Dataset Open Access
In the work described in the paper A dataset of rhythmic pattern reproductions and baseline automatic assessment system (ISMIR, 2019), Falcão et al. applied efforts on annotating a subset of the full MAST rhythm dataset with grades for a subset of performances.
Out of a total of 1040 student submissions comprised by the original dataset, 80 performances (equally distributed between 20 references) were graded by annotators using a custom annotation tool. During this process annotators could play each (reference, performance) pair as many times as desired before deciding on one of the available assessments: 4-Perfect, 3-Minor errors, 2-Major errors, 1-Completely off.
This dataset is a collection of audio files, onset features and user-annotated data. Audio files are stored into the Only Performances and Only References folders, together with some extra files containing their original onsets, quantized onsets and scaling parameters (for more information regarding this auxiliary data, please check the paper). An indexed list of files (listreferences and listperformances) is also stored in each of these folders in order to allow for mapping between audio files and auxiliary data. Example:
Also, in order to check the reference which relates to a specific performance, one must also look for audio file names in the list of files: the i-eth audio file listed in Only Performances/listperformances is a specific performance whose reference is defined by the i-eth line of Only References/listreferences. Example:
Such convention is also applied to the folder that contains the annotations (Performances Annotations). Each one of the annotator_X.txt file lists the grades assigned by the X-ith annotator, and the mapping between the assessments and their referred performances is indexed in listfiles.txt