Divide and Remaster (DnR) dataset v3
收藏资源简介:
Divide and Remaster (DnR) 数据集 v3 是由Netflix公司和乔治亚理工学院联合开发的电影音频源分离数据集,旨在解决多语言支持下的音频源分离问题。该数据集包含超过30种语言的对话内容,涵盖多个语系,如日耳曼语系、罗曼语系等。数据集包括训练、验证和测试三个部分,每个部分包含数千个音频片段,每个片段包含对话、音乐和效果等音频源。数据集的创建过程中,特别注意了语言多样性、音频质量及版权问题,确保了数据集的广泛适用性和合法性。该数据集主要应用于电影和流媒体服务的音频处理技术,以提高音频分离技术的准确性和通用性。
The Divide and Remaster (DnR) Dataset v3 is a movie audio source separation dataset jointly developed by Netflix and the Georgia Institute of Technology, aiming to address audio source separation challenges in multilingual scenarios. This dataset contains dialogue content in over 30 languages, spanning multiple language families such as the Germanic and Romance language families. The dataset is divided into three subsets: training, validation, and test sets, each holding thousands of audio clips. Each clip comprises multiple audio sources including dialogue, music, and sound effects. During the dataset's creation, special emphasis was placed on language diversity, audio quality, and copyright compliance, ensuring its wide applicability and legal validity. This dataset is primarily utilized for audio processing technologies in films and streaming services to enhance the accuracy and generalizability of audio source separation techniques.
数据集详情
Bandit: Cinematic Audio Source Separation
- 模型来源: 改编自 Bandsplit RNN
- 相关论文: IEEE OJSP Open-Access Paper
- 模型仓库: Model Repository
Banquet: Query-based Music Source Separation
- 模型来源: 改编自 Bandit + PaSST
- 相关论文: Accepted ISMIR 2024 Preprint
- 模型仓库: Model Repository
Divide and Remaster v3 (WIP)
- 数据集来源: 多语言重制版的 Divide and Remaster v2
- 相关论文: Preprint, submitted to IEEE IS2
- 模型仓库: Model Repository
- 数据集仓库: Dataset Repository (WIP)

- 1Remastering Divide and Remaster: A Cinematic Audio Source Separation Dataset with Multilingual Support音频算法,Netflix公司,洛斯加托斯,CA 95032,美国 · 2024年



