Mentions légales du service

Skip to content
Snippets Groups Projects

PSEval

Evaluation results of the engraving-based algorithms PSE and PS13b, for joint Pitch-Spelling and global and local Key Estimation.

The code of the algorithms can be found at https://gitlab.inria.fr/pse/pse

Evaluation Datasets

These algorithms have been evaluation on two different datasets:

  • the ASAP piano dataset the last results on the ASAP copora are in directory Results_ASAP/
  • the Lamarque-Goudard (LG) dataset with monophonic (complex) extracts the last results on this corpora are in directory Results_LG/
  • the directory Archives/ contains older evaluations and is obsolete.

Evaluation directories

Every evaluation directory contains:

  • the score files (MusicXML) annotated with
    • color codes for notes with spelling errors (see below for a precise description of colors)
    • the estimated local key for each measure
    • the estimated global key at the beginning of the score
  • one evaluation table in cvs format .csv file with 1 row for each score file evaluated

the firectory names have the following form: PSE_tons_C1_M1D1_filter_GRID_C2_M2D2 where

  • tons number of tonalities for evalution (typically 30, 135 or 165)
  • Ci = is the cost domain used at step i
  • Mi = if the computation of the initial state for spelling (modal/tonal) at step i
  • Di = wheter the computation of the spell table is deterministic (with chromatic scale) or exhaustive, for the step i
  • filter : % latency wrt best global for the pre-selection of global tons after step 1 100 means no selection (use all global for grid and step 2), and is not kept in the dir name
  • GRID: name of algo for the computation of the grid

Color code anotations

The notes in the score files (MusicXML) are coloured according to their comparison to the spelling in the ground truth score in the dataset considered. Both Pitch-Spelling algorithms PSE and PS13b processed in 2 steps:

  1. spelling
  2. rewriting of passing notes

The comparison of the spelling result with ground truth after each step determines the color code as follows:

step 1 step 2 color
comparison with ground truth = = black
comparison with ground truth = \neq violet
comparison with ground truth \neq = green
comparison with ground truth \neq \neq red

For instance, a green note indicates a spelling that was incorrect after step 1 but has been fixed by step 2 (rewriting), and a red note indicates a spelling incorrect after step 1 that was not fixed by step 2.