用于分数对齐的传统长笛数据集免费

jsaifc 48 2021-08-24 语音识别

资源介绍

用于分数对齐的传统长笛数据集 (http://ds.jsai.org.cn/) 语音识别 第1张Context The flute is a widely used instrument in compositional practices. Furthermore, it has a long tradition in [Electroacoustic Music][1] where instruments are combined with electrical sound production, generated for example with computers. On the other hand, audio to score alignment is the task of synchronizing an audio recording with the corresponding symbolic score, from where fields as [Computer Music][2] leverage in variety of applications. With the intention to contribute in the task of audio to score alignment with solo flute musical pieces this dataset was created out. Content This set of data is composed of 30 manually-labeled audio fragments and the corresponding scores with symbolic notation. Construction was made out from real solo flute pieces, through various recordings of 4 important musical pieces of flute repertoire. Selected musical pieces are based on the traditional execution techniques of the flute, where music can be described as notes (pitch and duration), leaving out the contemporary repertoire based on extended techniques. The musical pieces are: - **Allemande (first movement of BWV 1013):** composed by J.S. Bach - **Syrinx:** composed by C. Debussy - **Density 21.5:** composed by E. Varese - **Sequenza I:** composed by L. Berio From the total of fragments, 10 correspond to Allemande, 10 to Syrinx, 6 to Density 21.5 and 4 to Sequenza I. In addition all the flute register is covered (from C4 to D7) resulting in a total of 2245 musical events. Three kind of files compose the dataset: - **Audio files:** coded as 16bits @44100Hz with file extension '.wav'. - **Score files:** in two formats. On one hand, csv files with extension '.notes' generated through [Lilypond][3], and on the other hand midi files with '.midi' extension. - **Ground truth:** files generated by myself through manual annotations using [Sonic Visualizer][4], exported as csv files and saved with '.gt' extension. - **load.py:** python module with some functions for the dataset usage. Last but not least, in the synth folder you can find a synthetic version of the scores and the ground truth made with [Wind Synthesis Toolbox][5]. Referencing If you use this dataset in a publication, I would be grateful if you reference this page as follows: Juan P. Braga Brum (2018). "Traditional Flute Dataset for Score Alignment

END

发表评论