Intelligent automatic music curator and recommender system

智能自动音乐策展和推荐系统

• 批准号: 516172-2017
• 负责人: Stege, Ulrike
• 金额: $ 0.91万
• 依托单位: University of Victoria
• 依托单位国家: 加拿大
• 项目类别: Engage Plus 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=640440
• 关键词: Intelligent; automatic; music; curator; recommender

Describing music with natural language is not easy. The accuracy of any such description is typicallyproportional to one's level of expertise with the subject. The ability to search for music efficiently usinglanguage is therefore typically reserved for musicians.Apollo Music Store (AMS) specializes in curating and licensing music to fulfill the needs of advertisingagencies and brands around the world. To stay at the top of its field, AMS is experimenting with new ways ofletting its users find and select relevant music efficiently. AMS aims to eliminate the need to translate musicalideas to natural language, and instead allow users to input music tracks into the AMS search engine asexamples of what they are looking for.Building on results obtained during our Engage Grant titled Content-aware Music Recommender System usingEmotion Recognition, where our algorithms reached accuracies of up to 87% on given emotions, we intend topursue our research in designing machine-learning algorithms for the automated recommendation of music,based on the emotional and acoustic similarities to an audio example provided by the user. This projectcombines the quest to understand how to model human emotions with machine learning techniques and musictheory, and thus has far reaching applications in affective computing, cognitive science and artificialintelligence research.

用自然语言描述音乐并不容易。任何此类描述的准确性通常与一个人对该主题的专业知识水平成正比。因此，使用语言高效搜索音乐的能力通常是为音乐家保留的。阿波罗音乐商店（AMS）专门从事音乐策划和许可，以满足世界各地的广告代理商和品牌的需求。为了保持在该领域的领先地位，AMS正在尝试新的方法，让用户有效地找到和选择相关的音乐。AMS的目标是消除将音乐想法翻译成自然语言的需要，而是允许用户将音乐曲目输入到AMS搜索引擎中，作为他们正在寻找的内容的示例。在我们的Engage Grant项目中获得的结果的基础上，我们的算法在给定的情感上达到了高达87%的准确率，该项目名为“使用情感识别的内容感知音乐推荐系统”，我们打算继续我们的研究，设计机器学习算法的自动推荐的音乐，基于情感和声学的相似性，由用户提供的音频示例。该项目结合了对如何用机器学习技术和音乐理论来模拟人类情感的探索，因此在情感计算，认知科学和人工智能研究中具有深远的应用。