Huangmei Opera Dataset HOSD

High-quality annotated datasets serve as the foundation for optimizing the performance of speech and singing synthesis systems. Given the uniqueness of Huangmei Opera vocalization in terms of acoustic features and artistic expression, existing public opera datasets are not only scarce but also have coarse-grained annotation systems that do not align with the fine-grained modeling requirements of Huangmei Opera, making it difficult for deep learning models to accurately capture the essence of opera art. To address this research gap, this chapter constructs a dedicated acoustic dataset for Huangmei Opera vocalization synthesis, named Huangmei Opera Singing Dataset (HOSD). This dataset, which undergoes standard sampling in a professional acoustic environment, cross-dimensional fine-grained annotation, and rigorous quality control, comprehensively covers the vocal characteristics, effective vocal range, and ornamentation techniques (such as drag, tremolo, and glissando) of Huangmei Opera, providing sufficient data support and domain knowledge for neural network training.