HUST_Solfege is a public solfege dataset for solfege assessment, you can also use this dataset for solfege onset detection or note transcription.
HUST_Solfege contains 103 audios including 73 self-built solfege recordings and 30 singing recordings from MARG.
Among these recordings are 31 male recordings, 35 female recordings and 37 juvenile recordings. All of the audios are stored as monophonic, 44100Hz sampling rate and 16-bits wav files. Each audio includes a complete piece of music corresponding to a provided music score.
The statistics of HUST-Solfege is below.
| Solfege type | Number of recordings | Number of onset |
|---|---|---|
| Male | 31 | 1444 |
| Female | 35 | 1532 |
| Juvenile | 37 | 2092 |
| TOTAL | 103 | 5110 |
This dataset includes three files in repository: 1) wav, 2) onset&pitch and 3) alignment.
The description of wav files is stated above. All the audios are compressed in ZIP format.
The onset&pitch contains the notations for onset and actual pitch of each recording. The notations for offset haven't been done yet, please ignore the offset part, it will be completed in future work. The format of txt in onset&pitch is below.
| onset | offset | pitch |
|---|---|---|
| ... | ... | ... |
Note: The 30 recordings from MARG are just for onset detection. Pitch notations are not done for MARG recordings.
The alignment contains 30 txt file for alignment relationship between actual pitch and score note. The statistics of the test set is also available in statistics_of_testset.txt. Note: Test recordings are randomly selected from 73 solfege recordings.