Transcription of polyphonic piano recordings

A system for automatic identification of polyphonic piano recordings based on a heuristic approach is presented. The input to the transcription system is any piano piece sampled at 44.1 kHz. The system does not need any kind of training and there are no restrictions on the number of notes played simultaneously or on the notes frequency range. The transcription system initially divides the piano piece into attack slots. This temporal segmentation is based on onset detection and it is made by means of a sliding window procedure. Afterwards, a frequency analysis is performed on each attack slot. Finally, musical parameters such as beat, time signature, key signature and tonality are determined. It is shown that the proposed system achieves very good performance using simple signal processing techniques.