Silence detection is typically based on the RMS (Root Mean Square) energy value of an audio signal; when the RMS of an audio segment is below a preset threshold and its duration exceeds the minimum silence length, that segment is marked as a silence candidate.