Using an endpoint detection algorithm, the speech is selected from the two input waveforms. Then, their short-time magnitudes are determined. The short-time magnitude characterizes the envelope of a speech signal by lowpass filtering it with a rectangular window. The magnitude function follows these steps:

• The bounds of the signal are determined and each end is zero-padded.
• The signal is convolved with a rectangular window. As the window is swept across the signal, the magnitude of the signal contained within it is summed and plotted at the midpoint of the window's location. This function, avgmag.m, gives the envelope of the speech.

One magnitude plot is discrete time warped onto the other. The dot product of the two waveforms is computed, and this number is divided by the product of the signals' norms as mandated by the Cauchy-Schwarz Inequality. This calculation results in a percentage of how similar one signal is to the other.