Automatic segmentation and alignment of continuous speech based on temporal decomposition model.