Dataset for Evaluating Pedalling Techniques Recognition Using Gesture Data

With the help of a dedicated measurement system (see reference for the details of the system), the pedalling gestures and the piano sound can be synchronously recorded at an audio sampling rate and a high resolution. The measurement system was deployed on the sustain pedal of a Yamaha baby grand piano situated in the studios at Queen Mary University of London. Ten well known passages of Chopin's piano music were selected to form this dataset. Therefore the dataset consists of: - <strong>audio-data.zip</strong>: piano sound of the ten passages recorded at 44.1kHz, each saved as "<strong>PASSAGE.wav</strong>". - <strong>pedal-data.zip</strong>: associated gesture data recorded at 22.05kHz, each saved as "<strong>PASSAGE.npy</strong>". The gesture data correspond to the movement trajectory of the sustain pedal. - <strong>pedal-label.zip</strong>: label the "continuous" gesture data by "discrete" pedalling techniques at every 0.02 second. Label 0-4 represents none, 1/4, 1/2, 3/4 and full pedalling technique, respectively. Labels for gesture data "<strong>PASSAGE.npy</strong>" are saved in "<strong>PASSAGE-label.npy</strong>". - <strong>passage.zip</strong>: music scores of the ten passages, each saved as "<strong>PASSAGE.pdf</strong>". They were annotated with pedalling techniques by the experimenter in advance and then performed by a pianist, who was asked to follow the annotated scores. The resulting audio recording and gesture data formed the above "<strong>PASSAGE.wav</strong>" and the "<strong>PASSAGE.npy</strong>". The annotated score guided the labelling process and formed the "<strong>PASSAGE-label.npy</strong>".