19 Sep 2013

José R. Zapata defends his PhD thesis entitled "Comparative Evaluation and Combination of Automatic Rhythm Description Systems" on Thursday 19th of September 2013 at 16:00h in room 55.410.

The jury members of the defense are: Fabien Gouyon (INESC-Porto), Juan Bello (NYU), Xavier Serra (UPF).

Abstract: The automatic analysis of musical rhythm from audio, and more specifically tempo and beat tracking, is one of the fundamental open research problems in Music Information Retrieval (MIR) research. Automatic beat tracking is a valuable tool for the solution of other MIR problems, as it enables beat-synchronous analysis of music for tasks such as: structural segmentation, chord detection, music similarity, cover song detection, automatic remixing and interactive music systems. Even though automatic rhythm description is a relatively mature research topic in MIR and various algorithms have been proposed, tempo estimation and beat tracking remain an unsolved problem. Recent comparative studies of automatic rhythm description systems suggest there has been little improvement in the state of the art over the last few years. In this thesis, we describe a new method for the extraction of beat times with a confidence value from music audio, based on the measurement of mutual agreement between a committee of beat tracking systems. Additionally, we present an open source variant of the approach which only requires a single beat tracking model and uses multiple onset detection functions for the mutual agreement. The method can also be used identify music samples that are challenging for beat tracking without the need for ground truth annotations. Using the proposed method, we compile a new dataset that consists of pieces that are difficult for state-of-the-art beat tracking algorithms. Through an international evaluation framework we show that our method yields the highest AMLc and AMLt accuracies obtained in this evaluation to date. Moreover, we compare our method to 20 reference systems using the largest existing annotated dataset for beat tracking and show that it outperforms all the other systems under all the evaluation criteria used. In the thesis we also conduct an extensive comparative evaluation and combination of automatic rhythm description systems. We evaluate 32 tempo estimation and 16 beat tracking state-of-the-art systems in order to identify their characteristics and investigate how they can be combined to improve performance. Finally, we propose and evaluate the use of voice suppression algorithms for music signals with predominant vocals in order to improve the performance of existing beat tracking methods.