Our new paper entitled “Jazz Trio Database: Automated Annotation of Jazz Piano Trio Recordings Processed Using Audio Source Separation” is just published in Transactions of the International Society of Music Information Retrieval (TISMIR). This paper arises from work completed by Huw Cheston during his PhD at the CMS, in collaboration with Joshua Schlichting (McMaster University).
Scaling up computational music research requires large datasets for training models and algorithms. Creating such databases has been difficult for improvised genres like jazz as traditional scores are not available, necessitating lengthy processes of transcribing audio recordings by hand. We solve this problem by introducing an automated pipeline: pre-trained deep-learning models are applied to “unmix” audio recordings into separate signals for multiple instruments, which are then automatically transcribed into symbolic representations (MIDI). We use our pipeline to generate the Jazz Trio Database (JTD), a dataset of 44.5 hours of jazz piano solos accompanied by bass and drums, with annotations for each performer. We anticipate that JTD will be useful in a variety of applications, including artist identification, performance modelling, and symbolic music generation.
See the paper for more details! Link below:
Cheston, H., Schlichting, J. L., Cross, I., & Harrison, P. M. C. (2024). Jazz Trio Database: Automated Annotation of Jazz Piano Trio Recordings Processed Using Audio Source Separation. Transactions of the International Society for Music Information Retrieval, 7(1), 144–158. https://doi.org/10.5334/tismir.186.
Image credit: "Jazz trio" by Jorge Franganillo is licensed under CC BY 2.0