Data science, the ubiquitous technology nowadays, has brought huge changes and raised many issues. Its material, method, and product are big data, machine learning, and artificial intelligence (AI) respectively. It also plays an important role in the music industry. Departments including electrical engineering and computer science have been offering related courses in foreign and domestic universities such as MIT, NTU, NTHU, NCKU, and NCNU. By contrast, music departments in Taiwan rarely offer complementary courses relevant to data science. That situation builds, unfortunately, an unhealthy ecosystem for interdisciplinary higher education in music.
Thanks to Talent Cultivation Project for Digital Humanities (TCDH) by Ministry of Education (MOE), the author of this paper has the chance to create an undergraduate-level beachhead in Tainan National University of the Arts (TNNUA), where few (if not zero) students have STEM (science, technology, engineering, and mathematics) majors. The author has been offering STEM courses, e.g. statistics, data science, and music technology in TNNUA since 2019. He also provided parallel MOOCs (massive open online courses) and off-campus workshops with other music teachers and entrepreneurs in 2021 for outreach education to elementary and high school students.
Based on the author’s experience during the MOE TCDH projects, this paper proposes a course framework of music data science. The workflow in this subject may consist of three levels. First, the upstream level encompasses music theory, data science theory, and data ethics. Second, the midstream level contains expert manual annotation practices, programming languages, data mining, knowledge discovery, and machine learning. Third, the downstream level includes AI services or products, AI-assisted composition, improvisation, and audio mixing (as known as intelligent music production).
The learning or teaching sequence, however, is not necessary to follow those up-mid-down streams. This paper proposes a two-phase progression. The initial one is music artificial intelligence (MAI), which starts from the upstream level and jumps to the downstream one. By circumventing the midstream level, students will not encounter difficulty and feel frustration too early. After they enjoy AI products and appreciate AI-assisted works, the course(s) should gradually enter the next phase, music information retrieval (MIR). This phase corresponds to the midstream level so that students accomplish the whole workflow. Each of the two phases could be a quarter, a semester, or even an academic year, depending on the depth and width covered in the course framework.
Collaborative teaching is recommended to support the course framework. It may ease the instructor’s burden. Nevertheless, caution should be exercised when the course(s) is offered by multiple teachers. One of the challenges is the overall cohesion of instructions. The other is the cost of collaborators’ time and budget, especially when external guest lecturers travelling to rural campuses. Another challenge is how to track and evaluate students’ performances by diverse faculty throughout the course framework.
The course framework has been comprehensively experimented and revised in the Department of Applied Music at TNNUA for two years. Next year, the university will incorporate it into the curriculum structure of the newly established Graduate Institute of Sound Technology.