Statistical Learning and Estimation of Piano Fingering
This is the accompanying webpage for the following paper:
Eita Nakamura, Yasuyuki Saito, Kazuyoshi Yoshii
Statistical Learning and Estimation of Piano Fingering
Information Sciences, Vol. 517, pp. 68-85, 2020. [arXiv:1904.10237]

Source code

The source code for the piano fingering estimation methods based on 1st-order HMM, 2nd-order HMM, 3rd-order HMM, and chord HMM described in the paper together with programs for training model parameters and evaluating the results can be downloaded. Please see README.txt in the package.
Download the source code (127KB)

Estimation results by HMMs

1st-order HMM: Result_FHMM1.zip (115KB)

2nd-order HMM: Result_FHMM2.zip (115KB)

3rd-order HMM: Result_FHMM3.zip (114KB)

Some examples

The following examples show fingerings estimated by the 2nd-order HMM for one piece by Bach, one by Mozart, and one by Chopin, with different degrees of polyphony. Please click on an image to see a high-resolution version.

Bach: Invention No. 1
Mozart: Sonata K 330
Chopin: Etude Op. 10-3

Training and test data

The HMMs are trained and tested using different subsets of PIG Dataset. To get access to the dataset and see details of the data format, please visit the PIG Dataset webpage.

Visualizer

Picture This visualizer can be used for observing and comparing the above estimation results as well as the fingerings in PIG Dataset. It works on browsers such as Google Chrome, Safari, Firefox, Microsoft Edge, etc.