Pause concepts for audio segmentation at different semantic levels

Abstract

This paper presents work on the determination of temporal audio segmentations at different semantic levels. The segmentation algorithm draws upon the calculation of relative silences or pauses. A perceptual loudness measure is the only feature employed. An adaptive threshold is used for classification into pause and non-pause. The segmentation algorithm that determines perceptually relevant pause intervals for different semantic levels incorporates a minimum duration and a maximum interruption constraint. The influence of the different parameters on the segmentation is examined in experiments and presented in this paper. A new approach for evaluating segmentation accuracies is required. It is shown that the simple perceptual pause concept has a very high relevance when segmenting audio at different semantic levels.