VocalNotes Dataset

The VocalNotes project was established to understand why transcriptions of music can differ from person to person, even amongst experts. The project involved experts from five traditions (Russia, Japan, China, Jewish, and Alpine), who each independently transcribed a selection of vocal songs from their musical tradition, in order to compare transcriptions and understand the origins of differences.

The VocalNotes Dataset was created primarily using two tools,  Tony (https://www.sonicvisualiser.org/tony/) and Sonic Visualizer (https://www.sonicvisualiser.org/). Fundamental frequency was estimated using the pYIN algorithm, and manually corrected in Tony. Note-level annotation was performed manually instead of using Tony's automated note-annotation function. Note pitch was manually adjusted in Sonic Visualizer, in cases where the transcriber disagreed with the note frequency automatically assigned by Tony. Volume level annotation was calculated using a short-time Fourier transform (window=2048, hop=256), and integrating the signal energy over each window.

If you use this dataset we ask you to cite us as follows:
P. Proutskova et al., "THE VocalNotes DATASET", in Extended Abstracts for the Late-Breaking Demo Session of the 24th Int. Society for Music Information Retrieval Conf., 2023

VocalNotesAnnotations.zip -- contains the complete set of annotations (f0, note onset, note duration, note frequency, volume level) from the VocalNotes project. This includes separate annotations from at least two expert transcribers, for approximately 10 minutes of audio, for music from five different traditions: Russia, Japan, China, Jewish, and Alpine. 
Access to the original audio is restricted. Access can be granted only for academic research. To gain access, you need to agree to the Data Use Agreement:
https://forms.gle/Gpk9MDoXbGPfz4E78