BS-Roformer-HyperACE-Chinese-FolkOpera

专为中国民乐(二胡、板胡、高胡、琵琶、笛子等)与传统戏曲(京剧、秦腔、锡剧、粤剧等)唱腔优化的高精度人声/伴奏分离模型。

A specialized, high-performance vocal separation model adapted for Chinese Traditional Folk Instruments and Traditional Chinese Opera (Xiqu) singing, based on the BS-RoFormer architecture with HyperACE (Hypergraph Adaptive Context Enhancement).


模型亮点 / Highlights

  1. 民乐与戏曲专项增强:

    • 针对中国民族乐器(尤其是高胡、京胡、板胡、二胡等高亢擦弦乐器)与传统唱腔高频泛音高度交叠的难点场景进行了专项训练;
    • 在退化/老旧录音京剧测试集上,人声分离 SDR 大幅提升 +2.91 dB(SI-SDR 提升 +3.99 dB),极大改善了戏曲高音漏唱、破音与伴奏乐器残留问题。
  2. 保留极高通用流行乐水准:

    • 在常规流行乐歌曲测试集上 SDR 保持甚至微升 +0.05 dB,完全没有发生通用领域知识遗忘或性能倒退。
Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support