BPS-FH
#Dataset #Symbolic #Chord_Recognition
Beethoven Piano Sonata with Function Harmony (BPS-FH) dataset
MIDI形式ではない
元論文
Functional Harmony Recognition of Symbolic Music Data with Multi-task Recurrent Neural Networks
リンク
https://github.com/Tsung-Ping/functional-harmony
We represent the piano sonatas in the BPS-FH dataset as pianorolls with the pitch ranging from A0 to G#7 (middle C = C4), where the duration of each note is measured in term of crotchet beats, and is quantized to 32th note. The length of each sequence is 64, and each element in the sequence is a pianoroll segmented with window size of 33 and hop size of 2. As the input element for the Harmony Transformer, each pianoroll segment is flattened into a vector whose length is 84×33. We use 4-fold cross-validation for evaluation; the number of sequences of each fold varies from 368 to 585.
from Harmony Transformer: Incorporating Chord Segmentation Into Harmony Recognition