Word level precision of the NALIGN automatic segmen- tation algorithm

Abstract

This work presents an evaluation of the word level precision of an automatic segmentation algorithm, NALIGN. Measurements of the portion of temporal overlap between automatic and manual word level segmentations show that a precision of 90 to 95% can be achieved, partly depending on the type of speech material. These precision figures are furthermore only marginally lower than the gold standard obtained comparing independent manual segmentations.