HFV: A Multimodal Singing Dataset for Hehuang Hua'er

To facilitate the structured preservation and computational analysis of Hehuang Hua'er performance materials, this study presents HFV, a multimodal singing dataset for Hehuang Hua'er. HFV contains 228 multimodal singing samples with a total video duration of 14,301.72 s (approximately 3.97 h), covering 223 unique songs, 72 quling types, 7 sentence patterns, and 9 topic categories.

Each sample consists of a singing video, aligned audio directly extracted from the embedded video track, lyric text, cultural labels, and structured metadata. A unified sample identifier is used to establish sample-level associations among video, audio, text, and cultural attributes. Data quality is evaluated from four aspects: structural completeness, statistical associations between lyrics and cultural labels, basic acoustic analyzability, and the analyzability of singing-related visual information.