A Low-Cost Virtual 2D Spokes-Character Advertising Framework

Abstract

Live-streaming advertising has achieved huge success in modern retail platforms. However, small-scaled merchants are neither economically nor technically capable of having their own spokes-person. Addressing the need for the massive online interactive advertising, this paper proposes an economic-efficient approach, Virtual spokes-Character Advertising (VSCA). VSCA generates 2-D Virtual spokes-Character advertising video and provides it to the merchants as a supplementary marketing method. VSCA first generates the simplified natural language description of the merchandise from its original long title using text generation methods and then passes it to the Text-to-Speech model for the audio description. Secondly, VSCA remits the audio to our remodeled two-phases lip-syncing network to generate virtual advertising videos about the given merchandise. With our novelly designed two-phases lip-syncing network, it is the first in the industry able to generate lip-syncing video of given audio with human face image input instead of video input. As the industry’s first application on 2D spokes-character advertising, VSCA has its large potential in real world applications.