Deep learning is bringing breakthroughs to many computer vision subfields including Optical Music Recognition (OMR), which has seen a series of improvements to musical symbol detection achieved by using generic deep learning models. However, so far, each such proposal has been based on a specific dataset and different evaluation criteria, which made it difficult to quantify the new deep learning-based state-of-the-art and assess the relative merits of these detection models on music scores. In this paper, a baseline for general detection of musical symbols with deep learning is presented. We consider three datasets of heterogeneous typology but with the same annotation format, three neural models of different nature, and establish their performance in terms of a common evaluation standard. The experimental results confirm that the direct music object detection with deep learning is indeed promising, but at the same time illustrates some of the domain-specific shortcomings of the general detectors. A qualitative comparison then suggests avenues for OMR improvement, based both on properties of the detection model and how the datasets are defined. To the best of our knowledge, this is the first time that competing music object detection systems from the machine learning paradigm are directly compared to each other. We hope that this work will serve as a reference to measure the progress of future developments of OMR in music object detection.

It has been previously proposed that syllable and word boundaries in Gregorian chant texts can be used to segment chant melodies in a more meaningful way than segmentation methods that do not take textual information into account, based on how accurately the mode of a melody can be determined from the presence of such melodic segments. This was evidenced by empirical measurements on antiphons and responsories with fully transcribed melodies available from the Cantus database. We show that for antiphons, however, this result does not hold, as in these experiments, differentiae were not removed from the transcribed melodies. With appropriate data cleaning, the modality of a melody can be determined from segments that ignore textual boundaries just as accurately, and the resulting classification scores are not significantly better than those obtained from pitch profiles. Thus, while the idea is clearly attractive, there is currently no reason to suspect that textual boundaries lead to a more meaningful segmentation of chant melodies.

Gregorian chant was a central musical tradition in Medieval Latin Europe and one of the expressions of its cultural unity: any Latin Christian liturgy, such as the weekly Sunday mass, would have involved Gregorian chant as a major part of the prescribed ritual. The Gregorian legend of chant melodies’ divine origin required the practitioners to conserve them, to the extent that this requirement motivated the development of exact pitch notation. Nevertheless, surviving manuscripts document a considerable melodic diversity. Some systematic patterns within this melodic diversity have previously been observed in chant scholarship, especially during efforts to build a critical edition reconstructing the earliest possible forms of chant melodies with philological approaches. Taking an evolutionary perspective, we notice analogies between biological evolution and processes of chant transmission, which lead us to suggest recovering these ‘melodic dialects’ using phylogenetic methods instead. In this paper, we show that phylogenetic models recover historically plausible patterns of chant melody evolution. We observe that some, but not all, institutional networks play a more important role than geographical proximity. Phylogeny is shown to be a viable class of methods for studying chant melody, and we discuss next steps for a more comprehensive evolutionary approach to chant. This article is part of the theme issue ‘Transforming cultural evolution research and its application to global futures’.