This paper introduces the Wagner Ring Dataset (WRD), a multi-modal and multi-version resource on the large-scale opera cycle Der Ring des Nibelungen by Richard Wagner. The Ring comprises four music dramas organized into eleven acts and 21 939 measures in total. Concerning sheet music, we processed a publicly available piano reduction (822 pages) of the full score with optical music recognition followed by extensive manual corrections to create a high-quality, machine-readable symbolic score. Concerning audio data, our corpus covers 16 recorded performances of the full Ring (three of which are publicly available thanks to copyright expiry), each lasting about 14–15 hours. To musically synchronize these versions among each other, we manually annotated all measure positions for three performances, which we transferred to the remaining performances via automated synchronization techniques. The dataset further comprises annotations of key and time signatures, scenes, and singing voice regions (libretto). Moreover, we provide note event annotations for all performances derived from the piano score. The WRD thus constitutes a comprehensive resource for developing algorithms for various music information retrieval tasks, complementing existing datasets with a complex opera scenario. For computational musicology, the WRD serves as a structured dataset that allows for studying the composition and performances of the Ring.

Automatically detecting the presence of singing in music audio recordings is a central task within music information retrieval. While modern machine-learning systems produce high-quality results on this task, the reported experiments are usually limited to popular music and the trained systems often overfit to confounding factors. In this paper, we aim to gain a deeper understanding of such machine-learning methods and investigate their robustness in a challenging opera scenario. To this end, we compare two state-of-the-art methods for singing voice detection based on supervised learning: A traditional approach relying on hand-crafted features with a random forest classifier, as well as a deep-learning approach relying on convolutional neural networks. To evaluate these algorithms, we make use of a cross-version dataset comprising 16 recorded performances (versions) of Richard Wagner’s four-opera cycle Der Ring des Nibelungen. This scenario allows us to systematically investigate generalization to unseen versions, musical works, or both. In particular, we study the trained systems’ robustness depending on the acoustic and musical variety, as well as the overall size of the training dataset. Our experiments show that both systems can robustly detect singing voice in opera recordings even when trained on relatively small datasets with little variety.

Instrument activity detection is a fundamental task in music information retrieval, serving as a basis for many applications, such as music recommendation, music tagging, or remixing. Most published works on this task cover popular music and music for smaller ensembles. In this paper, we embrace orchestral and opera music recordings as a rarely considered scenario for automated instrument activity detection. Orchestral music is particularly challenging since it consists of intricate polyphonic and polytimbral sound mixtures where multiple instruments are playing simultaneously. Orchestral instruments can naturally be arranged in hierarchical taxonomies, according to instrument families. As the main contribution of this paper, we show that a hierarchical classification approach can be used to detect instrument activity in our scenario, even if only few fine-grained, instrument-level annotations are available. We further consider additional loss terms for improving the hierarchical consistency of predictions. For our experiments, we collect a dataset containing 14 hours of orchestral music recordings with aligned instrument activity annotations. Finally, we perform an analysis of the behavior of our proposed approach with regard to potential confounding errors.

Ballistic electron emission microscopy (BEEM) has been performed on both $\mathrm{Au}∕{\mathrm{Fe}}_{81}{\mathrm{C}}_{19}∕\mathrm{Si}$ (001) and $\mathrm{Au}∕\mathrm{Fe}∕\mathrm{Si}$ (001) Schottky diodes at $80\mathrm{K}$. The Schottky heights were measured to be $0.68\pm 0.02\mathrm{eV}$ and $0.70\pm 0.02\mathrm{eV}$ for the ${\mathrm{Fe}}_{81}{\mathrm{C}}_{19}∕\mathrm{Si}$ (001) and $\mathrm{Fe}∕\mathrm{Si}$ (001) interfaces, respectively. In addition, a second threshold voltage was observed for the $\mathrm{Fe}∕\mathrm{Si}$ (001) interface at $1.29\pm 0.04\mathrm{eV}$ and attributed to an additional conduction band minimum at the interface that arises from the bonding of the Fe to the Si. The hot electron attenuation lengths at $1.25\mathrm{eV}$ were measured to be $3.5\pm 1.0\mathrm{nm}$ and $3.0\pm 0.9\mathrm{nm}$ for ${\mathrm{Fe}}_{81}{\mathrm{C}}_{19}$ and Fe, respectively. The attenuation length of the ${\mathrm{Fe}}_{81}{\mathrm{C}}_{19}$ showed a decrease with increasing energy consistent with the universal curve for electron-electron scattering. However, the attenuation length for the Fe showed this decrease only until the onset of the second threshold after which it increased. It is proposed that this increase is attributed to the parallel momentum distribution of the additional conduction band minimum at the $\mathrm{Fe}∕\mathrm{Si}$ (001) interface.

Expressed by Fitts in 1954, there is an inverse relationship between movement time and target size and/or the distance between the participant and a target; smaller and further away targets require longer movement times than larger, closer targets. Here, the applicability of Fitts’ law to an in-vehicle, dual-task setting is under investigation. In the current experiment, different target sizes and spaces between target and non-target elements (target-element space), were investigated in simulated driving and non-driving scenarios in two age groups (18 – 35 and 45 – 65 years old). The aim of this experiment was to investigate whether target size and target-element space (i.e., the space between targets and non-targets) would have an effect on the mean movement times of younger and older participants for driving and non-driving conditions. Participants pointed to and touched targets on a smartphone touch screen in a vehicle mock-up. Movement times were recorded while concurrently performing a simulated driving task (dual-task condition) and while performing just the pointing task (single task condition). Errors were defined and are presented. Mean movement times in the dual-task condition were significantly faster than in the single task condition. Additionally, in line with Fitts’ law, in-vehicle movement time decreased as a function of target size. Elements that had the most target-element space were moved to most quickly, however, the medium and small spaces did not differ. Differences between the two age groups were not found