Figure 4—Every 110 ms, the square-wave input is sampled and several options are considered, depending on the state of the frame, zero-crossing, silence, and sound counts. The state machine effectively captures the input utterance, while rejecting short bursts and input errors due to excessive background noise.