Cleaning up clicks, pops, and crackle in recordings transferred from vinyl is perhaps the best-known application for audio-restoration (AR) software, but there are uses for it in virtually every situation where audio is recorded (for audio-restoration basics, see this month's “Square One” column). Whether you need to reduce background noise in film or video soundtracks, clean up audio for Podcasts, minimize the noise of computers and drives that leaks onto your music tracks, or repair clipped digital recordings, there's an audio-restoration product that can help you.
AR software comes in many varieties, from dedicated plug-in suites to standalone editors to noise-reduction features built into audio editors. (You may have such features included in software you own and not even realize it.) Although AR software titles offer the tools to tackle noise issues, these products will frequently yield the best results with subtle application. A heavy-handed approach to AR processing can often make a bad-sounding recording worse.
Full-featured AR programs or suites typically consist of several major tools. Common to most are a hum-and-rumble remover, a click-and-crackle remover, and a broadband-noise reducer. Manufacturers typically suggest that if you're going to use all three on a piece of audio, you'll get the best results by applying them in the order just mentioned. Some AR products also have noise gates, declipper modes for regenerating clipped sections, and other specialty tools.
In this article, I'll provide tips and techniques for using AR software in common audio-restoration scenarios. The information was gleaned from talking to professional users and software developers, and from my own experience with AR software. This is not a product roundup, but I will mention plenty of different programs and plug-ins along the way.
One of the most common reasons people turn to audio-restoration software is to reduce broadband noise. As the name implies, a broadband noise is one that covers a range of frequencies (as opposed to, say, 60 Hz hum). For instance, hiss, background noise, machine noise, and even people's voices would be considered broadband noise. Even with the prevalence of 24-bit recording and its improved dynamic range, hiss and background noise bedevil all types of recorded material, particularly dialog, interviews, and other spoken-word recordings (see the online bonus material “Denoise in the Name of the Law” at emusician.com).
Plug-ins like BIAS SoundSoap and SoundSoap Pro, Wave Arts MR Noise (from the Master Restoration Suite), and Waves Z-Noise and X-Noise (the latter is part of the Restoration Bundle); standalone applications like Enhancedaudio DC Seven and iZotope RX; and audio editors such as Adobe Audition, Apple Soundtrack Pro, BIAS Peak and Peak LE, Sony Sound Forge, and Steinberg WaveLab all offer denoising tools.
Denoising software requires that you sample a bit of program material, preferably a short section containing only noise, so that the software can calculate the frequencies and levels of the noise signal. This setting is sometimes called the “noise profile” or “noise print.” Then during playback, the software will use this profile setting to squash down any frequencies that fall below these levels at the specified frequencies. Essentially, it's a multiband gate.
FIG. 1: BIAS SoundSoap Pro''s denoiser module offers both a “learn noise” option and another that extracts a profile from a user-specified section of the audio that includes both source and noise.
“At every frequency,” says Bill Gardner of Wave Arts, “it knows what the level of noise is. So while the music's playing, if you're above that level, it decides it must be signal; it'll let it pass. But if it falls down to the level where it thinks it might be noise, it starts to attenuate it.” Although Gardner was speaking specifically about the MR Noise plug-in, what he said is descriptive of broadband-noise reducers in general.
Some denoising software can automatically detect the noise portion of the audio, even if you don't have a section of solo noise to sample. For example, Waves Z-Noise has a setting called Extract, and the standalone version of iZotope RX has one called Auto-Train. These functions search your audio for a section that contains only noise and sample it automatically. BIAS SoundSoap Pro (see Fig. 1) offers an option called Timed Learn Noise, which samples a user-specified segment of program material and then calculates an average noise profile from within it.
Tweak That Thing
Although a denoiser calculates its settings automatically from the noise profile, don't expect instant perfection. You'll typically need to do some additional tweaking of the threshold and reduction (sometimes called “attenuation”) controls — which are offered in virtually all full-featured noise-reduction software — to get the best results.
The threshold governs the level below which the reduction kicks in, at the frequencies identified in the noise profile. The reduction slider controls the amount of the process that's applied (similar to the ratio setting in a conventional noise gate). Experimenting with these settings helps you find the best compromise between reduction amount and sound quality. Too much reduction will cause glassy or watery-sounding artifacts, sometimes referred to as “musical noise” (see Web Clip 1; in his “Square One” column on audio restoration in this issue, Brian Smithers refers to these as “birdies”; I've also heard them called “space monkeys”). Other artifacts of the noise-reduction process include harsh-sounding audio and degraded transient response.
Most noise reducers offer other tweakable parameters to help you come up with the best trade-off between noise reduction and audio quality. It's important to learn the capabilities of these parameters to get optimal results.
One tool that can help a great deal in figuring out your settings is the noise-monitoring feature common to most AR software. It lets you listen separately to what you (or the program's automatic settings) have defined as the noise portion of the signal. At the press of a button, you're able to listen to the portion of the audio that's being taken out. This is particularly useful for checking how much of the target (source) signal, if any, you're affecting along with the noise.
Set and Don't Forget
Let me focus for the moment on reducing noise in a primarily spoken-word recording, such as in a video soundtrack or an audio interview. How you set the controls depends on whether you're aiming for ultimate sound quality or just intelligibility. You also have to assess the severity of the noise and come up with a realistic goal for the final result.
If the noise is relatively quiet compared with the target signal, it's usually pretty easy to clean it up without much degradation. However, as the relative noise level increases, getting both intelligibility and good sound quality can become much more difficult.
Curtis Crowe of Enhancedaudio, who has a great deal of restoration experience, terms a recording with more noise than target signal as being “upside down.” Fortunately, he notes, most of the time you won't be dealing with such an unfavorable balance. “Usually it's not upside down,” Crowe says. “As long as that's the case, we're likely to get excellent results.”
To tap into Crowe's audio-restoration methodology, I asked him how he would handle a hypothetical upside-down interview recorded on a busy and noisy convention floor. “That would be an incredibly difficult thing,” he says. “Obviously all the noise on a convention floor, same as in a restaurant, would be almost identical [to the source].” In other words, both the target audio (the interviewer and interviewee's voices) and the primary background noise (numerous people talking) are of a similar character, making the process of separating them out harder from a frequency standpoint.
“One thing that we'll always try,” says Crowe about such a situation, “is to use a continuous noise filter [broadband-noise reducer], and we know that we'll introduce a few artifacts. First we want to take a sample of the background noise. Typically I'll take one that's a little longer than normal, about 1 second. I'd rather have it without the target signal. If I can find a [full] second, fine; if not, I'll take what I can get.” The longer the sample of noise, the more accurate the settings that the software calculates will be.
FIG. 2: Punch and Crunch, the multiband compressor/expander in Enhancedaudio''s DC Seven program, can be used to home in on the target audio in particularly noisy passages.
“I'll start with my continuous noise filter [in DC Seven],” he continues, “doing almost nothing with the attenuation slider, and then I'll inch it [the slider] up, not trying to remove the noise, but trying to improve the signal-to-noise ratio by whatever amount I can. And when I move it up, I'm going to induce some artifacts as I reduce the noise, because my sample cannot match the noise correctly.”
Crowe would also apply another effect in that situation: an expander. “I'd use a multiband expander to try to lock on to the target voice to try to make it louder,” he says. “A tool in DC Seven is called Punch and Crunch [see Fig. 2]. It's a 4-band compressor/expander, and it shows you the energy in each band. And the bands are adjustable. So I'd probably start off with two bands covering 300 Hz to 3,000 Hz. And then, as I watched where the energy was in his voice, I'd try to narrow that to make the expander only effective when he's talking, as much as possible.”
Treat the Music
So far I've focused on spoken-word recordings, but what about noisy music tracks? People transferring from vinyl or cassette to digital often find source material that needs noise reduction (see the online bonus material “Restoring the Past”). Even in music that you've tracked in your own studio, you might have background-noise problems resulting from recording at too low a level, tracking a noisy source like an overdriven, compressed guitar signal, or picking up machine noise.
I spoke to producer (and EM author) Steve Skinner, who was having trouble with machine noise, particularly from his CPU and disk drives, getting on his tracks. “Disk-drive noise is particularly tricky because it goes up and down depending on the intensity,” he says.
He started using the Waves Z-Noise plug-in to clean up the tracks and has been very happy with the results. “I've been using it so far on guitar, upright bass, and spoken voice, and I think I've used it on the sung voice,” he says. “And as far as my ears can tell, I haven't heard it mess with the original signal.”
FIG. 3: Waves designed three presets into Z-Noise that let you tailor the plug-in''s response depending on the transient character of the source audio.
Here he describes his approach to using it on music tracks: “I'll open it up and put it on the learn function and do about 30 dB of noise reduction, and just move the threshold until the noise goes away. I'll then leave that setting and put it on the track using AudioSuite.”
Z-Noise offers a number of settings to adjust the behavior of the plug-in depending on the nature of the transients in your material. The Punch setting is designed specifically for situations where you want to preserve the transients as much as possible (see Fig. 3).
Bassist and producer Kevin “Brandino” Brandon (brandino.com) describes a massive audio-restoration job he did, making a CD (A Voyage Again/Con Mucho Gusto; N' House Records, 2006) from a cassette master of a live performance by jazz pianist Milcho Leviev from the Don Ellis Big Band, which was recorded back in the '80s. The playing was great, but the audio was quite noisy. Brandon and engineer Michael Gunderson turned to the software tools in Sony Sound Forge to clean it up (see Web Clips 2a and 2b).
They first used a pretty heavy dose of Sony's Noise Reduction (see Fig. 4), which is a broadband-noise reducer, but found that a subtler approach was necessary. “I listened back and discovered it was too clean,” Gunderson says. “It took some of the life away from the material.”
FIG. 4: Sony Sound Forge 9 with the built-in Noise Reduction effect open.
They then started over with a more gradual approach, applying denoising in layers, using both Noise Reduction 2.0 and the (fixed) noise-reduction algorithm in Sound Forge's Audio Restoration tool. “First I did one pass of Audio Restoration using a minimal to moderate setting, and it seemed to clean up the noise floor, but there still was some tape hiss,” recalls Gunderson. “So I did three passes of the Noise Reduction 2.0 to really strip away the tape hiss, using a very minimal setting each time. At this point you might wonder about any sound degradation. I would say it was minimal at best.”
Because audio quality is always paramount in music recordings, you have to be even more judicious in your use of AR processes than you might be in spoken-word applications. Be sure to solo and compare the track both processed and unprocessed. Be extra careful to apply settings that don't noticeably degrade audio quality.
Sometimes you're up against electrical hum, which is much more predictable in terms of its frequency and behavior than broadband noise typically is. The hum removers in AR software generally use EQ to notch it out, with presets at the basic hum frequencies of 50 or 60 Hz that extend up to harmonics of those frequencies. If your noise is electrically based, the hum remover can be very effective.
Jason England is a sound editor for film and television who uses BIAS SoundSoap Pro to deal with a range of noise problems, including hum. “I need something that's just click and go. The hum-and-rumble remover, you just turn it on. As an audio person you already have an idea of the frequency that you want to hit, and then you just slide it and do it.”
Even if you don't have AR software, you can use conventional EQ to try to reduce hum and rumble. For hum, set a steep cut with a very narrow bandwidth at 50 or 60 Hz and see if that helps. You can also use a highpass filter or brickwall filter to get rid of rumble. A good starting point for the former is to begin rolling off at about 60 Hz. With the brickwall filter, start out at 40 Hz. Of course, you'll want to experiment to find the best possible setting.
Pop Goes the Click
Pops, clicks, and crackle can plague an audio recording. The latter two are particularly problematic when transferring from vinyl. Luckily, AR programs and plug-in suites are all equipped to deal with these intermittent pests. Often you can set a click-and-crackle remover plug-in to run across the length of a track and automatically remove them.
But as with broadband-noise-reducing software, setting the remover correctly takes some work. Some people will listen to the noise-only signal and tweak the settings until they hear only clicks and crackle, and no source signal. Crowe takes a different approach.
He likes to listen to the full signal when setting the click remover in DC Seven. He finds a short, representative section, maybe 15 seconds long, and loops it. “I want to hear the same thing again and again and again. My ear gets used to it,” Crowe says. “I then adjust it with my mental focus on the clicks until I make them all go away. And then I bypass the filter, and instantly, while it's playing, it stops filtering. And then I hear all the clicks come back, and then I shift my mental focus to the target signal, the music, or whatever. And then I pop the filter in and out, in and out, while I'm listening. And if I can't hear any degradation in the target signal, then I've reached my point.”
But clicks and crackle don't occur just on vinyl. England is often bedeviled by RF static that gets onto the dialog audio after being picked up in the actors' wireless lavalier mics. He uses BIAS SoundSoap Pro to help remove these impulse noises.
England applies SoundSoap Pro's click-and-crackle remover differently depending on the severity of the problem. “I might zero in on that one little area where there was a little static,” he says, “but sometimes it's the whole take that's staticky, and I might just process the whole take.”
As with denoisers, overly high settings on a click remover can reduce transient response. So if you're declicking a rhythmic track, use moderate settings if possible. (For more about click removal, see the online bonus material “Click Away.”)
Distortions of Reality
Digital audio is very unforgiving when it comes to clipping. If you go into the red, you're likely to get nasty-sounding distortion. I used to think that if a take was clipped, it was ruined. However, while researching this story, I discovered that there are actually software-based remedies for clipping that can sometimes salvage a distorted recording.
Some AR programs, including RX and DC Seven, have declipping tools built in. These processes use interpolation; that is, they analyze the clipped waveform, calculate what it would look like if it hadn't gone over, and then use that information to generate a nonclipped replacement.
Crowe explains more specifically how the declipping algorithm in DC Seven works. “It measures the slope on the positive-going part of the wave and then on the negative-going part, and then it calculates what the clipped-off top should have been. But it is [only] a prediction.”
Declipping tools are not foolproof. They often require trial and error with different settings to get a satisfactory result. Sometimes they're simply not effective. Still, they give you a chance to recover otherwise ruined recordings.
Crowe suggests a surprisingly effective alternate tool for dealing with clipped audio: a de-esser. “Clipping results in harmonic bursts of energy that shouldn't have been there,” he explains. “And so they're all higher than the fundamental frequencies that we do want. A de-esser cuts the top-end frequencies, and you can see how that would help. I typically set it to around 2 kHz.”
While previewing the de-esser effect, Crowe adjusts the frequency range until he finds the right level of attenuation. “I'm going to adjust until I've gotten rid of the distortion without killing the high end. And I back it off until I reach the right balance, and you can't predict until you do it what that right balance point is.”
I tried Crowe's de-esser method on some clipped segments on an interview recording, and it worked quite well (see Web Clip 3). Because virtually all DAWs have de-essers, you don't even need dedicated AR software to apply this remedy. You will lose some high end, but for spoken-word recordings, that's not always critical.
FIG. 5: iZotope RX''s main view offers overlaid waveform and spectral editing views. RX''s Spectral Repair module uses interpolation technology to fix audio glitches.
Another effective tool for audio restoration is an editor with a spectral display. The spectral view sometimes allows you to see anomalies much more clearly than in a typical waveform display. Programs like RX (see Fig. 5) and Adobe Audition offer spectral views that you can actually edit in. One way this can be useful is if you have an audio problem that occurs within a limited frequency range. You can zero in on it visually and edit it, while leaving the rest of the audio unaffected.
RX, for example, lets you select by time (across the frequency spectrum), by frequency and time (a selected frequency range in a selected time range), and by frequency (a selected frequency range across the entire file). You can choose to listen to the selected area only, which is useful for homing in on the selection you want. You can then apply one of RX's many processes to the selected area.
A unique and incredibly effective tool in RX is its Spectral Repair module. It provides several options for repairing audio problems, including Replace mode, which uses interpolation to replace a small section of problem audio based on the audio surrounding it. For instance, it works great to remove string squeaks from acoustic guitar tracks, essentially replacing the squeak (which you've isolated in the spectral editing view) with audio generated from the note surrounding it. Like any AR tool, it takes some trial and error. I've found RX's Spectral Repair able to fix a range of otherwise unsolvable problems ranging from severe clipping to stray noises (see Web Clip 4). It's the tool I turn to when nothing else seems to work.
Noise No More
Audio restoration offers many options for quieting noisy recordings. Which product you use and how you use it are up to you, but the bottom line is that digital technology has made the repair of damaged and badly recorded audio much more possible than ever. Whether you're planning on buying a dedicated program or plug-in suite or you just want to learn to use the AR features available in your 2-track editor, I highly recommend that you dive right in. The knowledge and skills you gain are sure to come in handy sooner or later.
Mike Levine is EM's executive editor and senior media producer and the host of the monthly Podcast “EM Cast” (emusician.com/podcasts).
Audio-Restoration Bullet Points
- Always preview before applying destructive processes. Make sure you have a backup copy of the audio.
- If your software allows it, pop the AR processing in and out to compare it with the unprocessed audio.
- Don't rely strictly on automatic settings, especially in broadband-noise reducers. Experiment with the threshold and reduction sliders (or equivalents) as well as parameters like attack and release. If you start hearing artifacts, back off on the amount of the effect.
- Try to find as long a sample of noise for the “learn” function as you can. It will help with the accuracy of the settings.
- Use the “noise-only” monitoring feature to check the part of the audio being removed by the software.
- In heavy noise situations, consider a multiband expander or an EQ boost set to the frequency range of the target audio to help bring it out.
- If the audio contains rhythmic material, make sure the AR software isn't degrading the transient response.
- If you don't have declipping software, try de-essing in the area of 2 kHz (and higher) to minimize the distortion artifacts of clipped audio on spoken-word tracks.
- When editing out an anomaly from a file being used in a soundtrack situation, don't delete audio, or you'll change the timing. Eliminate it by reducing level instead.
- Don't edit out time when cleaning up a video soundtrack, or you'll likely affect the sync. Use attenuation to get rid of glitches instead.
In most DAWs, if you're denoising using a plug-in, you have the option to use it (or other AR processes) nondestructively throughout the entire duration of your track, or apply it destructively to specific sections where needed. The former scenario can be advantageous because instead of committing to a setting right away, you have the option to change up until the point that you actually bounce or run the mix. This is also helpful because it's easy to make your initial noise-reduction settings too strong.
But when you're dealing with intermittent broadband noise, or noise that changes in intensity or content over the duration of your track, applying AR processes destructively (such as using the denoiser as an AudioSuite plug-in in Digidesign Pro Tools), with specific targeted settings, is probably easiest. If you're using Waves' Z-Noise plug-in, you can turn on Adaptive mode, which is designed to change its settings automatically as the noise levels change in the program material. That gives you the advantages of nondestructive processing and the ability to handle a changing noise profile. (If you don't have Z-Noise, you could use automation to change your plug-in's settings over the course of the audio track.)
Before you apply destructive processes to your audio, it's always wise to make a backup copy in case you overdo it. In my work as a Podcast producer, I've had a number of situations where I used broadband-noise-reduction software on a track, only to decide later that I'd degraded the audio quality too much in the process. Because I had a backup, I had options to redo the processing in a more sonically pleasing way.
Sony Creative Software
Steinberg Media Technologies