I think that Doug's approach, using multi-channel A-D conversion is 
preferable to sequential passes. An item not discussed (as far as I've 
seen) in this thread is the sample-clock-to-audio phase lock. If you 
look at sampling theory, there can be a very large difference in 
individual numbers while still maintaining the same information. I ran 
into this doing a second pass on a tape and while zoomed out the 
waveforms looked the same (I was looking to cut into and out of a 
damaged section I had been able to get a better transfer of), once I 
zoomed in to the sample level, the individual samples were very 
difficult to match and it was hard to line up sample-accurate cut 
points. The way to do it is to look for relative energy under the curve 
as you are aligning.

All this suggests that a method of starting and maintaining the timing 
accuracy among the multiple copies AND the A-D sample clock (to the 
tolerance discussed below) seems to be a good idea.

The auto image alignment that is being done can be more or less 
successful, depending on the raw input. I'm impressed my hand-held cell 
phone camera can do as good HDR images as it does, but they are by no 
means as good as a tripod-based PhotoShop version.

I'm certain that software will progress to do this. Michael Gerzon 
alluded to some possibilities of track-splitting on tape to help improve 
capture in his 1976 article posted here:

As we discuss the new economy, I wonder how this work will be funded.



On 2012-06-22 10:30 AM, Doug Pomeroy wrote:
>>> Date:    Wed, 20 Jun 2012 10:44:42 +0100
>>> From:    Ted Kendall <[log in to unmask]>
>>> Subject: Re: audio from pictures
>>> <<SNIP>>
>>> Chris Hicks of Cedar did his doctorate on just this question of 
>>> multiple
>>> copies. When the algorithm could be persuaded to work, the results were
>>> much as you would expect - enhancement of the correlated wanted signal
>>> and reduction of the random noise. Unfortunately, pulling the 
>>> recordings
>>> into good enough sync for artifacts to be negligible was a frustrating
>>> and difficult business, even with recordings made one after the other
>>> (by me, as it happens) on the same kit on the same settings on the same
>>> afternoon, with meticulous centring. 
> I gave a talk on this subject at the ARSC conference in Nashville in 
> 1997.
> Two perfectly synchronized copies would provide a theoretical 
> signal-to-noise
> improvement of 3 dB, four copies an improvement of 6 dB and eight 
> copies an
> improvement of 9 dB. The first problem is finding so many copies in E 
> condition.
> But the bigger problem is how to achieve perfect synchronization.
> John S Allen of the Boston Audio Society discusses this matter in an 
> article he
> wrote for the Spring 1990 issue of the ARSC Journal.  He says the 
> accuracy of
> sync must be "about ten microseconds (72 degrees of phase shift at 20 
> kHz)".
> No real-world turntable is stable enough to achieve this result on 
> successive
> plays, and I proposed a stack of tables all powered by a single motor for
> playing the discs simultaneously, with use of digital delay lines to 
> align
> the audio from them after transferring.

Richard L. Hess                   email: [log in to unmask]
Aurora, Ontario, Canada           (905) 713 6733     1-877-TAPE-FIX
Quality tape transfers -- even from hard-to-play tapes.