Combination of Onset-Features with Applications to High-Resolution Music Synchronization

摘要

In this contribution, we described an approach for extracting novelty curves from music signals yielding good indicators for note onsets even in the case of only smooth temporal and spectral intensity changes within the signal. Furthermore, we showed how these features can be used for enhancing conventional music synchronization strategies, resulting in significant improvements with regard to the temporal accuracy of the alignments especially for music comprising instruments with a soft attack phase. For the future, we plan to employ the improved synchronization results for tasks such as performance analysis, where one objective is to extract expressive tempo information from music recordings.