Presentation on use of SVM for Instrument ID

Uwe Ligges and Sebastian Krey present their research on Instrument ID, "SVM based Classification of Instruments-Timbre Analysis" in these slides.

I'm hoping to soon get a copy of their publication "Krey, S. and Ligges, U. (2010): SVM based Instrument and Timbre Classification. In Locarek-Junge, H. and Weihs, C. (eds.). Classification as a Tool for Research, Springer-Verlag, Berlin. (in print)"

A very good (relatively up-to-date - 2007) overview of Classification in Music Research

Link:

http://http//www.statistik.tu-dortmund.de/fileadmin/user_upload/Lehrstuehle/Datenanalyse/weihs07classification.pdf

The abstract:

Since some few years, classification in music research is a very broad and quickly growing field. Most important for adequate classi cation is the knowledge of adequate observable or deduced features on the basis of which meaningful groups or classes can be distinguished. Unsupervised classi cation additionally needs an adequate similarity or distance measure grouping is to be based upon. Evaluation of supervised learning is typically based on the error rates of the classi cation rules. In this paper we fi rst discuss typical problems and possible influential features derived from signal analysis, mental mechanisms or concepts, and compositional structure. Then, we present typical solutions of such tasks related to music research, namely for organization of music collections, transcription of music signals, cognitive psychology of music, and compositional structure analysis.

More AMII researchers

Sebastian Krey

Dr. Uwe Ligges

Two papers available online are of interest:

• Weihs, C., Reuter, C. and Ligges, U. (2004): "Register Classification by Timbre", Technical Report 71/2004. SFB 475, Department of Statistics, University of Dortmund, Germany.
• Szepannek, G., Ligges, U., Luebke, K., Raabe, N. and Weihs, C. (2005): "Local Models in Register Classification by Timbre", Technical Report 47/2005. SFB 475, Department of Statistics, University of Dortmund, Germany.

Papers and posters of interest in this coming ISMIR

ISMIR is around the corner - the program is up.

In relation to AMII - these look interesting:

• "Musical Instrument Recognition using Biologically Inspired Filtering of Temporal Dictionary Atoms", Steven K. Tjoa and K.J. Ray Liu (My thanks to Steven for emailing me a copy of this today)
• "YAAFE, an Easy to Use and Efficient Audio Feature Extraction Software", Benoit Mathieu, Slim Essid, Thomas Fillon, Jacques Prado and Gaël Richard

The YAAFE toolbox is available here with an extension module here. Of the extension features, the SpectralIrregularity seems the most interesting IMO.

The available features in YAAFE basic are:

• AmplitudeModulation
• AutoCorrelation
• ComplexDomainOnsetDetection
• Energy
• Envelope
• EnvelopeShapeStatistics
• Frames
• LPC
• LSF
• Loudness
• MFCC
• MagnitudeSpectrum
• OBSI
• OBSIR
• PerceptualSharpness
• PerceptualSpread
• SpectralCrestFactorPerBand
• SpectralDecrease
• SpectralFlatness
• SpectralFlatnessPerBand
• SpectralFlux
• SpectralRolloff
• SpectralShapeStatistics
• SpectralSlope
• SpectralVariation
• TemporalShapeStatistics
• ZCR

Available feature transforms:

• AutoCorrelationPeaksIntegrator
• Cepstrum
• Derivate
• HistogramIntegrator
• SlopeIntegrator
• StatisticalIntegrator

AND TO THINK OF THE AMOUNT OF TIME I SPENT WRITING CODE TO EXTRACT THESE FEATURES!!!!

Another AMII researcher - Alexey Ozerov

I stumbled upon Alexey Ozerov - musical instrument recognition being just one aspect of his research.

Two papers are of significant interest to my own research:

• Ozerov, A., Essid, S. Tech Rep 2009 "Instrument recognition in polyphonic music based on NMF decomposition and SVM classification"
• J.-L. Durrieu, A. Ozerov, C. Févotte, G. Richard and B. David, "Main instrument separation from stereophonic audio signals using a source/filter model", In EUSIPCO, 17th European Signal Processing Conference, Glasgow, Scotland, August 24-28, 2009.

SVMs demystified...

I came across an excellent blog 'Onionesque Reality' by Shubhendu Trivedi. One article entitled "Demystifying Support Vector Machines for Beginners" does exactly what it says on the tin. Trivedi also blogs how SVMs can be used in Face Recognition and even attempts to answer why SVMs are so called.

Two books which are recommended caught my interest:

"Support Vector Machines and other Kernel Based Learning methods" by Nello Cristianini and John-Shawe Taylor.

"Learning with Kernels" by Bernhard Scholkopf and Alexander Smola. (Perfect book for beginners)

I've a lot of reading to do!

And before I forget, this lecture on SVM is meant to be awesome.

Another toolbox!

SHOGUN - A large scale machine learning toolbox. Will I or wont I...?

Validating predictive models

Will Dwinnell blogs about Validating Predictive Models with links:

• What are the population, sample, training set, design set, validation set, and test set?
• A generic k-fold cross validation matlab tool

Something different - Music movies

A superb list of music movies is provided by the UC library (Berkeley). Would be well worth checking some of these out.

Ensemble classifiers win PAKDD 2010

Taken from Data Mining and Predictive Analysis:

PAKDD-10 Data Mining Competition Winner: Ensembles Again!

The PAKDD-10 Data Mining Competition results are in, and ensembles occupied the top 4 positions, and I think the top 5. The winner used Stochastic Gradient Boosting and Random Forests in Statistica, second place a combination of logistic regression and Stochastic Gradient Boosting (and Salford Systems CART for some feature extraction). Interestingly to me, the 5th place finisher used WEKA, an open source software tool.

Evaluating classifier performance

An article in the excellent blog by Will Dwinnel, discusses evaluating classifier performance using Confusion Matrices, ROC, Lift charts and AUROC. Links to external resources are provided:

• AUC: a Statistically Consistent and more Discriminating Measure than Accuracy, by Charles X. Ling, Jin Huang and Harry Zhang
• Evaluating Performance, from “ROC Graphs: Notes and Practical Considerations for Researchers, by T. Fawcett
• The Use of the Area Under the ROC Curve in the Evaluation of Machine Learning Algorithms, by Andrew P. Bradley

In the blog entry "Model Performance Measurement", Matlab code is provided for various performance measurement routines:

• 'L-1' (mean absolute error)
• 'L-2' (mean squared error)
• 'L-4'
• 'L-16'
• 'L-Infinity'
• 'RMS' (root mean squared error)
• 'AUC' (requires tiedrank() from Statistics Toolbox)
• 'Bias'
• 'Conditional Entropy'
• 'Cross-Entropy'
• 'F-Measure'
• 'Informational Loss'
• 'MAPE'
• 'Median Squared Error'
• 'Worst 10%'
• 'Worst 20%'