The Impact of Label Noise on a Music Tagger
This addresses the problem of label noise for researchers and practitioners in audio music tagging, but it is incremental as it builds on existing methods without introducing new paradigms.
The study investigated how label noise affects learning in audio music tagging, finding that carefully annotated labels yield the best performance, but noisy labels still contain sufficient information for successful learning, with artificial corruption quantifying their contribution.
We explore how much can be learned from noisy labels in audio music tagging. Our experiments show that carefully annotated labels result in highest figures of merit, but even high amounts of noisy labels contain enough information for successful learning. Artificial corruption of curated data allows us to quantize this contribution of noisy labels.