A score identification parallel system based on audio-to-score alignment

  • PDF / 1,530,880 Bytes
  • 15 Pages / 439.37 x 666.142 pts Page_size
  • 29 Downloads / 181 Views

DOWNLOAD

REPORT


A score identification parallel system based on audio‑to‑score alignment A. J. Muñoz‑Montoro1   · R. Cortina2 · S. García‑Galán1 · E. F. Combarro2 · J. Ranilla2

© Springer Science+Business Media, LLC, part of Springer Nature 2020

Abstract This paper presents a parallel system for searching a digital score of classical music in a personal library. The application scenario of the system is for a musician who wants to search for a specific score in its own device by playing an excerpt of a few seconds of the composition. We propose a solution, based on audio-to-score alignment, which allows to identify the correct score in a database of musical pieces in real time. This is a challenging task because we focus on a real-time system targeted for handheld devices characterized by both mobility and low power consumption. Experimental results show that it is possible to achieve real-time execution in the tested scenarios using parallel computing techniques with ARM processors. Keywords  Score identification · Parallel computing · Real time · Audio-to-score alignment · DTW · Audio processing

* A. J. Muñoz‑Montoro [email protected] R. Cortina [email protected] S. García‑Galán [email protected] E. F. Combarro [email protected] J. Ranilla [email protected] 1

Department of Telecommunication Engineering, Universidad de Jaén, Linares, Spain

2

Department of Computer Science, Universidad de Oviedo, Oviedo, Spain



13

Vol.:(0123456789)



A. J. Muñoz‑Montoro et al.

1 Introduction Paper sheet music is often a source of annoyance for musicians for several reasons. Music scores can be voluminous to store and uncomfortable to handle while performing. They can become deteriorated from repeated use and brittle with age. When a music score is placed on a music stand, it can be easily dislodged causing some very anxious moments for the performer. Furthermore, nothing is more frustrating than thumbing through paper sheet music trying to find the desired music. In this way, advancements in technology during the last decade have resulted in rapid growth of software for displaying digital scores. Examples include Beatik,1 which is a page-turning software that adjusts to the musician playing in real time; Antescofo,2 which is a musical accompaniment software for classical musicians; Tonara,3 which is a music tracking platform focusing on the amateur musicians; and Newzik,4 which is a digital score display system. All of these applications allow users to store digital scores in their personal libraries. However, as users use this kind of tools, their personal libraries can grow to the point that searching a specific digital score can become a tedious task. This paper addresses the problem of searching a digital score of classical music in a personal library. Imagine a musician who wants to search for a specific score in its own device and decides to play an excerpt of a few seconds; the system should present the user with the title, the composer, and display the digital sheet music. Therefore, given a short audio query corresponding t