MAST rhythm dataset
Creators
- 1. Istanbul Technical University
- 2. Izmir Democracy University
Contributors
Data collectors:
- 1. Istanbul Technical University
- 2. Dokuz Eylül University
Description
The MASTrhythm dataset is a collection of 3721 audio files cropped from recordings of conservatory entrance examinations in Turkey (summer 2015 and summer 2016). (see the previous version with audio data here: https://zenodo.org/record/2620357#.Y1ZOVXVBwW0 ) This version of the dataset includes new annotation files.
4 level grades re-collected by blind-random-listening: A new set of grades were collected using an annotation tool that randomly plays audio recordings and ask the annotator to pick a grade from the set of grades: 1-Completely Off, 2-Major Mistakes, 3-Minor Mistakes, and 4-Perfect. The assessments were done by three individuals who specialized in music or have a musical background; the first annotator is a professor of musicology who has taken part as a jury member in many entrance exam auditions, the second and third annotator are music conservatory students of graduate-level programs. A rubric was explicitly defined and used for grading: If only one or two notes were wrong (missing, added, or played in a different rhythmic position), while the rhythmic accents were executed perfectly and without any delay or hurry, this type of error is considered as a "3-minor mistake"; If two-three notes were wrong, and were belonging to any form of hesitation with poorly executed accents, such performances were graded as a "2-major mistake". Any performance with more than three wrong strokes/durations was considered "1-completely off". The 4-level grades are shared in the format of text files: annotator#.txt
Naming convention:
The dataset is composed of audio files. All other information is coded in the file names: 'ref': reference recording, 'per': performance recording, 'fail': performance graded as 'fail', 'pass': performance graded as 'pass'
There exists 40 distinct rhythmic patterns. The ID for the melody makes up the first part of the file name. Examples:
'51_rhy1_per101559_fail.m4a': Rhythmic pattern ID: '51_rhy1', this is a candidate performance recording (candidate ID: 101559) graded as fail
'55_rhy2_ref280758.m4a': Rhythmic pattern ID: '55_rhy2', this is a reference recording (performed by a jury member) for candidate with ID: 280758.
Acknowledgement:
This new version of the dataset has been curated within the TUBITAK (The Scientific and Technological Research Council of Turkey) funded research project 1001-121E198 targeting development of automatic assessment tools for music performances.