Spectral processing refers to sound processing where sound is manipulated in the spectral domain rather than in the time domain. Typically, processes are applied to a sequence of FFT frames that are then converted back to the time domain using IFFTs and overlapped adds. The spectrum computed by the FFT is typically represented by an array of floating point numbers, and it is important to understand what each element of the array represents. To minimize artifacts due to the use of multiple FFT frames in sequence, FFT processing is usually performed on overlapping windows. Using raised cosine, or Hann, windowing before taking FFTs and again before overlapped addition of frames after the IFFT, perfect reconstruction can be achieved if the step size is 25% of the frame size (with 75% overlap). In Nyquist, spectral processing is assisted by special objects that iterate over a sound input, constructing a sequence of spectral frames. Another special function can convert a series of spectral frames back into a value of type Sound. The sequence of spectral frames can be processed by a user-defined function to alter the spectrum in any way.