Audio Data Indexing: Use of Second-Order Statistics for Speaker-Based Segmentation