noise category definitions, impulse noise, and workflow

[Craig Maier]

Hi,

What happens if you put two Paragraphic pre-emphasis (75 uSec each) in the multifilter and then apply the EZ Impulse Filter? Reversing that would just take two de-emphasis paragraphics in the Multifilter later in the process.

Something worth trying - - -

Craig

[Dan McDonald]

This might sound a little whacky but I have found the spectral filter in spectral difference mode to be very good at making whatever signal is there sound much better. I would get a pristine sounding recording that is similar (by the same artist if possible; if not,then similar instrumentation). Use that as the source and your recording as the target. It will really help make the sound that is there more "alive" and you can adjust the filter so it doesn't emphasize the noise above 5kHz,then use the premphasis.

[Marc Hildebrant]

Dan,

Could you add some details to the method ?

Marc

[Dan McDonald]

Hi Marc -
Sure. I use the forensics version of DC8. I find a recording that is comparable. If these are old Blues records, for example, you can get some very good quality early Chicago blues recordings by Muddy Waters for example - acoustic or electric - to match against the artist jFritz has, if there aren't any high-quality recordings of that artist available. Anyway, the trick is to get a high quality recording that should sound similar to the sound you think was present at the recording you want to restore (similar instruments, etc.).

Then I bring up the spectrum analyzer and put it in 'spectral difference' mode. I use the high-quality recording as a source and take a sample of it. Then I go to the recording I'm trying to restore, and take a sample of it. Then I preview the sample to hear what it sounds like. Also, check the difference signal. Sometimes it's better to just leave a portion of the spectrum alone, and, if so, you can move the bars in and out so that the filter doesn't affect the highs or lows.

Anyway, with something that is as low as 5kHz for the highs, working with tapes, I just bring everything down to -100dB above what I believe is the highest signal I can find. That way I don't increase the noise. I think that if I'm going to declick, though, I would leave the difference signal alone so that the highs are actually emphasized, so I could de-click better. After de-clicking, I'd use a brick wall or low pass filter to get rid of all the excess noise that was generated.

Anyway, this works real well with tapes and records that I've transferred. There is some trick to selecting the right 'source' record. Some things just sound horrible when you think they should be right, and sometimes something just 'clicks' and it really improves the overall sound.

[Craig Maier]

Dan,

Do you mean that you bring up the Spectral Filter not the Spectrum Analyzer?

Craig

[Dan McDonald]

Sorry... yes, the spectral filter. I always mix up those names.

[jfritz]

Thanks for the suggestions. I'll try doubling up on preemphasis. I should mention that the recordings that give me trouble are mostly from the 1920s, particularly the Paramounts. The two Paramount gags that come to mind is "The only thing electrical was the light bulb in the studio." and the thing about a special shellac mix that included sand. I'm fooling around with Blind Lemon Jefferson at the moment. A lot of his surviving records are nearly worn out. There are a few that have good copies available. They don't require much processing. It's interesting how different the vocals sound on the good copies. <g>

[Dan McDonald]

There are some pretty good Blind Lemon on CD though. Are these different songs?

[Marc Hildebrant]

I tried using the "Spectral Difference" filter for some rough 78 RPM Acoustic records...no improvement yet.

Question, what size fft is good ?

Marc

[Dan McDonald]

I bounce between the highest setting and about midway - it sort of depends on the material.

[Craig Maier]

Les Paul and company were using the Spectral Difference Filter to match the quality of optical sound-tracks to concorrently made magnetic recordings of the musical portion of the same shows. These were Llisterene sponsered TV shows from the early 1950s that featured Les Paul and Mary Ford. The optical track contained all the dialog plus the musical content. The magnetic tape only contained the musical portions. The sound quality of the magnetic media was substantially better than the optical. So, the Spectral Difference Filter was used to balance out the transitions between dialog and music to obtain a more realistic transition. The magnetic source was used as the reference to EQ the optical track using said filter. Some decent results ensued from that endeavour, but Les passed away before completing the restoration of all of the shows. I do recall that the process was described by Les and his tech as tricky business finding the right place to sample and determining the correct fft size to use. I think that larger ffts are not always better in this application, but it has been a few years since and my memory is fading on that issue.

Craig

[Dan McDonald]

Yes - I think (but I'm not sure, and I always check out different fft sizes)
That if I get a file that has almost no highs, I use a high fft size, and if I need a cleaner mid-range,I use a lower fft size,but again I have to experiment each time because it sort of interacts with the instrumentation.

[jfritz]

Re: Good Blind Lemon on CD. There are some songs that sound pretty good. Others, like the two takes of "Jack of Diamonds" are absolutely trashed. On the really bad recordings I'm more interested in making the vocals clear enough to understand the lyrics than doing a real restoration.

I tried double preemphasis on the Blind Lemon as well as some trial and error EQ of my own. Didn't have much luck. One thing this thread did was get me to pay more attention to the spectral analysis window. Some 78 reissues on CD have content that goes well past 10K Hz. As predicted by people who actually know something <g>, impulse noise reduction works considerably better on recordings like that.