MTG 2018 highlights
MTG 2018 highlights
by Xavier Serra
This blog post aims to overview the work done at the MTG during 2018, highlighting what we consider to be most relevant.
In relation to research, we first want to highlight the work done around the creation and use of our corpora and datasets. The largest effort was on the creation of FSD dataset and on the platform supporting it, which resulted in the organization of a large and successful machine learning challenge in Kaggle (Fonseca et al., 2018). We also continued improving our Dunya corpora, specially the Arab-Andalusian (Caro et al., 2018) and jingju music ones, and we published research using them (Pretto et al., 2018; Gong, 2018; Caro, 2018). Adding to the existing corpora, we started a new jazz corpus (Eremenko et al., 2018) that is being integrated into our Dunya platform. Another relevant and continued effort related to corpora was with AcousticBrainz, which resulted in our second MediaEval challenge on genre recognition (Bogdanov et al., 2018).
This last year we also continued our efforts on improving Essentia. We could highlight the new support for Tensorflow models and the inclusion of new algorithms related to fingerprinting, cover identification, and audio quality analysis. In general, there has been a significant increase of the research and development projects that use Essentia, both at the MTG and by other groups.
The increase of Deep Learning methods in our research has been considerable. We have explored different deep learning approaches for tasks such as genre classification (Kim et al., 2018; Oramas et al., 2018), acoustic scene classification (Fontseca et al., 2018), phoneme segmentation (Gong et al., 2018), audio tagging (Pons et al., 2018), or speech denoising (Rethage et al., 2018) and we have also worked on improving the actual learning models (Serrà et al., 2018).
An area of recent research activity at the MTG is the one related to music education. We have been developing an on-line environment for assessing musical exercises (Bozkurt et al., 2018), we have been working on the assessment of violin playing (Giraldo et al., 2018) and also assessing singing voice exercises (Gong et al. ,2018; Gómez et al., 2018). Another recent research interest of ours has been the study of musical emotions in the area of health (Ramirez et al., 2018).
The total number of research publications this last year was 60, including 6 journal articles and 34 peer reviewed conference articles. This number also includes the PhD thesis defended in 2018 by P. Herrera, R. Gong, C. Ó. Nuanáin, Á. Faraldo, D. Gómez, M. Miron, and R. Caro Repetto, which we consider the most relevant type of research outcomes of a group like the MTG. The master thesis of the SMC master finished last year, most of them supervised by researchers of the MTG, were also significant contributions, especially because they offered a great opportunity to explore new research ideas.
Most of the research at the MTG is carried out within funded projects. Let’s highlight some of their outcomes.
The largest project we have ever had at the MTG, CompMusic, despite having officially ended in 2017, was still quite active this last year and generated a number of relevant research outcomes, such as improvements to corpora and datasets and software tools, and new publications. The Riyaz app, being further developed and exploited by a spin-off of the project, had major improvements and attracted many new users. The last, and still active, funded initiative within CompMusic focused on the development of MusicCritic and its use in MOOCs like the one on Hindustani music. As a continuation of CompMusic, this last year we started MusicalBridges, a project funded by RecerCaixa that focuses on the education potential of our research results.
Two EU funded projects coordinated by the MTG that started three years ago, TELMI and AudioCommons, are finishing soon. With TELMI we have been able to support our research on multi-modal interactive analysis of violin performances and with AudioCommons we further developed the potential of Freesound (just went over 400000 sounds!) and its use by the creative industries. Important has been the successful consolidation of a system of micro-donations that should help sustain Freesound in the future.
In 2018 we started two new EU funded projects, TROMPA and MIP-Frontiers. TROMPA aims to enrich and democratize our publicly available musical heritage through a user-centered co-creation setup, and MIP-Frontiers is a training network within the area of Music Information Processing. Humaint is a special new project lead by Emilia Gómez at the EC Joint Research Centre for Advanced Studies in Seville aiming to understand the impact of machine intelligence on human behavior.
This last year was significant for the number of industrial projects we carried out as part of our extensive industrial collaborations. We got a Google research award to work on Freesound Datasets and we started projects with Kakao, La Cúpula, and Flitz, a part from continuing our long lasting relationship with Yamaha.
Most of the artistic and cultural projects of the MTG are developed within Phonos and this last year we made an important push to them. Ángel Faraldo was appointed Artistic Director of Phonos and with him we rethought the existing Phonos activities and created new ones. The programme for this academic year, that started last October, is more ambitious than ever. As MTG we also organized a number of dissemination activities to promote our field of research. Specially significant was the European Research Music Conference which brought together a significant number of the projects funded by the European Research Council that have music as their focus.
The MTG is a large and active research group in constant evolution and 2018 was no exception. Many students and researchers moved on to new adventures after having contributed to make the MTG a better place (former members) but also many new people joined our group with the aim to grow with us (team members). This great crowd of past and present MTG members is our main asset and what allowed us to carry out all the great things we did in 2018.