Identification of polyphonic piano signals

This paper presents a spectral identification method using a data base as a reference and applied to piano sounds. This method has a low cost in terms of computational time and allows the treatment of five octaves for polyphonic signals. The characteristic inharmonicity of piano notes and a sufficient number of partials present in a note spectrum are taken into account to treat the lower octaves and to avoid octave errors. An original method of linear search, based on the comparison between a data base of note spectra, used as a reference, and the incoming spectrum, allows a simple and fast identification.