The article presents an in-depth study of foundation models tailored for music informatics, a domain challenged by the scarcity of labeled data and generalization issues. The researchers conducted a comparative study among various foundation model variants, examining key factors such as model architectures, tokenization methods, temporal resolution, data, and model scalability. The performance of these models was assessed across diverse tasks in music information retrieval, particularly focusing on token-level and sequence-level classification. The results demonstrated robust performance of the model, surpassing existing models in specific key metrics. These findings contribute to the understanding of self-supervised learning in music informatics and pave the way for more effective foundation models in the field.
Publication date: 8 Nov 2023
Project Page: Not provided
Paper: https://arxiv.org/pdf/2311.03318
