In concatenative speech synthesis the fundamental factor with heavy influence on synthesized speech quality is the database of acoustic units. In case of bases received in automatic way, the key matter is suitable marking the borders of acoustic units. This article describes the algorithm of correction of acoustic units borders appointive in automatic way. It is based on two factors specified and tested here. It also describes worked out method of grade of acoustic units database, which allows to observe the influence of introduced correction on the base quality.