Time-Domain Segmentation and Labelling of Speech with Fuzzy-Logic Post-Correction Rules