At 12:37 PM 3/25/2003 -0500, aaron levinson wrote:
>I am doing a compilation of "garage salsa" groups of the 60's and 70's.
>I bought a CEDAResque program called Raygun that I  used in a few other
>instances in the past to clean up some of these recordings, which range
>from the truly superb to the fairly atrocious.
>Rather than going back to the same room and pay the going rate I've
>decided to learn it myself. One of the features Raygun offers is
>DC Offset. The tutorial was extremely vague around the details of the
>science that surrounded DC Offset itself.
>The gist was that in the process of the A to D conversion certain
>artifacts can be introduced which degrade the audio quality. Okay,
>fine. But how is it degraded, what suffers, what should I listen for? I
>do not want to put the data-stream through yet another algorithm if it
>is not going to audibly improve the final result.
>Using Occam's Law as both an aesthetic and technical
>maxim has worked for me thus far, can anyone better versed than I shed
>some light on this little-known phenomenon?

My experience is as an amateur, not a professional in the field. Still, it
may be of value.

DC offset is a source of even-harmonic, essentialy second-harmonic
distortion. That is, the positive-going and negative-going signals are not
balanced in the aggregate. (They are inherently not balanced locally unless
you are looking at artificial waveforms.) The other major source of
second-harmonic distortion is bias in microphones or pickups, but that
differs fundamentally in being noticeable only on large amplitudes.

The effect of DC offset is negligible until one edits the waveform since
the ear is not as sensitive to even harmonics as to odd. Operations such as
equalization and dyamic-range adjustment will magnify the distortion even
though they may not alter the offset itself. I find the effect to be an odd
generalized unreality to the sound, not the substantial change of emphasis
generated by odd harmonics.

As a trained scientist, I am frustrated by the vagueness of the above.
However, once I discovered how easy it is to correct for offset, I do it
routinely and need not face what happens if I don't. By the way, note that
the correction must be checked at the start and end of the signal. In many
cases I've seen, a nominally null signal begins at zero, moves quickly to
the offset value, then drops back to null at the end. If the offset is
corrected, the initial and final signals will not be null and will be heard
as clicks.

[log in to unmask]