The effect of speech sample length on interrater and intrarater reliability