Speaker-based segmentation for audio data indexing