Learning the Structure of Music

该项目旨在开发将新的概率机器学习技术应用于音乐分析的模型和工具。该项目的基本主题是学习将同一音乐作品中同时产生的不同数据联系起来的模式。数据的来源如下：a)乐谱(MIDI格式)，b)音频(作品的录音)，c)WORM数据(图表演奏信息)，d)EEG数据(听音乐的对象)和d)fMRI数据(听音乐的对象)。我们将寻找的链接模式涉及如下的数据流对：a)音乐乐谱与WORM数据，b)音乐乐谱与fMRI数据，以及c)音频与EEG数据。第一对将被用来识别特定表演者的典型表演模式。第二对和第三对将用于识别特定音乐模式(如旋律序列、音乐短语、和声进度等)在大脑中的影响。该项目将加深我们对音乐结构与表演和体验之间关系的理解。这种发展的潜力是相当广泛的，潜在的应用于音乐治疗和娱乐。例如，它将有助于开发在演奏者可能从未演奏过的曲子上模仿演奏者的风格的人工音乐演奏系统，以及为实现对听众的特定效果(或情绪)而量身定做的音乐创作系统。

This project is aimed at the development of models and tools for the application of novel probabilistic machine learning techniques to the analysis of music. The underlying theme of the project is the learning of patterns linking different data arising simultaneously from the same piece of music. The sources of data will be as follows: a) musical scores (MIDI format), b) audio (recordings of the pieces), c) worm data (charting performance information), d) EEG data (of subjects listening to the music) and d) fMRI data (of subjects listening to the music).The linking patterns that we will be seeking involve pairs of data streams as follows: a) musical scores with worm data, b) musical scores with fMRI data and c) audio with EEG data. The first pair will be used to identify typical performance patterns of particular performers. The second and the third pairs will be used to identify the effects in the brain of particular musical patterns (such as melodic sequences, musical phrasings, harmonic progressions, etc.).The project will advance our understanding of the relationship between musical structure and performance and experience. The potential of such developments is quite wide ranging, with potential application in music therapy and entertainment. For example, it will contribute to the development of systems for artificial performance of music imitating the style of a performer on pieces that he or she may have never played before and systems for musical composition tailored to achieve specific effects (or moods) on the listener.