SongDriver

SongDriver：零逻辑延迟、零曝光偏差的实时音乐伴奏生成系统

最新信息：SongDriver 现已完全开源！
代码、预训练权重、训练数据均已发布。

本仓库提供「开箱即用」的完整压缩包（含权重、代码与训练数据），对应论文：
《SongDriver: Real-time Music Accompaniment Generation without Logical Latency nor Exposure Bias》（ACM MM 2022）。

快速导航

• 最新代码与文档见：GitHub 仓库
• 一键完整下载（本页）：songdriver.zip 内含
  • 预训练权重（.ckpt）
  • 完整的训练 / 推理代码
  • 训练数据与预处理脚本

SongDriver 通过「编排阶段」与「预测阶段」并行，实现零逻辑延迟的实时伴奏生成，并显著降低曝光偏差。

Citation

If you use this work in your research, please cite our paper:

引用

如果本工作对您有帮助，请引用：

@inproceedings{10.1145/3503161.3548368,
author = {Wang, Zihao and Zhang, Kejun and Wang, Yuxing and Zhang, Chen and Liang, Qihao and Yu, Pengfei and Feng, Yongsheng and Liu, Wenbo and Wang, Yikai and Bao, Yuntao and Yang, Yiheng},
title = {SongDriver: Real-time Music Accompaniment Generation without Logical Latency nor Exposure Bias},
year = {2022},
isbn = {9781450392037},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
url = {https://doi.org/10.1145/3503161.3548368},
doi = {10.1145/3503161.3548368},
booktitle = {Proceedings of the 30th ACM International Conference on Multimedia},
pages = {1057–1067},
numpages = {11},
keywords = {automatic improvisation, music accompaniment generation},
location = {Lisboa, Portugal},
series = {MM '22}
}