MIP FRONTIERS Summer school open day: An Industrial perspective to Music Information Retrieval
Seminar organized in the context of MIP-Frontiers project, a multidisciplinary, transnational and cross-sectoral European training network for MIR researchers that aims to train a new generation of researchers.
13.05.2019
Date: Monday, 20th of May 2019
Time: 15:00 to 18:00
Location: Room 55.309, Tanger building, Campus Poblenou-UPF Barcelona
15:00 MIP-Frontiers project. What we do and who we are
Simon Dixon, Queen Mary University of London
Estefania Cano, A*STAR Singapore / Fraunhofer IDMT Germany
Emilio Molina, BMAT
16:30 coffee break
Daniel Wolf, Tido Music
Matthias Roeder, Herbert von Karajan Institute
18:00 Close
In this talk, I will give a brief overview of the role that application-oriented research institutions such as the Fraunhofer Society - Germany, A*STAR - Singapore, and AIST - Japan play in the research ecosystem. The importance of these institutions in bridging the gap between basic research and industrial/commercial exploitation will be highlighted. I will then present a series of research projects with direct industry application conducted at the Semantic Music Technologies Group at Fraunhofer IDMT. In particular, I will describe the process behind the transformation of Songs2See from a research idea to a commercial product.
BMAT tells everyone what music has been played so that artists get the recognition they deserve. For it, we continuously analyze the audio from more than 6000 channels (TVs, radios or venues) using a set of technologies: audio fingerprinting (against a catalog of tens of millions of tracks), music detection, and under some conditions, cover song identification. In this talk we will provide details about how these technologies relate with the various BMAT services and products, and we will provide our view about the research challenges that arise when applying these technologies at industrial scale.
Since the arrival of the "Web 2.0", user data has not only been at the heart of world-leading IT companies, but also relatively easy to acquire, collect and aggregate. Based on such data, available machine learning methods allow for the creation of useful tools such as search engines, recommendation, type correction and automatic translation. Recent technologies allow analysis and prediction on the basis of image and audio data, while much ground truth data such as tags are collected from user annotations. This combination of annotations and machine learning technology renders previously "hard" data types such as video, audio and image accessible to large-scale analysis and research. The new potential for exploitation also requires careful planning when data is collected, aggregated and published. Current regulations mostly protect straightforward cases with data such as email addresses, which are easily identified as personal identifiable information. Other data types may require a closer look. We will look at software and frameworks useful for data collection and planning for data protection.
Matthias Roeder: Why the music business is dead, and a computational understanding of music is the way out
My talk gives an overview on the development of the market for music, highlighting the deterioration of margins for usage of content in the digital space. What are some strategies that companies are developing to deal with this massive loss of revenue? How are consumers reacting to these developments? I argue that the contextualization of digital content as well as the usage of information derived from the content directly might provide additional value to stakeholders. Going forward I see an increased importance for interaction with content, as we can see from video games, as well as co-creation and open art forms.