Audio-track classification using deep neural networks

贝尔媒体公司是贝尔加拿大公司的大众媒体子公司。该公司面临的一个重大问题是缺乏有效的音轨识别方案。贝尔数字媒体的音轨可以是立体声、环绕立体声、单声道或语音(描述视频)。尽管有使用元数据标记音频频道的标准，但许多媒体制作设施(包括Bell)并不遵守这些标准。事实上，在许多情况下，没有元数据标签，或者没有正确地标记曲目。目前，该公司使用人工方法对未标记的音轨进行分类，效率低下，而且耗时。因此，本项目的目标是开发一种新的基于深度神经网络的音频轨道识别方法，以便贝尔公司能够有效和高效地在其音频轨道上使用。标记音轨将允许Bell提高其工作流程中的自动化程度，从而提高效率和准确性，同时改进其整体产品开发。*

Bell Media Inc. is the mass media subsidiary of Bell Canada Enterprises. A significant problem faced by the company is the absence of an efficient audio-track identification scheme. The audio-tracks of Bell digital media could be stereo, surround sound packages, mono or speech (described video). Although there are standards for tagging audio channels with metadata, many media production facilities (including Bell) do not adhere to these standards. In fact, in many cases there is no metadata tags, or the tracks are not labeled properly. Currently the company uses a manual approach to classify unlabeled audio tracks, which is inefficient, and time consuming. The objective of this project therefore, is to develop a new deep neural network based audio track identification that Bell can use on its audio tracks effectively and efficiently. Labeling the audio tracks will allow Bell to increase the amount of automation in their workflow, thereby increasing efficiency and accuracy, while at the same time improving their overall product development.****