Brand usage detection via Audio Streams

Abstract

With the constant creation of content on digital media platforms, it is becoming ever more difficult for companies to control the use of their brand. Such use can be found in a review of a competing product, feedback on a news article or merely part of a discussion in a public forum. The reference to the target brand can take the form of a direct textual mention, use of the corporate logo in an image form, or the mentioning of the brand in an audio format. Detection of textual and visual use of brand identity is well researched and considered in an advanced state, yet detection in audio formats is more challenging due to factors such as background noise, voice morphing and heavy accents to name a few. In this initial research, we seek to establish an ideal dataset for further research in this area and an initial review of current solutions for the detection of brand usage in audio streams. Our findings suggest that current solutions do not cater for heavy accents well, which will be the focus of the next stage of this research.