eduzhai > Applied Sciences > Engineering >

Deep Learning Based Source Separation Applied To Choir Ensembles

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 7 pages

Abstract: Choral singing is a widely practiced form of ensemble singing wherein a groupof people sing simultaneously in polyphonic harmony. The most commonlypracticed setting for choir ensembles consists of four parts; Soprano, Alto,Tenor and Bass (SATB), each with its own range of fundamental frequencies(F$0$s). The task of source separation for this choral setting entailsseparating the SATB mixture into the constituent parts. Source separation formusical mixtures is well studied and many deep learning based methodologieshave been proposed for the same. However, most of the research has been focusedon a typical case which consists in separating vocal, percussion and basssources from a mixture, each of which has a distinct spectral structure. Incontrast, the simultaneous and harmonic nature of ensemble singing leads tohigh structural similarity and overlap between the spectral components of thesources in a choral mixture, making source separation for choirs a harder taskthan the typical case. This, along with the lack of an appropriate consolidateddataset has led to a dearth of research in the field so far. In this paper wefirst assess how well some of the recently developed methodologies for musicalsource separation perform for the case of SATB choirs. We then propose a noveldomain-specific adaptation for conditioning the recently proposed U-Netarchitecture for musical source separation using the fundamental frequencycontour of each of the singing groups and demonstrate that our proposedapproach surpasses results from domain-agnostic architectures.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...