It is not clear why do you want to reduce the noise. If you want to reduce the blur on the display (I don't think you can call it blur, but probably you mean a blurry look of the spectrogram as it is shown in Jeff's work), you can reduce number of points on the spectrogram grouping close frequencies together and showing average amplitude for a group as abscissa value. Ask Jeff for more detail on his Article's page, see "Comments and Discussions" at the end of the page.
I don't think you can reduce the noise in the sense your are interested in by using auto-correlation method. The power of this method could help to detect hidden periodic signal (such in a rhythmical base of it) but hardly could clean up a "technical" noise generated by non-linear distortions or other defects of the recording process without strong distortion of the audio. See
http://en.wikipedia.org/wiki/Auto-correlation[
^], pay attention for discreet autocorrelation for signal processing. If you need to estimate effective average pitch of the recorded instrument sound (which is required by digital instrument tuners) or detect musical beats in the uniform-tempo musical samples, it could help; in all other cases, when all such characteristics "flow" at fast pace, you will only mess up the audio.
For other methods of noise control, see
http://en.wikipedia.org/wiki/Noise_reduction[
^],
http://en.wikipedia.org/wiki/Noise_control[
^].
—SA
Submit an article or tip