An algorithm which can extract precise GCI information directly from speech is described. By utilising the average pitch and the area informations, accurate GCI positions are obtained. The result compared with the manually classified GCI information using EGG signals only shows 0.2ms average error.