Visualization of Summaries

One of the main goals of the AMI project is the automatic generation of multi-modal meeting summaries. Apart from their generation, there are many ways of presenting these summaries to the user. A very appealing presentation style for extractive summaries has been realized by the SuVi-tool (Summary Visualizer). Based on actual audio and video data it generates a story-board layout to present the summary which aims to represent the chronological sequence of a meeting in a well-arranged style.

Appropriate images are taken from the video stream, where available, and the speech bubbles are extracted from the transcript according to importance. The topics of the meeting are shown in the order they were discussed in the meeting with important images, e.g., showing the main contributors and an excerpt of their contributions (see Figure 1).

SuVi uses a generic storyboard layout engine that combines constraint-solving and optimization techniques for an optimal presentation. Despite the static documents that have been generated in an earlier version of SuVi, we extended the system to generate proper SVG format graphics, fully hyperlinked to the transcript and video of the meeting.

To demonstrate the robustness and flexibility of the system, we applied our system to real-world data provided by Synthetron (see Figure 2).

SuVi IS1003a SuVi Synthetron
Figure 1: An example output of SuVi, applied to the meeting IS1003a from the AMI Meeting Corpus. Figure 2: The output of SuVi applied to the real-world data from Synthetron

List of relevant Publications

Contact Persons: