A Distributed Metadata Platform for Hybrid Radio Services

Content-based Recommendations for Radio Stations with Deep Learned Audio Fingerprints

“…Another issue for CF and DF systems are so-called Filter Bubbles, describing the creation of closed-off, synthetic environments, in which always the same items are recommended, disregarding the existence of contrary or different items beyond the bubble [Pa11,GLH19]. In contrast to that, an initial issue of CBF-based systems is the need for high-quality metadata precisely describing the items to recommend [Fr19]. The concept presented in this paper is part of the HRADIO Communication Platform, which provides a vast amount of metadata and audio information within a hybrid radio context [Fr19] and is completely open source.…”

“…In contrast to that, an initial issue of CBF-based systems is the need for high-quality metadata precisely describing the items to recommend [Fr19]. The concept presented in this paper is part of the HRADIO Communication Platform, which provides a vast amount of metadata and audio information within a hybrid radio context [Fr19] and is completely open source. For this reason, and to avoid issues like Filter Bubbles or the Cold Start Problem, it takes a CBF-based approach.…”

“…This chapter details the steps of the proposed analysis pipeline consisting of a Data Collector, the auDeep autoencoder, the Fingerprinter, and a Recommender, as depicted in Figure 1. In this context, the Data Collector unit (see Chapter 3.1) ensures the binding of the pipeline to the HRadio API (which is used for requesting a list of radio services from the HRADIO Communication Platform [Fr19]), records their audio streams and extracts audio bytes in predefined intervals. The raw audio snippets are transformed into mel-scaled spectrograms.…”

“…Therefore, on the one hand a large amount of wellstructured and rich data is needed, on the other hand different concepts for the analysis of meta and audio data have to be selected, implemented and carefully evaluated. This paper focuses on the latter and thus on the question of how meaningful recommendations for radio stations can be generated on the basis of rich metadata provided by the HRADIO Communication Platform, which was implemented by the authors within the scope of previous work [Fr19]. The findings and implemented concepts presented in this work are part of the open accessible HRADIO project7.…”

“…Thus, a CBF-based recommender system working with characteristics derived from available station data is preferred in context of this work. In [Fr19], it could be shown that station recommendation based on available metadata is possible, but only if the metadata reaches a certain quality level. This is provably often not the case.…”