MIR

Methods

e.g. amplitude distribution

Fill missing values

Smooth noisy data

Identify and remove outliers

Resolve inconsistencies

Provide a unified view of the data

Storing and arranging the database

Data reduction- represent a large set with a smaller subset

e.g. sample or bit depth reduction

Discretize a continuous value

In music data mining, discretization refers to breaking the music pieces down into relatively simpler and smaller parts, and the way these parts fit together and interact with each other is then examined.

e.g. PCA

e,g. Fourier transform, Wavelet transform

From a single piece of content

or from correlations, groupings or clusterings

Usually content based

Music similarity search

e.g. Music recommendation can be achieved by mining association patterns based on song co-occurrence

detecting associations among different acoustic features.

detecting associations among music and other document formats

e.g. Chord recognition in a MIDI file, model typical properties of drum sequences, exploring music structure

i.e. like Association, but now order and location are important

David Cope could fit here

Assign labels or categories to data

Classifiers

Audio classification

Genre classification

Artist Classification

Singer classification

Mood detection

Instrument recognition

partitioning a finite set of points in a multidimensional space into classes

Creation of a concise and informative extraction that best summarizes an

original digital content

generally the Euclidean distance of feature vectors

Multitask Learning (MTL)

Multiple Instance Learning (MIL)

Multilabel

Semisupervised learning

Tenso-based learning

Coded excitation LP

Source signal decomposed on a code book

Representation of the shape of the spectrum

Essentially an FFT of the magnitude spectrum

Cepstrum using Mel scale