[78-L] Removal of hiss on 78s

[DAVID BURNHAM]

The problem I see with your idea is the main problem with all noise 
reduction systems: how do you distinguish between signal and noise? 
In your proposal, when the system is interpreting a noisy section of 
groove, how is it to tell which undulations to trace and which not 
to? It's not enough to say that you would use the shape of the 
original cutting stylus - when the system comes to a lump or pit on 
the groove wall, it has to decide whether to trace this as if it were 
signal or skip past it as if it were noise. Imagine, for example, 
that the system encounters a widening in the groove that *could* 
indicate impulse noise. Which side of the groove should it trace? 
There could be a flaw on either side of the groove, and if you pick 
the wrong side you're eliminating signal instead of noise.



                        -        Jeff Lichtman

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

You still seem to be thinking dynamically, that is, that the record is being played, (perhaps a system could be developed which could do all the scanning and calculating so fast that the process could be done in real time but that's not where I'm going right now).  Think of digital audio from a CD.  Each sample carries no information whatsoever, it just represents a single binary number;  it is critical that the number be accurate, but it could be part of a Bruckner Symphony, a Michael Jackson hit, a Benny Goodman solo or a speech by Sherlock Holmes, or it may be an element of error correction, there is no evidence, no matter how closely you examine it, of what information it is part of, only when you string many samples together does it become a signal.  Similarly, in the system I'm describing, picture that there are 50,000 static scans per revolution.  Each scan carries no information about the content of the recording, just as a stylus sitting on
 a non rotating record offers no information except the location of the groove at the point of contact.  Each scan is only identifying what parts of the groove wall at that single point on the record surface correspond to the shape of the reference sample, (either the shape of the cutting stylus, if known, or an estimation taken from the record itself).  If there is any wear, dirt or groove damage it will cause the scan to deviate from the reference sample and that part of the scan will be discarded.  In other noise reduction systems, the equipment has to constantly make decisions about whether it is picking up noise or a cymbal crash.  The system I'm describing could accurately even clean up a RECORDING OF SURFACE NOISE!  It will often happen that the entire groove is damaged and there are no valid points on the scan, in which case the signal will have to be reconstructed through interpolation.  But any record in even average condition should
 have enough information that the audio can be lifted cleanly off the surface and be heard noise free.

As dl mentioned, this system could also reconstruct broken records or records damaged beyond any possiblility of being played, although it may take a couple of hours to work on a 2 minute piece.

dl