, and W. Curt Lefebvre
• Using the Fisher Kernel Method for Web Audio
  Classification
• Pattern Recognition with Support Vector Machines,
  SVM 2002 Proceedings
• Handbook of Neural Networks for Speech
  Processing, Shigeru Katagiri
• OSU SVM, http//eewww.eng.ohio-state.edu/maj/osu_
  svm/
• LIBSVM, http//www.csie.ntu.edu.tw/cjlin/libsvm/
• Auditory Tool Box, http//rvl4.ecn.purdue.edu/mal
  colm/interval/1998-010/