DBX Encoded Tapes

I have some old cassettes that I recorded using the dbx noise reduction format.  I would be interested in converting them to digital.  How do I replicate the dbx noise reduction using processing in order to make the audio as close as possible to normal sounding?

DBX is pretty much straight 2:1 compression, so what you have to do is 1:2 expansion to get the dynamics back correctly. The only significant problem is that ideally, you need to have some sort of idea of the time constants used, otherwise you will inevitably get some pumping effects.

But because DBX was applied as a constant ratio right across the amplitude range, and wasn't frequency dependent either, you shouldn't have all the difficuly that people have with Dolby NR, because essentially the processing is not signal level dependent, and Dolby, of course, is.

So really, it's down to doing a few experiments with the Dynamics Processor based on increasing the slope of the line from top to bottom in a 2:1 ratio, which will have most perceived effect on the quieter passages, obviously.

Well...

You can get close with 2:1 expansion, but there are a few little wrinkles.

1) You should use RMS level detection - this seems easy enough.
   The AA dynamics processor should do the job.  Attack and Decay
   settings also matter and I do not remember these.  A first guess
   would be 10mS attack and 100mS decay (I remember it was fairly
   quick).  Also the level sensing channel should be band limited.
   The LF and HF rolloffs were different between the two types of DBX.
   For cassette (I think) it was always type II, which sensed from about
   100Hz to 10Khz for the level sensing channel.  Here we have a problem,
   because the AA dynamics processor does not allow separate bandwidth
   settings for the sense channel versus the modulated channel like dbx
   had.  Maybe full bandwidth will be close enough.


2) There was a pre-emphasis added to the encoded signal which varied
    by which type of DBX was used (Type I or Type II).  I think this
    emphasis was added after the compression so you will need to de-
    emphasize before expansion.  Again - guessing a little (again), I think
    that  a two octave -6db dip centered at 10Khz should come close.

So first take the signal and run it through the de-emphasis filter and then
expand it 2:1 with the dynamics processor set to 10mS and 100mS and
you should be close.

There was a period of time when I used DBX Type II for just about everything, and Kihoalu's advice sounds very good.

However, any decoding experiments I've ever done in software have never produced an acceptable result.  If an investment is worthwhile to you, I'd recommend purchasing an outboard decoder on Ebay - a quick search with "DBX Type II" for keywords turned up quite a few for a reasonable amount of money.  

By the way, some years ago I am certain that we had a very helpful discussion about this very topic on the Syntrillium forum.  But I can't find it anywhere.  However, in my graphics archive I found the attached grungy image that gives you a very accurate idea of the de-emphasis curve, and that the curve is constant regardless of amplitude.  What isn't constant is the attack and delay timings, which seemed to be amplitude-dependent, which is another reason why decoding in Audition is hosed from the get-go.

Also, maybe if you did some digging, you probably could find an old DBX box somewhere, and buy it. That would solve the decoding problem perfectly.

I thought I'd drag this up again because I'm now in a position where I would like to try to decode a dbx encoded tape but I don't have a proper decoder and there are none on Ebay at the moment.

I've read the thread in the Syntrillium archives which gives the de-emphasis curves and they seem to sound about right but the one problem that I am having is getting the time constants correct. I still get the occasional dropout when the signal goes quiet. I've set the attack to something like 3mS and the release to 0.8 seconds. Should I set the release to something longer?

Cheers

James.

James, using your ears is the only acceptable solution, seeing as DBX was a patented process with secretively-implemented detection functions.  It would seem that the attack and release times were amplitude dependant.  If your recording has a really wide dynamic range, with a mix of percussive and gentle attacks, decoding DBX could prove to be a nightmare.

Don't cancel that Ebay search.