Music-visual feature mapping and deep tagging

音乐视觉特征映射和深度标记

• 批准号: 521282-2017
• 负责人: Tzanetakis, George
• 金额: $ 1.82万
• 依托单位: University of Victoria
• 依托单位国家: 加拿大
• 项目类别: Engage Grants Program
• 财政年份: 2017
• 资助国家: 加拿大
• 起止时间: 2017-01-01 至 2018-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=640433
• 关键词: Music; visual; feature; mapping; deep

Apollo Music Store is a Montreal based company that specializes on licensing and distributing digitally music.They represented hundreds of artists and more than 75% of them are Canadian. The goal of this project is toimprove the user experience of Apollo customers who are trying to find specific kinds of music for advertising,games and other applications. Two main objectives will be pursued. The first is to explore how toautomatically associate and adjust stock video with music tracks based on audio and visual feature extractionand mapping. The second objective is to use deep learning techniques to build software that automatically tagsthe music with descriptive keywords. Deep learning has been shown to be very effective in a variety ofmachine learning tasks. Data augmentation and semi-supervised learning are techniques used to train deepneural networks when the amount of labeled data is limited. Apollo Music will provide access to their digitalcatalog as well as existing meta-data their curators have provided about the tracks. The research group ofDr.Tzanetakis has a strong international presence in Music Information Retrieval and Machine Learning.which are research areas that will contribute ideas, concepts, and algorithms to the project. Two PhD studentswith complementary experience in computer vision/deep learning/audio signal processing and gesturerecognition/human-computer interaction will be funded through the project and work with the designers andengineers at Apollo Music Store.

阿波罗音乐商店是一家总部位于蒙特利尔的公司，专门从事数字音乐的授权和分销。他们代表了数百名艺术家，其中75%以上是加拿大人。这个项目的目标是改善阿波罗用户的用户体验，他们试图为广告、游戏和其他应用寻找特定类型的音乐。将追求两个主要目标。首先是探索如何基于音视频特征提取和映射来自动关联和调整库存视频与音乐曲目。第二个目标是使用深度学习技术构建软件，用描述性关键词自动标记音乐。深度学习已被证明在各种机器学习任务中非常有效。数据增强和半监督学习是在标记数据量有限的情况下用于训练深度神经网络的技术。Apollo Music将提供对他们的数字目录的访问，以及他们的策展人提供的有关曲目的现有元数据。博士的研究小组。Tzanetakis在音乐信息检索和机器学习方面拥有强大的国际影响力。这些研究领域将为项目提供想法、概念和算法。两名在计算机视觉/深度学习/音频信号处理和手势识别/人机交互方面具有互补经验的博士生将通过该项目获得资助，并与阿波罗音乐商店的设计师和工程师一起工作。