Automatic Identification of Simultaneous and Non-Simultaneous Singers for Music Data Indexing

Abstract

This study proposes a system to automatically identify multiple singers in a long audio stream that may have singing voices overlapping in time. The system is of great help in handling the rapid proliferation of music data. To achieve this, an audio stream is segmented into a sequence of consecutive, non-overlapping, fixed-length clips using a sliding window, and then undergoes solo/duet recognition, single singer identification, and duet singer identification. Furthermore, the authors propose a triangulation-based decision approach to improve the performance of singer identification. The results of the experiment conducted using a database of solo and duet a cappella demonstrate feasibility of the proposed system.