Stay in touch…

Blog

Read the latest Bitstream

RSS Feed

LinkedIn

Look for us at LinkedIn

Twitter

Follow us on Twitter

Electronic Musician

This article appeared in the July 2003 issue of Electronic Musician, and was written as a companion to a Windows restoration product survey provided by my friend, Gary Hall.

Audio Restoration On A Budget

Let’s look at the process of restoration, which always begins with the best possible transfer. As with most things in life, audio restoration conforms to the DIDO philosophy: DooDoo In – DooDoo Out. For you technically challenged folks out there, this mean that the more time you spend on getting a high fidelity transfer from the original medium through analog to digital audio, the better the quality will be in the end.

Since most restoration jobs start with an analog source, such a phonograph record, analog tape or optical film track, time should be taken to clean the original thoroughly, without removing the signal or worsening the noise. If the material is worth your time and effort, you may want to hire a specialist to do the transfer. If, however, your the type to live life to the fullest or you’re financially challenged, go ahead and do the transfer yourself. Parker Dinkins, President of MasterDigital in New Orleans, stresses… “how important the initial transfer is, i.e. stylus selection, disk cleaning, azimuth adjustment - because everything following the transfer will be affected by those initial efforts. You wouldn’t record Pavarotti with a crappy crystal lapel mic, but it’s amazing how many people fail to understand the significance of matching styli or adjusting azimuth during the initial transfer.” MasterDigital typically does over 120 restoration jobs each year and, when I talked to Parker, he had just spent almost an hour declicking less than 90 seconds of material. “I wanted it to be perfect for my client,” he said, “and my expensive, real–time hardware couldn’t touch it, so it was SonicStudio’s Type A, B, C and E — wash, rinse, repeat…you know the drill!”

If the original is on vinyl, and the budget warrants it, find a record cleaning machine in your area. These rather expensive devices, from SOTA, Moth, Keith Monks and others, mostly look like oversized turntables but, in place of a tone arm and pickup, there’s a squirt gun–like cleaning fluid applicator, scrubbing brush and a vacuum inlet to mop up the dirty leftovers.

Nitty Gritty NGM1
Nitty Gritty’s NGM1, about US$140

VPI HW165
VPI’s HW165, about US$500

Loricraft PCR3
Loricraft’s PCR3, about US$1750

For those budget jobs, use the manual approach; the good ol’ velvet pile record cleaning brush and some cleaning fluid. For cleaning fluid, a tiny amount of mild detergent and distilled water works wonders. Just be sure to rinse well with more distilled water and dry well. For seemingly terminal cases, some folks are successful playing a cleaned disk wetted with distilled water or Armor All. Be careful, though…If the record is an odd size, color or weight or smells differently from an average record (I’m serious!), you may have an acetate or transcription recording and should proceed with caution. These one–off shellac disks or “lacquers” have a record surface composed of a very soft compound and cannot tolerate even slight abuse or harsh cleaners. Another thing, remember that vinyl is an elastic material. Each time you play a vinyl record, you deform the groove walls but they return to their original shape if given the chance. If you repeatedly play it within a 24 hour period, the plastic is smeared and doesn’t have an opportunity to “relax” or flow back to it’s original shape, resulting in permanent damage.

Once the record surface is clean, the next task is to find a stylus that provides the best signal while reducing any surface noise. Styli shapes fall into 3 general categories, conical, elliptical and line contact. Each type will contact a different portion of the groove’s side wall and produce a different sound. The tracking force and anti–skate adjustment will also have an impact on the transcription quality. Again, if you have the budget, you may want to experiment or, just hire someone who has the equipment and expertise to do your audio justice.

For analog tape, very gently thread up the reel, after aligning the repro chain, of course. [see sidebar below – Hi, My Name’s MeRL] Run the tape at slow speed, bypassing the heads, guides and idlers if possible, and check for splices, tape shed or adhesion problems. Repair any sticky or broken splices and, if the tape is shedding, you should remove the reel right away and bake the tape to dehydrate and restore the binder before attempting a transfer. If your tape is a brown rather than grey or black in color, look out for shedding. Your tape may need baking to drive out excess moisture, so go low and slow. [see sidebar below – EasyBake Oven]

Wow, the hard part’s over so now we can go on to the digital restoration process. I always think of this as a five step process: analyze the material to determine the impairments, reduce fixed frequency noise, reduce impulse noise, reduce distortion and finally, reduce broadband noise. Let’s take each one in turn…

You can’t fix what you don’t know, so restoration starts with investigation. This usually means critical listening, logging and taking notes, then performing various measurements of frequency versus amplitude over time. All this serves to tell you what you’re up against when going to the next step, EQ. Analysis tools range in sophistication from, at the low end, an almost useless real or non–real time analyzer (RTA) and rockin’ good freeware such as Tom Erbe’s SoundHack to comprehensive measurement packages like Metric Halo’s SpectraFoo and the more sophisticated tools included with most restoration packages.

You see, most electrical recording contain some small signature of the AC power supply from which the gear was powered. This usually appears as 50 or 60 cycle hum or a low order harmonic such as 100 or 120 Hz. By measuring the exact frequency of the hum, you can determine the amount by which the speed of the recording is off. Remember that old analog gear relied on lots of mechanical stuff to make it all go and speed accuracy often suffered as a result. But, the AC or “mains” power coming out of the wall outlet is always spot on, frequency wise. So, all it takes is some tasteful sample rate conversion to nudge the recording back on to the correct speed.

Fixed frequency noise is just that, unwanted signal that is time invariant or close to it. Examples of this would be DC offset from crappy converters, hum and buzz from ground loops, the 15.75 kHz horizontal subcarrier from a TV (see figure 1 below) or acoustical noises from air handlers, old transformers or motors. The solution is EQ, good old fashioned filtering to reduce or eliminate the problem. There are as many tools available to attack this problem as there are vendors out there selling restoration products. Some are easy, quick and dirty while other are cumbersome but effective. All share the same approach: apply one or more filters to act on the unwanted signal. These tools are arguably the most easily abused as too radical a filter setting, usually a too narrow bandwidth or high Q value, will surely destroy any hope of a hifi result.

Sonogram
Figure 1 – Static (bottom) and dynamic (top) views of frequency vs amplitude

Some times it’s nice to toss out some range of frequencies but all you have is 1st or 2nd order filters, with 6 or 12 dB per octave slope respectively. If that’s the case, no problem — just cascade what you’ve got. Kellie Ware, President of Elemental Audio Systems suggests “…using several high or low pass filters together to get a fast roll-off or steeper slope.” In figure 2 below, you can see a screen shot of Elemental Audio’s Equim, where the white curve represents the selected 16 Hz high pass filter. The steep black curve is the result of adding three more high pass filters, for about 24 dB of cut at 30 Hz. You can also see the result of Equim’s Harmonic filters, which make short work of hum and buzz issues.

Equim
Figure 2 – Equim’s cascaded high pass & comb filters

Beware of notch filter for, by such things, you will be sad. Notch filter exhibit “infinite” cut, or at least as close to that as you get in the real world. As a result, they induce a ridiculous amount of group delay since group delay in a IIR filter is proportional to the severity of the setting. [see sidebar below – IIR vs FIR] A combination of analysis and critical listening will yield acceptable results for the next step, impulse noise reduction.

By the way, I’ve always found it useful on mechanical recordings to carefully apply a high pass filter. This removes three common problems: DC offset [see sidebar below – One Isn’t The Loneliest Number], low frequency “clunks” resulting from large stylus excursions, and rumble and other subsonic crap that can sneak into a recording The best way to do this is to listen for what you want to remove, not what you want to keep. So, starting with a low pass filter set to maximum cut, adjust the order (slope) and resonant frequency so you hear only low frequency garbage and no signal. Then, remember the settings and dial in a high pass filter with the same order and resonant frequency. Drop it in and out of circuit as you listen, just to make sure it doesn’t adversely impact the low end.

A word about minimalism is in order here. Fight the urge to EQ out high frequency noise in order to reduce hiss. This will totally mess up your ability to reduce impulse noise, the next step in the process. Throughout the restoration process, check your work against the original to make sure you’re not cleaning too aggressively.

On to impulse noise, those pops and clicks that inform us of the vinyl roots of a recording. Pops and clicks also pop up, so to speak, in the digital world when someone’s not paying attention to proper clocking procedures, batteries become low in a portable recorder or DC offset causes thunks at an edit point. Brief duration acoustic events, such as coughs and floor creaks, can sometimes be treated with impulse noise reduction tools as well. In general, these product use frequency analysis of the time just before and after an unwanted event, what Sonic Studio calls the “wings,” to cover up the problem. They range in sophistication from simple interpolators, that “stretch & blend” good samples over the offending region, to more clever versions that resynthesize novel samples to seamlessly match the wings. Earlier, I mentioned stereo transcription of a monaural source. A very simple “declicker,” that works with stereo playback of mono phonograph records, simply picks the better of two channels in real time. This assumes that pops and click do not often occur in both channels simultaneously.

Anyway, declicking is best done in stages. That is, the longest duration clicks are first removed, then sorter ones, then the smallest clicks perceptible as individual events. Andy Smith, Sales & Support Specialist for DARTech, says “…it's often a good idea to use multiple passes to get the most from your click removal tool. You’ll discover that finding a good “one size fits all” setting for click/pop removal will often not produce the same results as ‘focusing’ your click removal tool on specific kinds of clicks. For that reason, we recommend that you use at least two passes with any click removal tool.”

In the world of digital audio, there waits an especially insidious signal degradation and that’s the dreaded clip. Unlike analog, digital reveals it’s binary nature ’cause when you exceed the available dynamic range, you’re hosed. Jason Levine, Music Director for Syntrillium Software, waxes historical…“Since the beginnings of digital recording, the main problem that engineers face when recording is that the occasional rogue peak slips through, whether a exuberant plosive or an atypical snare hit, causing the signal to clip. In the past, there was no way to remedy this problem. But with Cool Edit Pro2’s Clip Restoration tool, it’s a no brainer…This is probably the most over-looked feature inside CEP, but it’s also one of the most powerful.”

When reducing impulse noise, beware of the pencil. Many digital editors have a “pencil” tool that allows you to “redraw” a waveform and, seemingly erase a pop or click. This usually isn’t a good choice as the pencil is operating in the amplitude domain while repairs should be performed in the frequency domain. In fact, the miraculous tools that make this restoration stuff possible all work in the frequency domain. Time domain data, like a WAV or AIFF file, is translated into frequency domain data using a mathematical slight of hand called a Fast Fourier Transform. The processing or repair is performed in the frequency domain and an inverse transform then returns the data to the time domain.

Once the pops, click, ticks, and zipper noises are taken care of, you may be able to reduce distortion. There are many sources of distortion but, usually, it’s the result of either of two mechanisms. One is poor gain staging and subsequent overload such as in a mic preamp or onto tape. The other is groove wall damage and surface noise on a phonograph record. Though only high end software can perform the seeming miracle of selective distortion reduction, some tool suites also have decrackling functions as well. In figure 3 below, notice that the Declick tool has a slider to adjust between CRACKLE and CLICK. This biases the detection algorithm to “pay attention” more to surface noise, the crackle, or discreet clicks. Also notice, at the bottom of figure 3 below, the patching diagram or signal chain. A sonogram first lets me visually identify problems in both time and frequency, then a filter provides a high pass filter to reduce subsonics. Next, a declicker reduces impulse and surface noise and, finally, a broadband denoiser delivers the coup de gras. By the way, in the TC Declick window of figure 3, the red parts of the waveform represent repaired samples…

Spark XL
Figure 3 – Spark XL’s impulse (TC Declick, left) & broadband noise reducers (TC Denoise, right)

Most impulse noise reducers provide a mode that lets you listen to the material being removed. Christoph M. Musialik, President at Algorithmix, suggests that “…for the best results, use your own ears in combination with the difference feature. Switch between the original input signal and the input/output difference, i.e., the part of the signal taken out by the de-scratching algorithms. This difference signal normally should not contain any parts of the original signal you want to preserve.” Christoph should know since, besides creating Sound Laundry, they’re also the developer of Waves’ Restoration package and all restoration plug-ins for Merging Technologies’ Pyramix to boot.

That final tool in my audio restorer’s arsenal, and probably the one most familiar to the average engineer, is a broadband noise reduction tool. Think of a broadband denoiser as a collection of bandpass filters, each followed by an expander (see figure 4 below). Denoising tools come in many configurations, from the now traditional approach pioneered by Sonic Studio’s NoNOISE range of products in the mid ’80s, shown in Waves’ X-Noise implementation (see figure 5 below) to the streamlined, two knob SoundSoap in figure 6 below. There are even graphic–free approaches, such as Arboretum’s Ray Gun Pro (see figure 7 below), the Swiss Army knife of restoration. [Since your ears are the final arbiter, a graphics-free approach isn’t as goofy as you’d think. - OM]

Boradband Denoiser
Figure 4 – Block diagram of a generic broadband noise reducer

Waves Xnoise
Figure 5 – The broadband noise reducer in Waves’ Restoration Bundle

Sound Soap
Figure 6 – Sound Soap’s broadband noise reducer

Ray Gun Pro
Figure 7 - Like David Brancaccio, Arboretum does the numbers

Most all denoisers share one feature, a “learning” mode. When given a small section of the noise of interest, uncontaminated by signal, the denoiser “learns” what is noise and what is not. It then “knows” at what amplitude to set the thresholds for all the expanders. Once the “fingerprint” has been taken, you’re free to mess with the controls until an acceptable result is obtained. If the noise is not uniform, you may want to try two passes, the first to reduce high frequency noise and the second to reduce noise in the midband.

Wideband noise is present, to some extent, in all recordings and, as the Bard might say, therein lies the rub. As mentioned above, your “ears” are the most important factor when removing broadband noise since too much reduction will step on the high end, causing the material to sound dull or wooly. In addition, some tools are quite capable of creating a characteristic “watery” sound when misadjusted. Subjectively, removing high frequency noise always seems to dull down the program, even if the was no high frequency content present in the original! So, think about the destination for the material and your audience’s expectations, since it’s likely they have never heard the original transfer. If you’re new to this, try generating a pink or white spectrum noise file then, with your available tools, become familiar with the look and sound of pure noise . I used the $25 Cacaphony from Richard F. Bannister to generate my test file but Black Cat Systems’ Audio Toolbox and other utilities also provide this function. For a low tech approach, record some interchannel noise from the FM band or check audio test CDs, which often have noise tracks you can digitally transfer.

Once you’ve trounced all over the audio with these DSP denizens, you may want to apply some tasteful EQ to what’s left. In the final analysis, it really comes down to expectations. Some clients expect pristine results from absolute crap and other know that DIDO controls our destiny: doo doo in really does equal doo doo out!

Sidebar – Hi, My Name’s MeRL
Those of you lucky or stupid enough to have an analog tape machine should already know about alignment tapes: created under controlled conditions, these costly recordings provide reference level of fluxivity or magnetization that are used to adjust or “align” the reproduce electronics or “repro chain.” After adjusting the mechanical path through the tape transport, repro alignment is followed by record alignment. the final step in analog tape machine setup. Magnetic Reference Laboratories, MRL, and STL or Standard Tape Labs, are the two most common vendors of alignment tapes.

Sidebar – EasyBake Oven
Back in the dark days of pop, the 80’s, when I was a young engineer doing rock and roll dates in Miami, we wrestled with what is now a well known problem. Old school analog tape is a carefully craft mix of iron oxide, aka “rust,” attached to a flexible plastic ribbon or “backing.” The chemists at the big tape manufacturers of the time, Ampex, 3M, BASF and Agfa, had decided they could “improve” their otherwise perfectly acceptable tape products. So, they all changed the formulation for their binder, the glue that hold their proprietary rust particles onto the backing. Trouble was, their new binders were hydrophillic or water–loving and, in humid Miami, it didn't take long for us to find our brand new master reels literally falling to pieces as the water softened binder stuck to the tape transport rather than staying attached to the backing. Major bummer.

Nowadays, a low temperature oven and plenty of patience will restore these rancid recording to like–new status. An inexpensive convection oven, set to 120 to 130° will drive out the moisture slowly, leaving a more robust tape. Raise the temperature slowly until you reach your comfort level, leave it for 4 to 20 hours depending on the setting and your patience level, then slowly ramp it back down to room temperature. Slowness is the key to prevent uneven drying and subsequent mechanical stresses to the tape. If you’re going to try this, spend some money on an accurate thermometer as too high a temperature spells toasted tape and certain doom. Be sure to store the tape in a sealed plastic bag or container after baking to prevent rehydration.

Sidebar – One Isn’t The Loneliest Number
Zero is…What the heck is DC offset? In the audio world, we often generalize when we discuss the audio passband, that 20 to 20 kHz range of frequencies that some of us can hear, sort of…As engineers, we should really be concerned with a much wider range of frequencies, from subsonics to ultrasonics. Subsonics, frequencies too low to hear, range from 20 Hz down to 0 Hz. Another name for zero Hertz is DC or direct current, as the zero cycles per second implies that the signal polarity never changes.

Analog audio is AC (alternating current) but a DC signal, also called DC offset, can creep into a recording by various means, from a badly matched phantom powering circuit to an El Cheapo™ A/D converter. Since DC offset only robs you of dynamic range and heats up your woofer’s voice coil, it’s best to get rid of it early on in the restoration process.

Sidebar — IIR vs FIR
This is a scary concept but here goes: All filters suck, some more than others. Whew, I’ve gotten that off my chest. Now, what’s the bottom line here? Well, filters or EQ can be built in two flavors, either Infinite Impulse Response (IIR) or Finite Impulse Response (FIR). Most filter that we interact with on a daily basis are of the IIR type. They trade off a frequency–dependent amplitude boost or cut in return for some amount of frequency–dependent phase shift, also known as group delay. Unfortunately, that’s a problem…although we expect them to operate in the frequency domain, they also muck up the time domain. To understand this group delay thing, think of your basic two way loudspeaker (see figure 8 below). The tweeter and woofer are both mounted on the front baffle. The tweeter’s tiny voice coil is very close to the baffle while the woofers large cone offsets it’s voice coil back quite a bit from the baffle. Now, the voice coil and cone comprise the motor that makes a speaker move air so broadband sound launched from the woofer is offset in time from the tweeter’s output by the physical displacement between the two devices. That’s why many loudspeakers have slanted back-leaning front baffles, which time aligns the drivers.

acoustic group delay
Figure 8 – An simplified version of group delay via frequency-dependent acoustic delay

This is classic group delay, where the high frequencies from the tweeter arrive at your ear before the woofer’s low frequency content. Think of what this “time smear” does to a broadband sound like a tasty kick drum, a signal with both the HF snap of the beater and the LF boom of the shell’s resonant cavity. Anyway, this same time smear or group delay occurs in all IIR filters to a greater or lesser degree. The only difference is, unlike a speaker, in electronics the high frequencies lag while the low frequencies lead.

As to FIR filter, they exhibit a constant group delay regardless of frequency, so no wonky phase shift problems but, they have a different problem: pre–echo. FIR filters have an annoying tendency to present a small amplitude version of their input at the output, before the input has been applied! I know, what have I been smoking? Nothing, old boy, I simply haven’t mentioned that FIR filters cannot be built in the analog world, only in the bits and bytes of a digital implementation. So, a rip in the time–space continuum that I’ve just mentioned is taken for granted in digital signal processing circles…It’s just one more thing a designer must contend with.

The upshot is that FIR filters, since they lack group delay, don’t “sound” like EQ as we know it. Because of their sonic neutrality, they’re usually used only for specialized correction like in mastering, forensics…or restoration! Though they don’t exhibit group delay, they do impose significant latency due to significant computational overhead. The folks at Elemental Audio Systems offer Firium, a smooth FIR filter in either VST or Audio Units format. Download the demo and take a critical listen…