We introduce MuVGM, a large‑scale Multimodal dataset for Video Game Music research, combining symbolic transcriptions, reference audio, gameplay videos, box art imagery, textual descriptions, audio captions, and affective annotations across a broad range of titles and platforms. Video game music is uniquely suited to multimodal research because it is composed for specific, directly observable visual environments and narrative contexts, providing explicit functional grounding absent from most existing music datasets. The dataset comprises 7129 tracks from 1484 games across 39 consoles, spanning 37 years of game music history, and was constructed through a six‑stage pipeline combining web scraping, Wikipedia matching, cover song detection, audio fingerprinting, retrieval‑augmented text generation, and automatic music captioning. The dataset, including an interactive web‑based dataset explorer, is available at https://tylerhmc.com/muvgm/.
Artificial intelligence (AI) has become central to debates about the future of music, particularly with the rise of generative systems and their societal, economic, cultural, and legal implications. Public and policy discourse, however, often follows commercial narratives that narrow music‑AI to automated generation and treat music primarily as data or commodified output. This obscures both the broader range of AI applications in music and the long‑standing contributions and responsibilities of the music information retrieval (MIR) community. Addressed primarily to this community, the article argues that MIR is well positioned to contribute to current debates on trustworthy music‑AI but must also critically examine the limitations of its own research practices and incentives. We briefly situate MIR’s methodological development, applications, and shared infrastructures, emphasizing the culturally and socially situated character of music data. We then examine ethical, legal, and governance challenges in music‑AI and propose transparency, accountability, provenance, and sustainability as criteria for assessing current practice and guiding future work. Finally, we identify six interconnected priorities: mission‑oriented collaboration; sustainable open infrastructures; rigorous and context‑sensitive evaluation; cultural diversity and responsibility; engagement across research, industry, policy, and society; and interdisciplinary education and mutual capacity‑building. We conclude by identifying ways in which the MIR community can translate these priorities into practice through its research, infrastructures, evaluation and publication processes, educational activities, and engagement with musical, industry, policy, and civil society communities.
Signature Sound Vienna is an interdisciplinary digital musicology investigation of the Vienna Philharmonic Orchestra’s New Year’s Concert series, conducted by reference to a collection of relevant concert recordings and scores assembled for this purpose. In this paper, we present an overview of the workflows and software tools developed co‑creatively between the project’s music scholars and technologists, aiming to facilitate the publication, analysis and dissemination of findings pertaining to musicological research data while adhering to FAIR principles of data management. These workflows and tools include automated conversion scripts to publish metadata of large collections of music media, retrieved and validated using the MusicBrainz Picard Tagger, as Linked Data adhering to the Music Ontology; mei‑friend, a browser‑based editor for music encodings; Listen Here!, an environment for tool‑assisted close listening to score‑aligned collections of performance recordings; and Primal, a platform for review and interaction with music annotation Linked Data. Building on Web standards and established ontologies, including the Solid platform for social Linked Data, the Web Annotation Data Model, the Music Annotation Ontology and the schema of the Music Encoding Initiative (MEI), our modular tools integrate on shared Linked Data structures. This supports complex musicological investigations through comparatively simple applications focusing on specific tasks and modalities, as well as enabling further extensions beyond the scope of our project. We present a case study of the use of this approach within our own scholarly investigations, showcasing findings from performance analyses of collections of the most frequently performed compositions of the New Year’s Concert series, and discuss the implications of our approach on broader digital musicology contexts.
Extreme vocal techniques like growling and shrieking are common in metal and related genres, giving the music a characteristic intensity. Although widely used, these techniques have received limited attention in Music Information Retrieval (MIR) and musicology, particularly in fine‑grained categorisation. Their niche status and marginalisation have led to limited representation in digital music libraries. Since current classification systems prioritise dominant subgenres due to a lack of detailed vocal categorisations, extreme vocal styles are often oversimplified or misrepresented, hindering tagging, retrieval, and analysis. Automation could enable consistent and nuanced vocal annotation, supporting the technical goals of MIR and analytical aims of musicology. Foregrounding vocal technique as a classification axis contributes to a richer understanding of vocal performance and genre formation. To assess machine learning as a reliable alternative to human judgment, this study developed a coarse frequency‑based qualitative taxonomy from interviews with 11 extreme vocal practitioners and compared human and machine performance using the Extreme Metal Vocals Dataset (EMVD), including a listening test with 158 expert participants. Perceptual results achieved 76.2% Unweighted Average Recall (UAR), while a Support Vector Machine (SVM) model trained on ComParE features achieved a UAR of 90% on average, using singer‑independent 3‑fold group cross‑validation. Feature group, ablation, and human prediction analyses suggest that the human‑machine gap reflects differences in cue prioritisation and classification performance. Results illustrate the feasibility of machine learning to annotate vocal styles, while showing that human confusion patterns remain informative for category boundaries. This highlights how MIR methods could facilitate analysis of underrepresented genres.
