Automatic Classification of Pitched Musical Instrument Sounds

In this chapter we first discuss the methodological elements required for developing systems for the automatic classification of sounds. We then introduce a number of acoustic features that are useful for describing the sounds of pitched musical instruments, some techniques to further distil the information they provide (i.e., transformation and projection), and some strategies for selecting the best features when large sets of them are available. Several techniques for automatic classification, complementing those explained in other chapters of the book, are presented in another specific section. Once the reader has been introduced to all the tools this problem calls for, we present a review of the different systems and approaches that have been proposed for the automatic classification of instruments in isolated sounds, solo phrases, duets, and polyphonic audio. Conclusions and topics for future research compose the closing section of the chapter.

1 Introduction
1.1 Classification in humans and machines
1.2 Taxonomies of musical instruments
1.3 Timbre spaces
1.4 Outline of this chapter
2 Methodology
2.1 Databases
2.2 Validation and generalization
3 Features and their selection
3.1 Energy features
3.2 Spectral features
3.3 Harmonic features
3.4 Temporal features
3.5 Feature transformation
3.6 Feature selection
4 Classification techniques
4.1 Instance-based learning
4.2 Discriminant analysis
4.3 Decision trees
4.4 Artificial Neural Networks
5 Classification of isolated sounds
5.1 Flat classification systems
5.2 Classification based on hierarchical decisions
6 Classification of sounds from music files
6.1 Classification of instruments in solo phrases
6.2 Classification of instruments in duets
6.3 Classification of instruments in complex mixtures