This dataset systematically collects the audio samples of single notes and octaves played in real-world environments. four musicians (two pianists and two guitarists) with different professional levels are paid and asked to play single notes and octave dual-tones in all possible pitch ranges, with different musical dynamic levels (forte, mezzo forte and piano) and playing techniques (normal, pedal and staccato for piano and normal for guitar). All piano sound samples are home-recorded, meaning more noises, reverberation and other possible defects prevalent in real-world recordings. The sound samples of ‘Guitar 1’ are recorded in a studio, while those of ‘Guitar 2’ are home-recorded.
For piano, the pitches of single-note data range from A0 to C8, while octave data range from A0+A1 to C7+C8. The pitch range of guitar is narrower, ranging from E2 to D6 for single notes, with some duplicated pitches played on different strings, and E2+E3 to D5+D6 for octave data.
If you make use of this dataset for academic purposes, please cite the following publication:
Li Su, Li-Fan Yu, Hsin-Yu Lai, Yi-Hsuan Yang, Resolving octave ambiguities: A cross-dataset Investigation, Proc. Sound and Music Computing Conference (SMC), 2014.
Please download the full datasethere(1.46 GB).
For more information, please contact Li Su (firstname.lastname@example.org).