Construction of music databases is one of the most important themes in the fields of music studies. Corpora of scores, audio recordings, and information on books of composers and musicians have been collected and used in the analysis of music styles, structures, and performance expressions. Recently, meta-text information, such as the names of composers and musicians has been attached to large-scale digital-databases and then been used in the analysis of music styles, structures, and performance expressions from the view point of social filtering. In spite of there being a number of active music database projects, there have been few projects dealing with music performance expression, such as dynamics, tempo and the progression of pitch.
Performances included in the CrestMusePEDB are transcribed into MIDI-level data through an iterated listening process by plural experts (who have master's degrees in musicology) using a commercial DAW and original alignment software. Then data on deviation is obtained.
The current version of the database is focused on traditional Baroque style music from the early twentieth century and includes music from Bach, Mozart, Beethoven, and Chopin. CrestMusePEDB ver. 3.1 is available. We are going to choose around a hundred music pieces, including those referred to often by previous music studies in the past couple of decades.