Feature Learning of Critical Live Performance Audio Characteristics for a Virtual Sound Engineer

现场音乐表演通常涉及调整控制台的几个参数，以优化观众的声音质量。这通常是由音响工程师或音响技术人员通过在演出前的声音检查中调整混音控制台的控制来完成的。许多音乐家或小场地的租户将受益于自动生成的表演指导声音调整。虽然有一般的规则可以遵循，但大多数工作主要是基于声音工程师的直觉和经验，并且不容易转化为计算机算法。这个项目的产业合作伙伴，高档科技公司，就致力于开发这样一个智能自动虚拟音响师。实现这一目标的第一步是确定在现场音乐表演中被认为是“好声音”的声音特征。在这个项目中，将开发特征学习技术来包含需要优化的声音特征，以提供令人满意的音频体验。

Live music performance typically involve the adjustment several parameters on a console to optimize the sound quality for the audience. This is usually down by a sound engineer or sound technician, by adjusting controls on a mixing console during a sound check prior to the performance.Many musicians or tenants of smaller venues would benefit from automatically-generated guidance for performance the sound adjustment. While there are general rules to follow, most of that work is primarily based on the sound engineer intuition and experience, and do not translate easily into a computer algorithm. The industrial partner in this project, Upscale Technology Inc., aims at developing such an intelligent automatic virtual sound engineer. One of the first step towards that goal is to identify the sound characteristics that are common to what is considered a "good sound" in live music performances. In this project, feature learning techniques will be developed to encompass the sound charactistics that need to be optimized to provide a satisfying audio experience.