Towards Cross-Version Harmonic Analysis of Music

Proceedings of the 2009 joint international conference on Digital libraries - JCDL '09, 2009

Music retrieval systems for Western tonal music digital libraries have to consider rhythmic, timbral, melodic and harmonic information. Most existing retrieval systems only take into account melodies. Melody comparison may induce errors since two musical pieces can be very similar whereas their melodies may differ in a significant way. In this paper, we propose to investigate and experiment a retrieval system based on the comparison of chord progressions. The definition of chords may be ambiguous but their properties can be precisely described and represented. We detail the adaptations of alignment algorithms, successfully applied for the estimation of symbolic melodic similarity, for chord progression retrieval. Several experiments, performed on symbolic databases, show that the system described is robust to variations and outperforms a recent chord retrieval system.

2019

Automatic harmonic analysis is challenging: rule-based models cannot account for every possible edge case, and manual annotation is expensive and sometimes inconsistent, undermining the training and evaluation of machine learning models. We present an interactive workflow to address these problems, and test it on Bach chorales. First, a rule-based model was used to generate preliminary, consistent chord labels in order to pre-train three machine learning models. These four models were grouped into an ensemble that generated chord labels by voting, achieving 91.4% accuracy on a reserved test set. A domain expert then corrected only those chords that the ensemble did not agree on unanimously (20.9% of the generated labels). Finally, we used these corrected annotations to re-train the machine learning models, and the resulting ensemble attained an accuracy of 93.5% on the reserved test set, a 24.4% reduction in the number of errors. This versatile interactive workflow can either work i...

Proceedings of the 21st international conference companion on World Wide Web - WWW '12 Companion, 2012

In this paper we compare the use of different musical representations for the task of version identification (i.e. retrieving alternative performances of the same musical piece). We automatically compute descriptors representing the melody and bass line using a state-of-the-art melody extraction algorithm, and compare them to a harmony-based descriptor. The similarity of descriptor sequences is computed using a dynamic programming algorithm based on nonlinear time series analysis which has been successfully used for version identification with harmony descriptors. After evaluating the accuracy of individual descriptors, we assess whether performance can be improved by descriptor fusion, for which we apply a classification approach, comparing different classification algorithms. We show that both melody and bass line descriptors carry useful information for version identification, and that combining them increases version detection accuracy. Whilst harmony remains the most reliable musical representation for version identification, we demonstrate how in some cases performance can be improved by combining it with melody and bass line descriptions. Finally, we identify some of the limitations of the proposed descriptor fusion approach, and discuss directions for future research.

2017

This paper presents a fast and accurate alignment method for polyphonic symbolic music signals. It is known that to accurately align piano performances, methods using the voice structure are needed. However, such methods typically have high computational cost and they are applicable only when prior voice information is given. It is pointed out that alignment errors are typically accompanied by performance errors in the aligned signal. This suggests the possibility of correcting (or realigning) preliminary results by a fast (but not-so-accurate) alignment method with a refined method applied to limited segments of aligned signals, to save the computational cost. To realise this, we develop a method for detecting performance errors and a realignment method that works fast and accurately in local regions around performance errors. To remove the dependence on prior voice information, voice separation is performed to the reference signal in the local regions. By applying our method to results obtained by previously proposed hidden Markov models, the highest accuracies are achieved with short computation time. Our source code is published in the accompanying web page, together with a user interface to examine and correct alignment results.

Journal of Open Source Software, 2021

Music can be described and represented in many different ways, including as sheet music, symbolic representations, and audio recordings (Müller, 2015). For each of these representations, different versions (e.g., recordings performed by different orchestras and conductors) that correspond to the same musical work may exist. Music information retrieval (MIR) aims at developing techniques and tools for organizing, understanding, and searching this information in a robust, efficient, and intelligent manner. In this context, various alignment and synchronization procedures have been developed with the common goal of automatically linking several types of music representations, thus coordinating the multiple information sources related to a given musical work. In the design and implementation of synchronization algorithms, one has to deal with a delicate tradeoff between efficiency, robustness, and accuracy-requirements leading to various approaches with many design choices. In this contribution, we introduce a Python package called Sync Toolbox that provides open-source reference implementations for full-fledged music synchronization pipelines and yields state-of-the-art alignment results for a wide range of Western music. Using suitable feature representations and cost measures, the toolbox's core technology is based on dynamic time warping (DTW), which brings the feature sequences into temporal correspondence. To account for efficiency, robustness and, accuracy, our toolbox integrates and combines techniques such as multiscale DTW (MsDTW) (Müller et al., 2006; Salvador & Chan, 2004), memory-restricted MsDTW (MrMsDTW) (Prätzlich et al., 2016), and high-resolution music synchronization (Ewert et al., 2009). While realizing a complete system with presets that allow users to reproduce research results from the literature, our toolbox also provides well-documented functions for all basic building blocks required for feature extraction and alignment. Furthermore, the toolbox contains example code for visualizing, sonifying, and evaluating synchronization results, thus deepening the understanding of the techniques and data.

Large music content libraries often comprise multiple versions of a piece of music. To establish a link between different versions, automatic music alignment methods map each position in one version to a corresponding position in another version. Due to the leeway in interpreting a piece, any two versions can differ significantly, for example, in terms of local tempo, articulation, or playing style. For a given pair of versions, these differences can be significant such that even state-of-the-art methods fail to identify a correct alignment. In this paper, we present a novel method that increases the robustness for difficult to align cases. Instead of aligning only pairs of versions as done in previous methods, our method aligns multiple versions in a joint manner. This way, the alignment can be computed by comparing each version not only with one but with several versions, which stabilizes the comparison and leads to an increase in alignment robustness. Using recordings from the Mazurka Project, the alignment error for our proposed method was 14% lower on average compared to a stateof-the-art method, with significantly less outliers (standard deviation 53% lower).

IEEE Transactions on Audio, Speech, and Language Processing, 2011

In this paper we present algorithms for the automatic time-synchronization of score-, MIDI-or PCM-data streams which represent the same polyphonic piano piece. Since the waveform-based PCM-data streams do not contain any information on the notes we extract in a preprocessing step note parameters such as onset times and pitches in order to make the PCM-data comparable to symbolic score-like representations. In the extraction step we use novelity curves for onset detection and filter bank tree techniques in combination with note templates for pitch extraction. To handle ambiguities such as trills or arpeggios in the score data-stream we introduce the concept of fuzzy-notes. Further suitable normalization and quantization of the involved data streams are necessary to generate the input data of our synchronization algorithms which are based on the technique of dynamic programming. The decisive ingredient for our approach are carefully designed cost functions which will be explained in detail. Our synchronization algorithms have been tested on a variety of classical polyphonic piano pieces recorded by MIDI-and standard acoustic pianos or taken from CD-recordings.

2022

Partitura is a lightweight Python package for handling symbolic musical information. It provides easy access to features commonly used in music information retrieval tasks, like note arrays (lists of timed pitched events) and 2D piano roll matrices, as well as other score elements such as time and key signatures, performance directives, and repeat structures. Partitura can load musical scores (in MEI, MusicXML, Humdrum **kern, and MIDI formats), MIDI performances, and score-to-performance alignments. The package includes some tools for music analysis, such as automatic pitch spelling, key signature identification, and voice separation. Partitura is an open-source project and is available at https://github.com/CPJKU/partitura/.

2003

Music alignment links events in a score and points on the audio performance time axis. All the parts of a recording can be thus indexed according to score information. The automatic alignment presented in this paper is based on a dynamic time warping method. Local distances are computed using the signal's spectral features through an attack plus sustain note modeling. The method is applied to mixtures of harmonic sustained instruments, excluding percussion for the moment. Good alignment has been obtained for polyphony of up to five instruments. The method is robust for difficulties such as trills, vibratos and fast sequences. It provides an accurate indicator giving position of score interpretation errors and extra or forgotten notes. Implementation optimizations allow aligning long sound files in a relatively short time. Evaluation results have been obtained on piano jazz recordings.