Audio-Visual TV Broadcast Signal Segmentation

Abstract

Research in the field of audio-visual broadcast programs transcription and indexing has been solved for more than 20 years. Great progress has been made mainly in the area of broadcast transcription from audio signal. In the last 10 years, this research has become more intense, mainly due to the use of deep or convolutional neural networks and because of large IT companies (Google, Microsoft, IBM, Amazon) that can rely on a large number of custom embedded multimedia databases. Very important part of system for audio-visual broadcast signal transcription is subsystem for signal segmentation. Signal segmentation is usually solved separately for audio and visual signal. In this paper, a methodology for audio-visual broadcast signal segmentation is presented and described. The result from audio signal segmentation is used for improving of visual signal segmentation.