SCIENTIFIC DEFINITION OF AI VIRTUAL SINGERS: COMPARISON WITH AI CLONING SINGERS AND AI-GENERATED MUSIC

Abstract

This study aims to establish a scientific definition of AI Virtual Singers by distinguishing them from AI-cloned singers and AI-generated music. Data were collected from the Chinese video-sharing platform Bilibili using the Chinese search terms “虚拟歌手” (virtual singer) and “AI虚拟歌手” (AI virtual singer) on April 20, 2026, China Standard Time (UTC+8). Platform observation, content analysis, and comparative analysis were used to examine the top ten video results by playback volume and comprehensive ranking for each search term. The results show that, for “虚拟歌手,” the playback-volume list consisted entirely of voice synthesis systems (100%), while the comprehensive-ranking list contained 70% voice synthesis systems and 30% AI-generated music. For “AI虚拟歌手,” the playback-volume list included 50% AI-cloned singers, 30% voice synthesis systems, 10% AI-generated music, and 10% unclear cases; the comprehensive-ranking list included 50% voice synthesis systems, 40% AI-generated music, and 10% unclear cases. Based on these findings, this study defines AI Virtual Singers in the narrow scientific sense as digital human voice synthesis systems underpinned by deep-learning acoustic models, usually with an independent virtual persona or voice-library IP, capable of generating highly realistic vocal sounds with micro-parameter controllability from melodies (MIDI) and lyrics provided by human creators.