Automatic Melody Harmonization via Reinforcement Learning by Exploring Structured Representations for Melody Sequences

We present a novel reinforcement learning architecture that learns a structured representation for use in symbolic melody harmonization. Probabilistic models are predominant in melody harmonization tasks, most of which only treat melody notes as independent observations and do not take note of substructures in the melodic sequence. To fill this gap, we add substructure discovery as a crucial step in automatic chord generation. The proposed method consists of a structured representation module that generates hierarchical structures for the symbolic melodies, a policy module that learns to break a melody into segments (whose boundaries concur with chord changes) and phrases (the subunits in segments), and a harmonization module that generates chord sequences for each segment. We formulate the structure discovery process as a sequential decision problem with a policy gradient RL method selecting the boundary of each segment or phrase to obtain an optimized structure. We conduct experiments on our preprocessed HookTheory Lead Sheet Dataset, which has 17,979 melody/chord pairs. The results demonstrate that our proposed method can learn task-specific representations and, thus, yield competitive results compared with state-of-the-art baselines.

Christian textual amulet on a lead sheet from Rękawczyn in Greater Poland Province

The subject of this article is the discovery of a lead amulet with an engraved Latin inscription. It was found in Rękawczyn (the eastern part of Greater Poland) in 2018 during research conducted by the District Museum in Konin. This artefact is the first and so far only find of its kind in a Polish context. It was identified by comparing it to similar finds from Germany and Bohemia from the 11th-13th centuries that bear apotropaic Latin inscriptions. The text on the amulet consists of 24 lines, which are barely legible and which contain a long quotation from the Gospel of John followed by an apotropaic formula with magical words. The reconstruction of the text is accompanied by a tentative interpretation and commentary. The palaeographical analysis of the script (Gothic minuscule) dates the amulet to the second half of the 14th century. The study presents the amulet in the wider context of similar inscriptions of European origin.

On the Adaptability of Recurrent Neural Networks for Real-Time Jazz Improvisation Accompaniment

Jazz improvisation on a given lead sheet with chords is an interesting scenario for studying the behaviour of artificial agents when they collaborate with humans. Specifically in jazz improvisation, the role of the accompanist is crucial for reflecting the harmonic and metric characteristics of a jazz standard, while identifying in real-time the intentions of the soloist and adapt the accompanying performance parameters accordingly. This paper presents a study on a basic implementation of an artificial jazz accompanist, which provides accompanying chord voicings to a human soloist that is conditioned by the soloing input and the harmonic and metric information provided in a lead sheet chart. The model of the artificial agent includes a separate model for predicting the intentions of the human soloist, towards providing proper accompaniment to the human performer in real-time. Simple implementations of Recurrent Neural Networks are employed both for modeling the predictions of the artificial agent and for modeling the expectations of human intention. A publicly available dataset is modified with a probabilistic refinement process for including all the necessary information for the task at hand and test-case compositions on two jazz standards show the ability of the system to comply with the harmonic constraints within the chart. Furthermore, the system is indicated to be able to provide varying output with different soloing conditions, while there is no significant sacrifice of “musicality” in generated music, as shown in subjective evaluations. Some important limitations that need to be addressed for obtaining more informative results on the potential of the examined approach are also discussed.

Comparing Word Affect and Tone Affect: Comment on Sun and Cuthbert 2017

In the article "Emotion Painting: Lyric, affect, and musical relationships in a large lead-sheet corpus", Sun and Cuthbert (2017) explored the correlations between affect-carrying lyrics and musical features such as beat strength, pitch height, consonance, and mode. Several musical features did indeed turn out to be highly correlated with the affect of the lyrics. However, correlations between other features, particularly mode-related musical features and lyric affect, were either insignificant or even contradicted previous research. In the current commentary, it is argued that the difference between the musical features that show significant correlations and those that do not is that the former have a local musical effect whereas the latter tend to affect the mood of a whole phrase or piece, and that the way Sun and Cuthbert estimate lyric affect for sentences or song may not be appropriate. Furthermore, a few remarks are made about the way Sun and Cuthbert treat multi-syllable words and about some basic assumptions concerning the relation between music and lyrics in a song. Nevertheless, the authors are praised for their innovative and interesting work, while several alternative and additional analyses (for example with scale-degree qualia and syncopations) are proposed.

Partimenti and the Power of Improvisation

Like a lead sheet in popular music, a partimento gives a performer a single line of music to aid in the performance of a multivoice composition. Lead sheets provide a simplified melody and symbols for chords. Partimenti provide a bass and sometimes figured-bass numbers to indicate specific intervals. In both cases the reconstruction works well if the performer has a good knowledge of the style of music involved and a memory for the kinds of musical patterns needed. In Naples, children played the written partimento with their left hands at a small harpsichord. With their right hands they improvised the types of melodies, chords, and counterpoints implied by the bass. Beginners may have improvised mostly simple chords. Intermediate-level students improvised melodies and counterpoints. And advanced students developed highly contrapuntal realizations that included partimento fugues.

Minimising Defect Formation in Sand Casting of Sheet Lead: A DoE Approach

Sand casting of lead sheet is a traditional manufacturing process used up to the present due to the special features of sand cast sheet such as their attractive sheen. Similarly to any casting process, sand casting of lead sheet suffers from the presence of surface defects. In this study, a surface defect type, hereby referred to as ‘grooves’, has been investigated. The focus has been laid on the identification of the main factors affecting defect formation in this process. Based on a set of screening experiments performed using Scanning Electron Microscopy (SEM) as well as the existing literature, a number of factors affecting the formation of such defects was identified and their corresponding significance was estimated using the Analysis of Variance (ANOVA) technique. The obtained results suggest that the most significant factor affecting defect formation in sand casting of lead sheet is the composition of the moulding mixture. Defect formation was also proven to be dependent on the sand grain fineness, the quality of the melt and some of the interactions between the aforementioned process parameters. Finally, an optimal set of process parameters leading to the minimisation of surface defects was identified.

Representing Harmony in Computational Music Cognition

Cognitive theories of harmony require unambiguous formal models of how listeners internally represent chords and chord progressions. Previous modeling work often uses representation schemes heavily reliant on Western music theory, such as Roman-numeral and lead-sheet notation; however, we argue that such work should be complemented by models using representations that are closer to psychoacoustics and rely less on Western-specific assumptions. In support of this goal, we compile a network of 13 low-level harmonic representations relevant for cognitive modeling, organised into three symbolic, acoustic, and sensory categories. We implement this collection of representations in an easy-to-use object-oriented framework written for the programming language R and distributed in an open-source package called hrep (http://hrep.pmcharrison.com). We also discuss computational methods for deriving higher-level representations from these low-level representations. This work should ultimately help researchers to construct high-level models of harmony cognition that are nonetheless rooted in low-level auditory principles.

Transcribing Lead Sheet-Like Chord Progressions of Jazz Recordings

The vast majority of research on automatic chord transcription has been developed and tested on databases mainly focused on genres like pop and rock. Jazz is strongly based on improvisation, however, and the way harmony is interpreted is different from many other genres, causing state-of-the-art chord transcription systems to achieve poor performance. This article presents a computational system that transcribes chords from jazz recordings, addressing the specific challenges they present and considering their inherent musical aspects. Taking the raw audio and minor manually obtained inputs from the user, the system can jointly transcribe chords and detect the beat of a recording, allowing a lead sheet–like rendering as output. The analysis is implemented in two parts. First, all segments with a repeating chord progression (the chorus) are aligned based on their musical content using dynamic time warping. Second, the aligned segments are mixed and a convolutional recurrent neural network is used to simultaneously detect beats and transcribe chords. This automatic chord transcription system is trained and tested on jazz recordings only, and achieves better performance than other systems trained on larger databases that are not jazz specific. Additionally, it combines the beat-detection and chord transcription tasks, allowing the creation of a lead sheet–like representation that is easy to interpret by both researchers and musicians.