Beijing opera percussion pattern dataset

CompMusic

doi:https://doi.org/10.34810/data441

Beijing opera percussion pattern dataset

License

Citation

CompMusic. Beijing opera percussion pattern dataset [dataset]. Repositori Digital de la UPF: Barcelona, 2016 [citat 17 novembre 2016]. Disponible a: http://hdl.handle.net/10230/25720

Abstract

Beijing Opera Percussion Pattern (BOPP) dataset is a collection of 133 audio percussion patterns covering five pattern classes. The dataset includes the audio and syllable level transcriptions for the patterns (non-time aligned). It is useful for percussion transcription and classification tasks. The patterns have been extracted from audio recordings of arias and labeled by a musicologist.

Description

-Dataset-/n/nThe dataset is a collection of 133 audio percussion patterns spanning five different pattern classes as described below. The scores for the patterns and additional details about the patterns are at: http://compmusic.upf.edu/bo-perc-patterns/n/n-Audio Content-/n/nThe audio files are short segments containing one of the above mentioned patterns. The audio is stereo, sampled at 44.1 kHz, and stored as wav files. The segments were chosen from the introductory parts of arias. The recordings of arias are from commercially available releases spanning various artists. The audio and segments were chosen carefully by a musicologist to be representative of the percussion patterns that occur in Jingju. The audio segments contain diverse instrument timbres of percussion instruments (though the same set of instruments are played, there can be slight variations in the individual instruments across different ensembles), recording quality and period of the recording. Though these recordings were chosen from introductions of arias where only percussion ensemble is playing, there are some examples in the dataset where the melodic accompaniment starts before the percussion pattern ends. /n/n-Annotations-/n/nEach of the audio patterns has an associated syllable level transcription of the audio pattern. The transcription is obtained from the score for the pattern and is not time aligned to the audio. The transcription is done using the reduced set of five syllables described in Table 1 of [1] and is sufficient to computationally model the timbres of all the syllables. The annotations are stored as Hidden Markov Model Toolkit (HTK) label files. There is also a single master label file provided for batch processing using HTK (http://htk.eng.cam.ac.uk/). /n/n-Dataset organization-/n/nThe dataset has wav files and label files. The files are named as /n<pID><InstID>.<extension>/nThe pID is as in Table 1, instID is a three digit identifier for the specific instance of the pattern, and extension can be .wav for the audio file or .lab for the label file. pID ϵ {10, 11, 12, 13, 14}, InstID ϵ {1, 2, ..., NpID}. e.g. The audio file and the label file for the fifth instance of the pattern duotuo is named 12005.wav and 12005.lab, respectively. The master label file is called masterLabels.lab/n/n-Availability of the Dataset-/n/nThe annotations are publicly shared and available to all. The audio is from commercially available releases. It cannot be publicly shared but can be made available on request for non-commercial research purposes. In the future, the dataset would be available for viewing and download through an interface in Dunya (http://dunya.compmusic.upf.edu).