An identification system of musical tones

The aim of this paper is to present a system for the automatic identification of musical tones from a monophonic music melody in progress, among available alternatives of a library of the same previously recorded tones. This study was born by the demand to have a tool based on a comparison criterion to measure the fidelity of reproduction of some musical tones during a musical piece in execution. The algorithm is realized in two main distinct steps. At first, digital processing techniques are used with the purpose to obtain a pattern vectors from the original waveform. Thanks to the analysis techniques of Short-Time Fourier Transform, it has been possible to extract these patterns, so that they could to reflect precise energy-dependent features of the original signal, relevant to the identification. This resulting patterns are, subsequently, elaborated using the Theory of the Least Squares Optimal Filtering. The Least Squares Criterion is, here, regarded as purely deterministic, that is there is no presumed knowledge of the statistical properties. Therefore the algorithm has several desirable features. There is no upper limit on the frequency search range, so the algorithm is suited for high-pitched tones. The algorithm is relatively simple and may be implemented efficiently and with low latency on DSP processors. A preliminary investigation of the problem was developed in cooperation between the Norwegian University of Science and Technology of Trondheim and the National Research Council of Pisa, in Italy, in the framework of a stage at NTNU within the European Project 'Mosart' (2003).