Discover the research challenges of the MTG

At the Music Technology Group (MTG), we conduct cutting-edge research at the intersection of science, technology, and the arts to understand, model, and generate sound and music. Our work addresses key societal and cultural challenges, with applications in:

  • Artistic Creation: Developing tools that empower creativity and expression.

  • Cultural Preservation: Enhancing access to and understanding of the world's musical heritage.

  • Music Education: Creating technologies that support and enrich music learning experiences.

  • Health and Wellbeing: Exploring how music technologies can contribute to personal and social wellbeing.

  • Sustainable Development: Designing methodologies to describe and preserve our sonic environments.

To advance these areas, we engage in diverse research topics, including:

Extracting and identifying relevant perceptual and musical features from audio signals is an essential step in the analysis and understanding of music. Our current research on audio feature analysis covers: pitch analysis, harmonic analysis, tonality analysis, onset analysis, dynamics analysis, timbre analysis, and audio problems detection.

References: Tamer et al. 2023; Perez et al. 2022; Narang et al. 2022; Plaja-Roglans et al. 2022;  H. Cuesta, 2022; Won et al. 2020; Alonso-Jiménez et al. 2019; Essentia algorithms.

This line of research focuses on learning general-purpose representations of sound and music that capture perceptual, semantic, and contextual properties and can be reused across multiple tasks. We develop audio embeddings trained on large-scale audio and metadata collections to support applications such as music classification, recommendation, and sound event recognition, moving beyond task-specific features toward reusable foundation representations.

References: Alonso-Jímenez 2024; Zinemanas 2023; Ferraro et al. 2021; Fonseca et al. 2021; Oramas 2017; Essentia models.

Machine listening enables computers to understand and interpret sound in real-world environments. Our research advances sound event classification, sound scene analysis, and automatic sound retrieval, contributing to technologies for media production, acoustic monitoring, and digital content management. 

References: Anastasopoulou et al. 2025; Anastasopoulou et al. 2024; Weck et al. 2023; Weck et al. 2022; Weck et al. 2021;  E. Fonseca, 2021; X. Favory, 2021; Zinemanas et al. 2021.

Music is much more than audio recordings, it is multimodal and therefore it can be studied through many types of signals. Related with this multimodality, we do research on topics such as: audio to score alignment, automatic audio captioning, musical gesture analysis, and music score analysis.

References: Weck et al. 2024; Weck et al. 2023; Zinemanas et al. 2023; Ramoneda et al. 2022; Weck et al. 2022; Ramoneda et al. 2022; Morsi et al. 2022; M. Won 2022; Weck et al. 2021; Favory et al. 2021; D. Cabrera Dalmazo, 2020; A.Morsi, 2020.

In the last few years there has been an incredible advancement on the topic of audio source separation. In our research we take different approaches depending on the application being targeted and considering specific trade-offs.

References: Plaja-Roglans et al. 2025; Shankar et al. 2025; Plaja-Roglans et al. 2023; Chanda et al. 2022; P. Chandna, 2021; O. Slizovskaia, 2021; Lluis et al. 2019; M. Miron, 2017.

Building on our extensive experience working on a variety of sound synthesis methodologies we are currently focusing on deep learning approaches to synthesize the singing voice and other musical instruments, such as drums.

References: M. Blaauw 2022; Ramires et al. 2022; Bonada et al. 2021; Blaauw et al. 2020; Ramires et al. 2020.

Creating music is a very complex process that current Artificial Intelligence approaches are trying to address. Related to this, we do research on symbolic and audio music generation and on the ethical issues related to it.

References: Batlle-Roca et al. 2025;  Haki, 2024; Batlle-Roca et al. 2024; Evans et al. 2024; Haki et al. 2022; Nuttall et al. 2021; Gómez et al. 2020; Haki et al. 2020.

The classification of musical signals is a major research topic within the field of music information retrieval, which supports many music applications currently being developed. We have been working on topics such as auto-tagging, mood classification, genre classification, instrument recognition, and representation learning.

References: Alonso-Jimenez et al. 2023; Buisson et al. 2022; Drysdale et al. 2022; Won et al. 2021; Gómez-Cañón et al., 2021; Won et al. 2020; O. Slizovskaia, 2020; J. Pons, 2019; Essentia models.

Music retrieval based on the analysis of audio signals relates to many specific research problems. In particular we have been working on: music version identification, music recommendation, search, and exploration, music metadata, and algorithmic biases.

References: Batlle-Roca et al. 2023; Schedl et al. 2022; L. Porcaro, 2022; P. Tovstogan 2022; F. Yesiler, 2022; A. Ferraro, 2021.

This research applies data-driven and computational methods to the large-scale study of musical cultures and repertoires. We focus on culture-specific music analysis, enabling new approaches to the study of musical traditions and supporting cultural heritage preservation and access.

References: Nuttall et al. 2025; Nuttall et al. 2024; Nuttall et al. 2023; Plaja-Roglans et al. 2023; Srinivasamurthy et al. 2023; Nuttall et al. 2022; R. Caro Repetto, 2018; Bogdanov et al. 2017; CompMusic.

The study of musical performances is especially relevant in the context of music education. Within that context we have been doing research on automatic performance assessment, score difficulty analysis, and music transcription.

References: Kim et al. 2025; Ramoneda et al. 2024; Kim et al. 2024; Tamer et al 2023; Morsi et al. 2023; Ramoneda et al. 2023; Kim et al. 2023; Kim et al. 2022; Ramoneda et al. 2022; Ramoneda et al. 2022; Eremenko et al. 2020; MusicCritic demos.

Adequate corpora and datasets are the basis for doing data-driven research and creating machine learning models. Over many years we have been creating large corpora for tackling broad sound and music research tasks and also we have been developing datasets for creating specific machine learning models. Related to these, it has been important the development of evaluation methodologies and of annotation protocols.

References: Shankar et al. 2024; Araz et al. 2024; Narang et al. 2023; Manco et al., 2023; Cortès et al. 2022; Bogdanov et al. 2022; Tamer et al. 2022; Srinivasamurthy et al. 2021; Fonseca et al. 2020; Ramires et al. 2020; Freesound; Open corpora and datasets.

Our research involves the development of efficient, reliable and robust software implementations and tools. For example, we have been working with Web audio, on efficient implementations of deep learning algorithms, on real-time implementations, and on complete MLOps systems.

References: Fuentes et al 2024; Anthon et al. 2023; Marcos-Fernández et al. 2022; Correya et al. 2023; Correya et al. 2021; Ramires et al. 2020; Zinemanas et al. 2020; Alonso-Jiménez et al. 2020; Alonso-Jiménez et al. 2020; Essentia.js; Open software.

Musical interfaces are needed to create music and a number of research topics focus on them. For example we work on topics such as: tangible interfaces, brain interfaces, hardware interfaces, and adaptive digital music instruments.

References: A. Hugues et al. 2021; F. Font, 2021; S. Jordà, 2017; Z. Vamvakousis et al. 2016

Music is known to have the power to induce strong emotions and can effectively impact the health and well-being of individuals. Music interventions have started to incorporate music technology for improving different conditions, e.g. emotional disorders, autism, cancer, Alzheimer, and stroke rehabilitation.

References: Ramirez et al. 2023; Ramirez et al. 2018; Ramirez et al. 2015.

Research outputs

Projects