Using duration models to reduce fragmentation in audio segmentation

Abstract

We investigate explicit segment duration models in addressing the problem of fragmentation in musical audio segmentation. The resulting probabilistic models are optimised using Markov Chain Monte Carlo methods; in particular, we introduce a modification to Wolff's algorithm to make it applicable to a segment classification model with an arbitrary duration prior. We apply this to a collection of pop songs, and show experimentally that the generated segmentations suffer much less from fragmentation than those produced by segmentation algorithms based on clustering, and are closer to an expert listener's annotations, as evaluated by two different performance measures.

Melodic Grouping in Music Information Retrieval: New Methods and Applications

“…Working at this level of abstraction, the task is to gather events (represented in metrical time as they might be in a musical score) into sequential groups. Research on segmentation from sub-symbolic or acoustic representations of music is not discussed as it generally operates either at the level of larger sections of music differing in instrumentation (e.g., Abdallah et al, 2006) or at the lower level of separating a continuous audio stream into individual note events (e.g., Gjerdingen, 1999;Todd, 1994). Furthermore, the present work emphasises melody (although not exclusively) reflecting the predominant trends in theoretical and computational treatments of perceived grouping structure in music.…”

Non-linear recurrent analysis of the behavior of a complex technological object

“…The papers [6,7] describe a state of natural or artificial systems that, as a rule, tend to change over time. The study of these complex processes is an important task in many disciplines, enabling to understand and describe their essence.…”

Content-Based Music Information Retrieval: Current Directions and Future Challenges

“…Fig. 8 shows an automatic segmentation using the method detailed in [63] of a pop track, along with an expert's annotation of segments from the same track. Note that the segmentation matches the annotation closely: apart from the identification of the solo ending and the transition, each annotation is algorithmically given a distinct label.…”