The offset should cover only the silence at the beginning of each sample while the fixed part (and the cutoff) should be set to a length/position that allows the samples to be stretched by the synthesis engine without quality issues. Pre-utterance goes at the transition between the consonant and the vowel and overlap is set at a value lower than the pre-utterance, although the correct value for overlap may be dependent on the type of consonant. Unvoiced "hard" consonants like "k" or "t", for example, may require negative overlap values.