First extract *N* samples of speech starting at and then window:

Now apply the DFT.

The result is continuous, but it can be sampled at *n* points without
loss of information.

This equation defines an algorithm that takes an array of *N* complex
numbers (or equivalently an array of *N* real numbers and *N* imaginary
numbers) and returns an array of *N* complex numbers.

The inverse short-term DFT can be similarly defined:

Speech Vision Robotics group/Tony Robinson