A HMM based system for fine tuning in automatic speech segmentation

Abstract

In this study, a HMM based system for fine tuning the results of automatic speech segmentation results, is proposed. The phonetic boundaries of an automatic segmentation system are used as input to this system. This system includes diphone and diphone class based HMMs. The average absolute boundary error for /y/- /uu/ boundary is decreased by 44% and the average absolute boundary error for /t/-/p/ ve /uu/-/o/ class boundary is decreased by 63% using the proposed fine tuning system.