Robust features for representing structured signal components

In the last years we have developed several features for robustly representing repeating signal components [1]. A main focus in this was the robust detection of such signal components in real-world audio recordings. Applications included bioacoustical monitoring [2] (e.g., detection of repeating bird calls or sounds of marine mammals) and speech detection [3]. In the latter, the harmonic structure of voiced signal parts constitute the repeating components. One of my interests during the Dagstuhl seminar was to discuss possible applications of such features in music structure analysis. Generally speaking, repetitions can be seen as building blocks for more complex structure elements of audio signals, which is particularly obvious for music. Thus, another interesting thing discussed at the seminar was that of possible generalizations of the repetition-based features proposed in [2] to represent such complex structures. As a third possible application, the usability of such features to the extraction and separation of mixtures of repeating components (e.g, for multipitch extraction [3] or the detection of overlapping rhythmic components) was discussed