Music information indexing based on timbre helps users to get relevant musical data in large digital music databases. Timbre is a quality of sound that distinguishes one music instrument from another among a wide variety of instrument families and individual categories. The real use of timbre-based grouping of music is very nicely discussed in (Bregman, 1990). Typically, an uncompressed digital music recording, in form of a binary file, contains a header and a body. A header stores file information such as length, number of channels, rate of sample frequency, etc. Unless being manually labeled, a digital audio recording has no description on timbre, pitch or other perceptual properties. Also, it is a highly nontrivial task to label those perceptual properties for every piece of music object based on its data content. Lots of researchers have explored numerous computational methods to identify the timbre property of a sound. However, the body of a digital audio recording contains an enormous amount of integers in a time-order sequence. For example, at a sample frequency rate of 44,100Hz, a digital recording has 44,100 integers per second, which means, in a oneminute long digital recording, the total number of the integers in the time-order sequence will be 2,646,000, which makes it a very big data item. Being not in form of a record, this type of data is not suitable for most traditional data mining algorithms. Recently, numerous features have been explored to represent the properties of a digital musical object based on acoustical expertise. However, timbre description is basically subjective and vague, and only some subjective features have well defined objective counterparts, like brightness, calculated as gravity center of the spectrum. Explicit formulation of rules of objective specification of timbre in terms of digital descriptors will formally express subjective and informal sound characteristics. It is especially important in the light of human perception of sound timbre. Time-variant information is necessary for correct classification of musical instrument sounds because quasi-steady state, where the sound vibration is stable, is not sufficient for human experts. Therefore, evolution of sound features in time should be reflected in sound description as well. The discovered temporal patterns may better express sound features than static features, especially that classic features can be very similar for sounds representing the same family or pitch, whereas changeability of features with pitch for the same instrument makes sounds of one instrument dissimilar. Therefore, classical sound features can make correct identification of musical instruments independently on the pitch very difficult and erroneous.
Automatic content extraction is clearly needed and it relates to the ability of identifying the segments of audio in which particular predominant instruments were playing. Instruments having rich timbre are known to produce overtones, which result in a sound with a group of frequencies in clear mathematical relationships (so-called harmonics). Most western instruments produce harmonic sounds. Generally, identification of musical information can be performed for audio samples taken from real recordings, representing waveform, and for MIDI (Musical Instrument Digital Interface) data. MIDI files give access to highly structured data. So, research on MIDI data may basically concentrate on higher level of musical structure, like key or metrical information. Identifying the predominant instruments, which are playing in the multimedia segments, is even more difficult. Defined by ANSI as the attribute of auditory sensation, timbre is rather subjective: a quality of sound, by which a listener can judge that two sounds, similarly presented and having the same loudness and pitch, are different. Such definition is subjective and not of much use for automatic sound timbre classification. Therefore, musical sounds must be very carefully parameterized to allow automatic timbre recognition. There are a number of different approaches to sound timbre (Balzano, 1986; Cadoz, 1985). Dimensional approach to timbre description was proposed by (Bregman, 1990). Sets of acoustical features have been successfully developed for timbre estimation in monophonic sounds where mono instruments were playing. However, none of those features can be successfully applied to polyphonic sounds, where two or more instruments were playing at the same time, since those features represent the overlapping sound harmonics as a whole instead of individual sound sources.