Both in interactive music listening, and in music performance research, there is a need for automatic alignment of different recordings of the same musical piece. This task is challenging, because musical pieces often contain parts that may or may not be repeated by the performer, possibly leading to structural differences between performances (or between performance and score). The most common alignment method, dynamic time warping (DTW), cannot handle structural differences adequately, and existing approaches to deal with structural differences explicitly rely on the annotation of ``break points'' in one of the sequences. We propose a simple extension of the Needleman-Wunsch algorithm to deal effectively with structural differences, without relying on annotations. We evaluate several audio features for alignment, and show how an optimal value can be found for the cost-parameter of the alignment algorithm. A single cost value is demonstrated to be valid across different types of music. We demonstrate that our approach yields roughly equal alignment accuracies compared to DTW in the absence of structural differences, and superior accuracies when structural differences occur.
We present a new neural network based peak-picking algorithm for common onset detection functions. Compared to existing hand-crafted methods it yields a better performance and leads to a much lower number of false negative detections. The performance is evaluated on basis of a huge dataset with over 25k annotated onsets and shows a significant improvement over existing methods in cases of signals with previously unknown levels.
JF - Proceedings of the 6th International Workshop on Machine Learning and Music CY - Prague, Czech Republic ER - TY - CONF T1 - Local Group Delay based Vibrato and Tremolo Suppression for Onset Detection T2 - Proceedings of the 14th International Society for Music Information Retrieval Conference (ISMIR 2013) Y1 - 2013 A1 - Sebastian Böck A1 - Widmer, Gerhard KW - local group delay KW - onset detection KW - tremolo suppression KW - vibrato suppression AB -We present SuperFlux - a new onset detection algorithm with vibrato suppression. It is an enhanced version of the universal spectral flux onset detection algorithm, and reduces the number of false positive detections considerably by tracking spectral trajectories with a maximum filter. Especially for music with heavy use of vibrato (e.g., sung operas or string performances), the number of false positive detections can be reduced by up to 60% without missing any additional events. Algorithm performance was evaluated and compared to state-of-the-art methods on the basis of three different datasets comprising mixed audio material (25,927 onsets), violin recordings (7,677 onsets) and operatic solo voice recordings (1,448 onsets). Due to its causal nature, the algorithm is applicable in both offline and online real-time scenarios.
JF - Proceedings of the 14th International Society for Music Information Retrieval Conference (ISMIR 2013) CY - Curitiba, Brazil ER - TY - CONF T1 - Maximum Filter Vibrato Suppression for Onset Detection T2 - Proceedings of the 16th International Conference on Digital Audio Effects (DAFx-13) Y1 - 2013 A1 - Sebastian Böck A1 - Widmer, Gerhard KW - maximum filter KW - onset detection KW - vibrato suppression AB -
We present SuperFlux - a new onset detection algorithm with vibrato suppression. It is an enhanced version of the universal spectral flux onset detection algorithm, and reduces the number of false positive detections considerably by tracking spectral trajectories with a maximum filter. Especially for music with heavy use of vibrato (e.g., sung operas or string performances), the number of false positive detections can be reduced by up to 60% without missing any additional events. Algorithm performance was evaluated and compared to state-of-the-art methods on the basis of three different datasets comprising mixed audio material (25,927 onsets), violin recordings (7,677 onsets) and operatic solo voice recordings (1,448 onsets). Due to its causal nature, the algorithm is applicable in both offline and online real-time scenarios.
JF - Proceedings of the 16th International Conference on Digital Audio Effects (DAFx-13) CY - Maynooth, Ireland ER - TY - CONF T1 - Refined Spectral Template Models for Score Following T2 - Proceedings of the Sound and Music Computing Conference (SMC) Y1 - 2013 A1 - Korzeniowski, Filip A1 - Widmer, Gerhard AB - Score followers often use spectral templates for notes and chords to estimate the similarity between positions in the score and the incoming audio stream. Here, we propose two methods on different modelling levels to improve the quality of these templates, and subsequently the quality of the alignment. The first method focuses on creating more informed tem- plates for individual notes. This is achieved by estimating the template based on synthesised sounds rather than generic Gaussian mixtures, as used in current state-of-the-art systems. The second method introduces an advanced approach to aggregate individual note templates into spectral templates representing a specific score position. In contrast to score chordification, the common procedure used by score fol- lowers to deal with polyphonic scores, we use weighting functions to weight notes, observing their temporal relationships. We evaluate both methods against a dataset of classical piano music to show their positive impact on the alignment quality. JF - Proceedings of the Sound and Music Computing Conference (SMC) CY - Stockholm, Sweden ER - TY - Generic T1 - Rhytmic Pattern Modeling for Beat and Downbeat Tracking in Musical Audio T2 - Proceedings of the 14th International Society for Music Information Retrieval Conference (ISMIR 2013) Y1 - 2013 A1 - Krebs, Florian A1 - Sebastian Böck A1 - Widmer, Gerhard AB -Rhythmic patterns are an important structural element in music. This paper investigates the use of rhythmic pattern modeling to infer metrical structure in musical audio recordings. We present a Hidden Markov Model (HMM) based system that simultaneously extracts beats, downbeats, tempo, meter, and rhythmic patterns. Our model builds upon the basic structure proposed by Whiteley et. al, which we further modified by introducing a new observation model: rhythmic patterns are learned directly from data, which makes the model adaptable to the rhythmical structure of any kind of music. For learning rhythmic patterns and evaluating beat and downbeat tracking, 697 ballroom dance pieces were annotated with beat and measure information. The results showed that explicitly modeling rhythmic patterns of dance styles drastically reduces octave errors (detection of half or double tempo) and substantially improves downbeat tracking.
JF - Proceedings of the 14th International Society for Music Information Retrieval Conference (ISMIR 2013) ER - TY - CONF T1 - Tracking Rests and Tempo Changes: Improved Score Following with Particle Filters T2 - Proceedings of the International Computer Music Conference (ICMC) Y1 - 2013 A1 - Korzeniowski, Filip A1 - Krebs, Florian A1 - Andreas Arzt A1 - Widmer, Gerhard AB -In this paper we present a score following system based on a Dynamic Bayesian Network, using particle filtering as inference method. The proposed model sets itself apart from existing approaches by including two new extensions: A multi-level tempo model to improve alignment quality of performances with challenging tempo changes, and an extension to reflect different expressive characteristics of notated rests. Both extensions are evaluated against a dataset of classical piano music. As the results show, the extensions improve both the accuracy and the robustness of the algorithm.
JF - Proceedings of the International Computer Music Conference (ICMC) CY - Perth, Australia ER -