SingWithExpressions is the accompanying repository for the submission of the ISMIR paper - Automatic Estimation of Singing Voice Musical Dynamics.
The dataset consists of 509 performances synchronized with 163 score files, curated semi-automatically using state-of-the-art source separation, and audio-to-score alignment techniques. The resulting dataset can serve as a valuable resource for various MIR research endeavors such as transcription, synthesis, or vocal pedagogy.
The performances are sources from YouTube and the scores are sources from OperScore Lieder Corpus. For isolating vocals, we Demucs, followed by automatic alignment with the scores from Lieder corpus. More details can be found in the accompanying paper.
SingWithExpressions is licensed under CC BY-NC-SA 4.0
