Leveraging noisy online databases for use in chord recognition

• DOI: 10.5281/zenodo.1418311
• Publication Date: 2011-10-24
• Authors: Matt McVicar; Yizhao Ni; Tijl De Bie; Raúl Santos‐Rodrí...
• Keywords: Chord (peer-to-peer); Training set; Performance Improvement; Boosting

The most significant problem faced by Machine Learningbased chord recognition systems is arguably the lack of highquality training examples. In this paper, we address leveraging availability annotations from guitarist websites. We show that such can be used as partial supervision a semi-supervised method—partial since accurate timing information lacking. A particular challenge in exploitation these data their low quality, potentially even leading to performance degradation if directly. demonstrate however curriculum learning strategy automatically rank according potential for improving performance. Using strategy, our experiments modest improvement simple major/minor alphabet, but highly much larger alphabet.