TABLE OF CONTENTS 1: INTRODUCTION................................................................................................................5 Intended Audience for this Guide.........................................................................................................5 About the 2014 Edition............................................................................................................................6 Additional Resources..........................................................
8: BROADBAND NOISE REDUCTION........................................................................... 57 Hum and Tonal Noise Reduction........................................................................................................ 62 Tonal Noise.............................................................................................................................................. 70 Dialogue Denoising.............................................................................................
APPENDIX A: GETTING SET UP TO REPAIR AND RESTORE AUDIO...................121 Equipment................................................................................................................................................121 Monitoring Audio...................................................................................................................................122 Transferring Media to the Computer.............................................................................................
1: INTRODUCTION In today’s world, audio and video recordings can be made by almost anyone. From smartphones to sophisticated studios and sound stages, large amounts of media content are created daily. And as our world becomes increasingly and audibly congested, the rate of ruined recordings is rising in tandem.
ABOUT THE 2014 EDITION The 2014 edition of this guide has been revised and updated by the experts at iZotope HQ, based on years of research and consultation. ADDITIONAL RESOURCES If you’re interested in exploring audio repair and restoration in greater depth, there are plenty of resources available. The iZotope YouTube channel (www.youtube.com/izotopeinc) has a number of audio repair and restoration tutorial videos aimed at anyone from the beginner to the advanced audio expert.
2: WHAT IS AUDIO REPAIR AND RESTORATION? When you hear the words “repair” and “restoration,” you might be inclined to think of dusty vaults filled with aging master tapes and records. Some audio repair and restoration projects may indeed involve taking old recordings and reviving them, but the methods used are useful for a wider range of scenarios. Every time you record audio—whether at home, in the studio, or on location—there’s always the chance of encountering unexpected and unwelcome audio “guests.
3: AUDIO REPAIR AND RESTORATION BASICS DEFINING THE OUTCOME The goal of good audio repair and restoration is to render the best possible sonic result with the least audible human intrusion. In essence, your intervention in the original recording should be transparent and not introduce new artifacts that distract the listener. Sometimes it’s possible to solve an audio problem entirely, and other times it’s about finding the right balance between reducing the problem and preserving the original audio.
• Declickers are used to reduce and remove intrusive clicks and pops. These can be caused by anything from dust and scratches on an old record, a CD skipping on playback, or even mouth clicks and lip smacks from a voiceover. • Decracklers are closely related to Declickers, but are optimized to help reduce and remove a more continuous, quieter stream of clicks that blend together to cause what the human ear perceives as a general crackle.
GENERAL RECOMMENDATIONS WHEN REPAIRING OR RESTORING AUDIO We highly recommend that you educate yourself about the function of individual tools in your toolbox. Dedicated audio repair and restoration tools can do a fantastic and fairly autonomous job, but learning when, where, and to what degree of strength to use a specific tool can yield better, faster, and more transparent results. It’s also beneficial to establish your aim before setting out.
RX TIP • Using RX 4 document files and saving module-specific presets can save you the trouble of writing out all of the parameters on a recall sheet as you would in the analog domain. 5. Back up your work—the first and last rule of any audio editing project! You never know when a hard drive, backup device, or original master might fail. Again, always back up your work! THE TOOLS OF THE TRADE In the following chapters, we’ll briefly examine some of the essential audio repair and restoration tools.
4: UNDERSTANDING SPECTROGRAMS / IDENTIFYING AUDIO PROBLEMS As with medical diagnostics, the key to successful audio restoration lies in your ability to correctly analyze the subject’s condition. This can be a life-long, never-ending quest—constantly honing the ear to distinguish the noises and audio events that need to be corrected. To get started, it’s important to identify the problems with your file and identify which tool(s) will give you the results you want.
PRINCIPLES OF SPECTROGRAM DISPLAYS So what is a spectrogram? A spectrogram is a very detailed, accurate image of your audio, displayed in either 2D or 3D. Audio is shown on a graph according to time and frequency, with brightness or height (3D) indicating amplitude. Whereas a waveform shows how your signal’s amplitude changes over time, the spectrogram shows this change for every frequency component in the signal.
One thing you’ll notice when looking at the waveform display is that it’s good at showing audio amplitude, but less effective at showing what’s happening at different frequencies. For example, we can easily see here that the sine wave is the same level for the entire duration of the file. However, we can’t tell much about how the pitch or frequency changes over time. Now let’s look at this same audio file using a spectrogram.
Now let’s look at something more complex: the human voice. Here’s a short, spoken phrase as seen through a waveform display: What we’re seeing here is the amplitude of the spoken words over time. If we switch to the spectrogram view, we’ll see many things we can’t see in the waveform view: The human voice is much more complex than it might seem from looking at the waveform view.
This is why having a detailed spectrogram display is so important to doing audio restoration. It helps you clearly see the problems that you’re trying to fix. SPECTROGRAM TYPES Not all spectrograms are created equal. An algorithm known as the “Fast Fourier Transform,” or FFT for short, is used to compute this visual display.
RX TIP • RX 4 has an Auto Adjustable mode that automatically combines different FFT sizes on different areas of the frequency spectrum, so it’s always easy to see everything that’s going on. Now we’ll move on to some specific examples of how to visually identify noise and other audio problems. Learning to identify these problems by sight will greatly help you—it means you’ll be able to use any software that includes spectrogram technology.
USING A SPECTROGRAM TO IDENTIFY AUDIO PROBLEMS Hum Hum is usually the result of electrical noise somewhere in the recorded signal chain. It’s normally heard as a low-frequency tone based at either 50 Hz or 60 Hz depending on whether the recording was made in North America or Europe. If you zoom in to the low frequencies, you’ll be able to see hum as a series of horizontal lines, usually with a bright line at 50 Hz or 60 Hz and several less intense lines above it at harmonics.
TIP • Hum Removal is ideal when frequencies of hum do not overlap with any transient useful signal. Buzz In some cases, electrical noise will extend up to higher frequencies and manifest itself as a background buzz. See the example below: TIP • Hum-removal tools usually focus on low-end hum, so when the harmonics extend to higher frequencies, a denoiser is more effective at removing the problem.
TIP • Denoisers are very effective at dealing with this type of broadband noise. Clicks, Pops, and Other Short Impulse Noises Clicks and pops are common on recordings made from vinyl—but can also be introduced by digital errors, including recording into a DAW with improper buffer settings, or making a bad audio edit that missed a zero crossing. Even mouth noises such as tongue clicks and lip smacks fall into the clicks category. These short impulse noises appear in a spectrogram as vertical lines.
Clipping Clipping is an all-too-common problem. It can occur when a loud signal distorts on input to a sound card/ converter, mixing console, field recorder, or other sound capture device. A spectrogram is not particularly useful for identifying clipped audio—for this you’ll want to work with a waveform display. RX TIP • Move the slider below the spectrogram to the left to superimpose the waveform display. As you’ll see in the image below, the clipping appears as “squared-off” sections of the waveform.
TIP • Declipping tools can intelligently redraw the waveform to where it might naturally have been if the signal hadn’t clipped. TIP • Sometimes, heavily limited audio will also appear “squared off” when zoomed out, but this doesn’t necessarily mean it will sound as heavily distorted. You can zoom in to see if individual waveform tops are clipped. Intermittent Noises Intermittent noises are different than hiss and hum—they may appear infrequently and may not be consistent in pitch or duration.
Gaps and Drop Outs Sometimes a recording may have short sections of missing or corrupted audio. These are usually very obvious to both the eye and the ear! See the example below: TIP • Noises and dropouts like the example above are often unpredictable, and usually need to be removed or patched manually using a visual/spectral editing tool.
5: WHAT IS RX 4? As we explore the various ways one can approach audio repair and restoration, we’ll be using iZotope RX 4 to demonstrate key concepts and showcase some unique tips and tricks. RX 4 is a complete audio repair and restoration suite, consisting of both a standalone audio editor and software plug-ins for use within a DAW. RX 4 offers advanced visual representations of your audio and innovative new processing technology that removes noise and repairs audio.
6: DENOISING “Denoising” in the context of audio typically refers to the reduction or removal of steady-state background noise. As explained in an earlier chapter, steady-state noise might include constant ambient noise, tape hiss, or electrical buzz and hum. WHAT’S THE GOAL OF DENOISING? There are many situations where removing steady-state background noise can greatly improve the quality of the material.
1. Reduce > Remove. Often, if the goal is to improve the listening experience or increase headroom, the best result isn’t necessarily obtained by attempting to remove the noise entirely. Rather, reducing the noise to an acceptable or indistinguishable level often yields better results. 2. Doing multiple, gentle processes will often yield a more natural result than one harsh process of denoising.
4. Applying different levels of noise reduction to different areas of the frequency spectrum can help reduce artifacts. For example, try applying more gentle processing to the high-mids than you would to the low-end, which usually contains less important audio information. TIP • Denoisers with per-band parameter adjustments are designed to accommodate this workflow. RX 4 Advanced makes this available should you need it.
7: TIPS AND TRICKS FOR EDITING DIALOGUE The process of recording, editing, and mixing dialogue for mediums such as radio, TV, and film is often challenging. From news reports to reality TV, the current emphasis on field recording (made possible by advances in portable recording technology) also introduces audio problems such as distortion and background/ambient noise, to name a few. These problems are by no means reserved for location-recorded dialogue.
Removing clipping may not remove the audible distortion entirely. For more serious issues, Decrackle and Deconstruct can help mitigate harmonic distortion, as demonstrated in Steps 2 and 3. Step 1: As seen above, the waveform is truncated. Open Declip, and click Suggest. This sets the Threshold(s) just below the point of clipping.
If the clipping occurs only on one side of the waveform, or perhaps is more serious on one side than the other, RX 4 Advanced can de-link the Threshold sliders. This helps obtain a better sound, by avoiding over-processing the side of the waveform with less clipping. The negative Makeup gain and the Post-limiter assist in preventing the redrawn waveform from clipping again, which it otherwise is likely to do. Click Process. The repaired audio now looks and sounds more natural.
If distortion artifacts are still audible, and adjusting the Declip parameters doesn’t achieve the desired result, move on to Step 2: Step 2: The above waveform was heavily clipped. Declip can repair the truncation, but will not fix the harmonic distortion that occurred in the high frequencies. The problem is hard to see visually, but easy to locate manually using the next step.
Use the Frequency selection tool combined with the Play frequency selection tool to sweep around until it’s clearly audible where the problem is located.
Open the Declick module, and select the Decrackle tab. Decrackle is designed to handle a more continuous stream of irregularities in audio that become blended together at lower amplitude levels, resulting in what we hear as crackle. This helps clean up the additional distortion artifacts. Make sure you still only have the specific frequency range selected, then click the Play button to audition the overall net sonic result.
Step 3: Make the same or a similar frequency specific selection, and open the Deconstruct module (RX 4 Advanced only). Deconstruct can adjust the levels of tones and noise independently. Since bad harmonic distortion in dialogue often manifests as noise in the high end, using Deconstruct to attenuate the noise in a frequency specific selection is the last step in helping reduce distortion. These three techniques are key to reducing distortion in dialogue recordings.
TIPS FOR TACKLING BACKGROUND NOISE (STEADY STATE) Steady background noise in dialogue, if left untreated, is likely to create problems down the road when EQ and Compression are applied. Frequency boosts or cuts, and dynamic range adjustments can make the noise more noticeable. Most types of interference, especially those that happen on dialogue recordings, can usually be tackled with RX 4’s Dialogue Denoiser in Auto mode, which is the most efficient way to reduce noise.
RX TIP • In the spectrogram, this type of background noise appears as the steady image in-between the speech fragments. The quickest way to reduce this noise in the dialogue mix is to simply insert RX 4’s Dialogue Denoiser as the first plug-in insert on the dialogue track. By default, this works in Auto mode and will instantly reduce the noise. As well as saving time, Auto mode also allows the Denoiser to react to changing noise profiles, which are common in audio recorded outside.
Setting it to Manual mode allows you to teach it a specific noise profile, and automate the strength of processing in different frequency areas, which may achieve a better sounding result. The advantage to reducing noise in this way is that it’s non-destructive, so you can always go back and adjust the amount of noise reduction if you so choose later in the mix, rather than committing before you’ve had a chance to get any further in the mixing process.
Step 2: In this audio file, the background noise contains a lot of tonal elements that can be recognized as pitches or high-frequency buzz. This often comes from HVAC units and fluorescent lighting. RX 4’s Spectral Denoiser is well equipped to tackle these problems, which you should do inside RX 4 before re-importing the audio into your session. RX TIP • RX includes a plug-in called RX Connect that can send and receive audio directly from your timeline to the RX standalone application for processing.
Open the Denoise module, and select the Spectral tab. In Manual mode, use the Learn feature to teach RX the noise profile. Adjust the amount of noise reduction as needed, and click Process.
• Reduction curve: Use this feature to adjust the strength of processing in different frequency areas. Backing off the processing around 5 kHz, where the dialogue’s sibilance is most pronounced, helps avoid artifacts. • Tonal vs. Noisy reduction: Delinking these sliders allows independent control over the tonal and noisy elements in a noise profile.
TIPS FOR TACKLING PLOSIVES AND MOUTH SOUNDS Plosives are often unavoidable, especially when microphones are in close proximity to the presenter or voiceover artist. These pesky bursts of low-frequency information are usually found between 0 - 150 Hz, but can extend as high as 300 Hz, and appear as bright, momentary bumps on the spectrogram.
PLOSIVES Step 1: Right click on the Frequency scale to the right of the Spectrogram and select Extended Log. This display mode shows more information in the lower frequencies, so you may clearly see the plosives.
Step 2: Using the Brush Tool, draw around the plosive(s). Open Spectral Repair, and use the default Attenuate settings to process the selection. If the plosive is still audible, step back and try processing with a higher Strength value.
TIP • If your file is long, and drawing out each individual plosive would take some time, try using the Frequency selection tool to make a blanket selection between 30 Hz to just under where the dialogue sits, and process this selection with Spectral Repair’s Attenuate in Vertical mode and a weighting of After.
Step 3: If the plosive is reduced, but still impactful and audible, try further reducing the impact with Clip Gain. Right-click on the Spectrogram, and select View Clip Gain.
Left-clicking adds nodes, and drawing in a slight, momentary volume reduction can help reduce the impact of the plosive.
MOUTH NOISES Step 1: Open the Declick module and select the M-band (random clicks) algorithm as well as the Click type “Click”. Step 2: With Clicks only selected, click Preview, and adjust the Sensitivity slider up until you start to hear bits and pieces of words and phrases. Since you are currently auditioning only the clicks RX will be removing, this means you’ve gone too far; so dial back the processing, uncheck Clicks only, and click Process.
Step 3: If the mouth noises persist, it is possible Decrackle may also be able to attenuate them. Open Declick, and select the Decrackle tab.
Use the Frequency selection tool to select the area where the mouth noises are most commonly heard, usually between 3 - 20 kHz. Click Preview in Decrackle and adjust the Strength slider until you hear the mouth noises start to disappear. Click Process to reduce the mouth noises within your selected area. TIPS FOR TACKLING MATCHING Matching different pieces of recorded dialogue together can be a challenge for the dialogue editor.
Matching EQ profiles Step 1: Open both the audio source that sounds good, and the audio source that requires matching in RX. Open EQ Match, and navigate to the tab that holds the good-sounding audio source. Click Learn to teach RX the EQ profile of this sound.
Step 2: Navigate to the tab with the audio source that requires fixing, and click Process. This will automatically recognize the sonic signature of the audio, and conform it to fit that of the source audio, thus matching the two different sounds together. Step 3: If the result isn’t perfect, you may wish to adjust the Amount slider. Typically an amount of between 85 100% is good. 100% is the maximum amount of matching, and may sometimes be too much.
Step 1: If working within a host, select the area in your audio track containing the clips that need fixing, and use RX Connect to send it to RX 4. Or, render out an audio file containing these regions and load it in RX. Step 2: Open Ambient Match and click Learn. You do not need to select what ambience is present in the file, as Ambient Match automatically identifies this for you.
Step 3: Click Process to apply the Ambient throughout the audio file, filling the holes. You may find upon listening back that the ambience applied within the holes is slightly quieter, causing a volume jump in the ambience that is still jarring. See Step 4 for a solution to this issue. Step 4: The Trim control is a smart tool that can be used to add or reduce gain to only the ambience that’s inserted where ambience didn’t exist before.
To create unlimited amounts of ambience for use as ambient beds to underscore scenes, add to ADR, or to add to an asset list for future use, try the following: Step 1: Working in RX: Open the audio file containing the ambience you want to capture, and then also create a new, blank RX Document in RX. Working in RX (from RX Connect): Send a clip containing ambience as a reference to RX, and then select a blank region in your host and send it as a repair clip.
Step 2: Learn the ambience from the reference clip, and then move to the tab that contains the blank audio file. Click Process to apply the ambience across the entire blank area.
Step 3: If you were using RX Connect, send this back to your host and render. If you were working from an RX Document, simply export your ambience as an audio file to use elsewhere.
8: BROADBAND NOISE REDUCTION In the spectrogram chapter, we learned how to use a spectrogram to visually understand what’s happening in your audio. Broadband noise appears as random speckles or a haze over the audio file—similar to the static you see on your TV when there is no signal. Below are screenshots of a noisy voiceover sound recording: FIG. 1 FIG. 2 Fig.
Step 1: For broadband noise problems, always use a Denoiser. The first step is always to create a noise profile. This may be done manually by isolating the noise, or automatically. A good denoiser offers both manual and automatic modes. Automatic modes—such as Adaptive in the RX 4 Advanced Spectral Denoiser—listen to the audio and learn the noise profile on your behalf. These modes are most suitable for speech and scenarios with a changing noise profile, or when you want a faster workflow.
Select an area of the signal that contains nothing but noise, and click Learn (or the equivalent in your Denoising software). As a general rule, a selection of 100 ms should be the minimum, but the longer your selection of noise, the more accurately the computer can reduce the noise. We recommend a selection of at least 1 second, preferably 2-4 seconds or longer if possible.
Step 2: Begin to adjust the level of noise reduction and listen closely to the changes that occur in the audio. Some people prefer to slowly increase the amount of noise reduction until it sounds right, and stop once they start hearing artifacts. Other engineers prefer to go hard, and then dial it back until it sounds appropriate.
RX TIP • Use the Output Noise Only feature in RX 4’s Spectral Denoiser to listen to just the noise. If you hear any desirable signal bleeding through, like low-level bursts of dialogue or music, you should reduce the amount of noise reduction. Levels of between -5 and -15 dB of noise reduction are common. If you find yourself approaching -15 dB of noise reduction and you aren’t satisfied, consider doing a gentler pass of around -7 dB. Relearn the noise profile, and perform a second gentle pass of -8 dB.
Step 3: Once your noise reduction is beginning to sound effective, direct your attention to any smoothing filters your tool may offer, such as the RX 4 Spectral Denoiser’s Artifact Control slider. Adjusting these smoothing tools can help eliminate any artifacts and preserve the main goal of improving the listening experience.
It’s usually easy to identify the fundamental base frequency, as it’s usually the most visible. Often, the fundamental frequency will be 50 Hz (Europe) or 60 Hz (North America) due to those regions’ difference in electrical delivery. This audible hum originates from bad ground connections or inducted power sources such as AC mains or transformers.
We’ll start by using the Hum Removal tool. Hum removal tools, such as the one in RX 4, are extremely precise filters, designed to notch out very specific frequencies. For basic hum with anywhere from two to seven harmonics, this is an effective tool.
Step 1: To begin, identify the frequency of your hum. Your ears and the spectrum analyzer module may help. Set the base frequency to 50 Hz or 60 Hz, depending on where the audio was recorded. If the hum isn’t located at 50 Hz or 60 Hz, you may use Free mode, which unlocks the filters and allows you to set the base frequency yourself.
RX TIP • Use the time selection tool to select an area of hum and click Learn. This automatically sets the filters to the correct frequencies based on your selection.
Step 2: Next, adjust how aggressively Hum Removal attacks the primary frequency by pulling down the first frequency node. The deeper the cut, the more hum will be removed—but you may also adversely affect wanted audio in that frequency range, so tread carefully.
Step 3: If you can see and hear additional harmonics, start reducing the gain of these as well. RX TIP • Use the Number of Harmonics control to select up to seven harmonics above the primary frequency. The spectrogram display makes it easy to identify the number of hum harmonics in your project.
Step 4: As you start reducing the additional harmonics, it’s a good idea to ensure you aren’t removing any desirable audio. A good hum removal tool allows you to adjust the “Q,” or width value of the filters. Generally speaking, narrower filters will provide a better result, as they can more accurately notch out frequencies without affecting the surrounding audio. However, they also produce more ringing, i.e. time smearing of transients.
TONAL NOISE Tonal noise, such as buzz from fluorescent lighting, is more difficult to trace back to a primary frequency and may extend up into high frequencies that are out of the scope of what RX 4’s Hum Removal module can handle. Tonal noise is different from the broadband noise discussed previously in that it tends to be concentrated at certain frequencies. For this, we can use the RX 4 Denoiser module, which gives us separate control over the tonal and broadband noise elements in a noise profile.
To remove Tonal Noise using the Denoiser, follow the steps for dealing with broadband noise outlined earlier in this chapter, but with the following additions: • When adjusting the level of noise reduction in Step 2, de-link the sliders by clicking on the Chain-link icon. Then, increase the amount of tonal noise reduction using the now de-linked Tonal slider. This will start to clamp down more heavily on the tonal elements of your noise profile.
• Make sure the reduction curve is showing on the noise profile graph. Click the toggle box to show it. Now you may use this reduction curve as an envelope to notch certain areas of the noise profile to be affected more heavily. Where you see tonal peaks, you can direct the Denoiser to be more aggressive in those areas.
DIALOGUE DENOISING Recorded dialogue may have broadband or tonal noise problems, so why not just treat dialogue with the aforementioned methods? Well, you can—and would likely get a good result! However, as mentioned in the denoising chapter, some denoisers are designed specifically for use on dialogue and vocals.
Step 1: You can use the Dialogue mode in the RX 4 standalone application’s Denoiser module, or use the dedicated plug-in in your DAW. Insert your dialogue-specific denoiser on your audio track, and allow the audio to begin playing. iZotope’s Dialogue Denoiser has two modes, Manual and Auto. Make sure you start with Auto.
A good dialogue-specific denoiser will offer multiband adjustment, allowing you to quickly adjust the amount of noise reduction applied to different areas of the frequency spectrum. In the Dialogue Denoiser Manual mode, you can adjust these controls. This is useful if your audio has a lot of hiss or low-end rumble, for instance. TIP • Apply gentler processing to the high-mid frequencies, as this tends to be where vocal artifacts are more audible.
Step 2: Increase the amount of reduction until you hear the noise begin to disappear. Listen closely, as you’ll want to set the amount of reduction slightly below the level at which it begins to affect the voice. TIP • Remember that a higher value for the denoiser’s reduction parameter doesn’t mean more noise, it means more reduction (and therefore less noise).
GENERAL DENOISING TIPS 1. Softly, softly. Be gentle and do multiple passes if necessary. This will often lead to a better sonic result than one harsh denoising pass. 2. Reduce with caution. The amount of reduction is the most important control, so listen very closely as you adjust it. It’s possible to over-compress or over-saturate an audio signal using conventional mixing tools, and it’s also possible to overuse a denoising tool. 3. Know your noise.
9: REMOVING INTERMITTENT NOISES AND GAPS Intermittent noises can include a wide range of intrusive sounds that don’t fall into the steady-state or impulse categories. These can include a cell phone ringing at a violin recital, a door hinge squeaking during an interview, or traffic noise interrupting the dialogue being recorded for a film. Gaps, dropouts, and short sections of corrupted audio are also a common audio problem. These can be caused by everything from a loose audio cable to digital errors.
RX TIP • In addition to the common horizontal and vertical time/frequency selection tools, RX 4 includes several advanced selection tools, such as Lasso, Brush, and Magic Wand. These provide much more control over selecting sudden audio events that change frequency and move about.
Terms such as brush and lasso are common across visual editing platforms. Here’s what they mean in the context of audio repair: • Lasso: A lasso tool lets you use your mouse to outline a freeform selection of an image. • Brush: A brush tool lets you use your mouse to outline a freeform selection with a defined brush size. The brush size is usually adjustable.
Once you’ve highlighted certain events, the audio processing that follows represents the final step of an audio repair.
There are several things that make intermittent noises especially hard to fix: • They can be wildly unpredictable in frequency and timing. • Unlike broadband noise, hum, clicks, and crackles, noises like this can’t be removed with an automated process and can be time consuming to fix. • Most traditional audio editing tools cannot effectively remove them without leaving many artifacts or damaged audio.
USING SPECTRAL REPAIR IN RX 4 In this chapter, we’ll explore using the RX 4 Spectral Repair module to remove intermittent noises and fill in unwanted gaps. We’ll begin with these algorithms as they work in particularly unique ways. Please note that other audio repair and restoration solutions may not include the following functionality, or might achieve the intended results differently. The process of using Spectral Repair requires making precise selections using either the basic or advanced tools.
Spectral Repair offers four different algorithms that you can use for audio repair: 1. Attenuate is an intelligent gain adjustment. It can be used to push unwanted audio events into the background—particularly useful if the unwanted audio events don’t completely obscure the desired signal. 2. Replace is used to replace damaged audio, including entire gaps and dropouts. It can resynthesize audio using the audio information surrounding the damaged area.
3. Pattern is suited to patching badly damaged audio that contains repeated components, such as instrumental vibrato. It’s an intelligent copy and paste function that incorporates advanced blending techniques. 4. Partials+Noise is a more advanced version of Replace. It’s more effective on heavily harmonic content, and focuses on detecting and resynthesizing harmonics.
INTERMITTENT NOISES Step 1: Begin with identifying unwanted noise in the spectrogram. See the spectrogram chapter for more details on using the spectrogram to hone in on different types of problematic audio. Using the selection tools, isolate the noise as precisely as possible. You can either draw a freehand selection around the edges of the unwanted audio, or use the Magic Wand tool in RX 4 to make the selection automatically.
Step 2: Once you’ve isolated the unwanted noise, open Spectral Repair and ask yourself the following questions: • Do you want to push the noise gently into the background? If so, use Attenuate to process the selection. • Do you want to completely remove the noise? If so, use Replace or Partials+Noise to process the selection. As a general rule, Replace is a good starting point for removing sounds entirely.
Step 3: Now that you’ve identified, isolated, and treated unwanted noise, it’s time to play back your audio. Listen out for any changes that may have been made to the desirable audio, such as the introduction of artifacts. Remember that listeners will only hear what you’ve left in the mix, and not what you’ve taken out. If you fail to identify and treat leftover artifacts, they may distract the listener.
In this second pass, you can select and patch the initial attack of the bell, which is a more mechanical sound. Understanding the different components that make up unwanted noise can help you identify how best to treat it, and whether one or several passes would be most effective. This process is similar to how you identified different types of noise and the ways to treat it in the chapter on denoising.
AUDIO DROPOUTS Step 1: Audio dropouts are easily spotted in either a waveform or a spectrogram view. Working with a spectrogram view does make the repair process easier. First, you’ll want to make sure you have the gap highlighted. A full-bandwidth selection tool is most preferred for this, and you should highlight a very small amount of audio to the left and the right of the gap. TIP • It’s best to select a small amount of audio to the left and right of the gap as clicks will be present.
Step 3: As mentioned above, listening back is important—particularly when performing repair on full bandwidth audio dropout. Repairing the audio dropout may work the first time, but sometimes you may need to do a couple of passes.
For instance, in the above image, we have an opera singer and an orchestra. Pattern was used to replace the dropout, resulting in this: Although this may sound good, we can see that the slowly evolving orchestral chord has been repaired abruptly.
In situations where a whole signal does not transition smoothly, it’s advisable to perform a second pass. Target the specific frequency area and process it with the appropriate algorithm. In this case, you should use Partials+Noise to replace the harmonic content. In this image, the audio dropout has been repaired and some additional steps have been taken to ensure a seamless edit.
GENERAL AUDIO REPAIR TIPS 1. Trial and error is good. Although you’ll likely get great results the first time around, you’ll become more effective at using Spectral Repair the more you experiment and work with it. 2. Provide more information. If you fail to get a perfect result first time, simple tricks like extending the surrounding region length parameter may help. 3. Look around you.
10: REMOVING CLICKS AND POPS Clicks and pops can occur at almost any stage of the recording process. They can be caused by surface noise from mechanical media (especially discs), static electricity, power lines, cell phones, mouth noises, inadvertent physical contact with a microphone, and bad audio connector cables. Occasionally, digital errors will result in unexpected clicks, too.
PRINCIPLES OF REMOVING CLICKS AND POPS Removing clicks and pops effectively is not possible with conventional tools, nor is it possible by notching with a precise EQ. This is because clicks usually cover a wide frequency bandwidth and are extremely momentary. Declicking tools are designed to identify and recognize the sonic signature of a click, enabling the user to then attenuate or remove it entirely, either one by one, or by processing the entire audio file at once.
USING DECLICK IN RX 4 The Spectral Repair chapter explains how to eliminate certain clicks and pops, but for other kinds of distractions, an even better tool is available for use. The Declick module in RX 4 repairs and reduces clicks, pops, and other impulse noises within the waveform. It has three modes: Declick, Decrackle, and Interpolate. For click shorter than a few ms, use Declicker. For clicks longer than a few ms, use Spectral Repair.
RX TIP • Interpolate performs one-off manual click repairs that Declick is unable to treat. Interpolate can also be used to fix very slight audio dropouts. Clicks occurring in the analog domain, whether caused by the surface of a mechanical playback medium or a mouth, are sonically and visually different to digital clicks. The first image demonstrates a click that occurs during playback of a vinyl record. It’s a solid frequency event on the spectrogram, and can be considered random.
The RX 4 Declicker features both a Random and a Periodic mode that you can use to tackle these problems respectively. ANALOG CLICKS Step 1: Open Declick. Select the Declick tab, and then select M-band (random clicks) mode. “Click” will be the click type selected by default. If you believe that the click is being caused by a low-end thump, you should select Thump as your click mode instead.
Step 3: After choosing an acceptable level of click reduction, listen carefully to ensure you aren’t unintentionally harming any transients. Using a declicking tool on its highest settings can audibly soften transients. If you can hear that the transients are losing their edge, turn the strength of the declicker down a little. RX TIP • Use the Clicks Only feature in RX 4 to listen out for the isolated clicks. If you hear desirable audio creeping in, your settings are likely too harsh.
GENERAL TIPS FOR REMOVING CLICKS AND POPS 1. Two is better than one. As with denoising, there are scenarios where two processes of any declicker are better than one, particularly if you’re dealing with an old recording full of unwanted clicks. Performing one process of declicking will remove the most obvious clicks, allowing the second process of declicking to remove the quieter ones. 2. When is a click not a click? The simple answer is: when it’s a transient.
11: REMOVING CLIPPING Live concerts and on-location interviews frequently become victims of signal overload or clipping. This can happen in both the analog and digital domain, as well as during the A/D process. This is often the result of time constraints in setting up and sound checking.
It’s technically possible to obtain moderately good results by using the mouse to slowly redraw each sample—restoring it to what it might have been prior to clipping. Declipping tools, however, use more advanced techniques to intelligently redraw the waveform. This saves the audio engineer lots of time, and removes manual work. In RX 4, you can use the Declip module. It uses advanced interpolation techniques to intelligently rebuild the peaks of clipped audio.
USING DECLIP IN RX 4 Step 1: If you can actually see where the clipping occurs, Declip is both easier to use and more likely to succeed. Once you’ve identified exactly where the clipping occurs, you can open Declip in the RX 4 standalone application and set a suitable threshold. There are two ways you can do this: 1. Zoom in on the waveform and drag the threshold down until it sits just below the truncation. 2.
RX TIP • The height of the histogram represents the high and low levels of the audio, while the width of the bar indicates the most frequent levels. Here, a white line that covers the full width indicates that clipping is present. RX TIP • Clicking on the Suggest button will cause Declip to automatically place the threshold. This may save you some time and manual work. Step 2: Declipping tools redraw a truncated waveform.
Step 3: Occasionally the audio may have what’s called asymmetric clipping—a scenario where clipping occurs at a different level on one side of the waveform. When confronted with this scenario, click the linked icon to delink the two thresholds. This will allow you to set different threshold values for the positive and negative sides of the waveform, which offers the best sounding results when removing clipping. This feature is exclusive to RX 4 Advanced.
GENERAL TIPS FOR REMOVING CLIPPING 1. If your declipping tool does not work for you, you may be able to use RX 4’s Spectral Repair tool to fix short corrupted segments. See the Spectral Repair chapter for more details. 2. To avoid volume fluctuations you may want to process the entire audio file rather than specific sections.
12: REMOVING REVERB Reverb is used to add certain spatial characteristics to audio. It’s often used on vocals and instrumentation, but can be used on sound design elements too. It’s a relevant topic to discuss, even for audio repair. Technologies that are able to attenuate reverberations in an audio signal are not very common, and there are only a few effective tools that do this job well. RX TIP • RX 4 includes technology created by iZotope that reduces reverb.
PRINCIPLES OF REMOVING REVERB The above image shows a reverberant signal versus the same signal without reverb. The difference between the two is most obvious in the decay that follows the transient, known as the reverb tail. Technologies that seek to attenuate reverb often look to these tails in order to identify and distinguish the reverb. This is so that attempts can be made at reducing its effect.
USING DEREVERB IN RX 4 Please note that reverb reduction tools are very few in number, and you may find that the controls mentioned here are specific to RX 4 Dereverb. Step 1: Listen to your audio. Try and discern how long the reverb tail is, and then set the tail length accordingly. Using an incorrect tail length setting can cause more harm than good. RX TIP • The Learn feature may help here, as this will automatically set the band thresholds for you. For best results, learn from the entire audio file.
Step 2: RX 4’s Dereverb is multiband. There are four sliders allowing you to adjust the amount of reverb reduction in different areas of the frequency spectrum. If you used the Learn feature in Step 1, these sliders may have been initially positioned for you. Step 2 is all about making sure. Using the Solo function, listen to each band and adjust the slider until you’re happy with the amount of reverb being reduced.
GENERAL TIPS FOR REMOVING REVERB 1. Reverb is often centered in the mid range, and focusing your attention on the low and hi-mid multiband sliders will help ensure a good result. 2. Use a gentle pass of 2-4 dB of denoising before removing reverb and then doing some heavier denoising. A slight, gentle pass can help remove the top “silky” layer of noise in the audio signal, allowing reverb removing algorithms a better peek at the reverb underneath.
13: EXPORTING AND DELIVERING AUDIO Once you’ve completed the repair and restoration process, you’ll want to take the necessary steps to export and deliver your work. This might be as simple as rendering/bouncing your mix from a DAW, or saving an audio file that you can use elsewhere. But sometimes, particularly in the case of audio forensics or audio for TV and broadcast, delivery requirements might be somewhat demanding.
RX TIP • If any of the RX 4 plug-ins have been using a lesser quality setting for latency purposes (such as Low or Medium in Declip), change this to high before bouncing. When using the standalone application, you have several options at your disposal. Although these options are described within the context of RX 4, the principles of maintaining session data and multiple copies of your work still apply across any form of audio repair and restoration work. Step 1: Click on File in the upper menu bar.
• Export Selection allows you to export the audio you’ve highlighted, which is particularly useful if you tend to break apart larger projects. • Export Regions to Files allows you to export individual regions from one audio file as separate, unique audio files. Step 2: Decide on your output format. The file format options available to you are WAV(including BWF), AIFF, FLAC, OGG, and RX 4 document. Should you choose WAV or AIFF, RX 4 allows you to save in 32-bit floating point.
TIP • If you’re interested in learning more about the concept of dithering, iZotope has produced this informational video: http://youtu.be/vVNzylf9sGo. If you prefer to read, you can download iZotope’s free dithering guide at: http://www.izotope.com/ozone/guides. 2. Whatever your audio repair and restoration solution, make sure you document the steps you take. Keep copies of the original audio, save presets, and note as much information as possible about the process.
RX TIP • RX 4 Advanced includes iZotope Insight, a comprehensive metering suite that monitors and ensures loudness compliance. For more information, watch this useful video: http://youtu.be/XpmOLjDSo0c. And for more information on Insight, visit www.izotope.com/insight.
4. Ensure everything is the correct sample rate, as required for delivery. A project destined for CD would require a sample rate of 44.1 kHz, whereas the audio for a video project would typically require 48 kHz. This may involve either downsampling or upsampling your audio. Please note: upsampling audio does not enhance sonic quality. RX TIP • RX 4 includes iZotope’s Sample Rate Conversion (SRC) technology as part of the Resample module. It supports sample rates between 11.
14: SUMMARY We hope this guide has increased your knowledge of audio repair and restoration and, as a bonus, given you some ideas of how to use RX 4 effectively. When conducting audio repair and restoration, it’s important to understand that each noise problem is different. Learning the fundamentals, such as those outlined in this guide, will allow you to quickly and effectively hone in on which tool you need for the job, and how best to use it.
15: ABOUT THE AUTHORS In addition to the hard work of iZotope’s development, QA, and content development teams, we also wish to extend our thanks to Nat Johnson whose writing, restoration expertise, and audio examples contributed greatly to the original version of this guide published in 2008. Nat Johnson Nat Johnson started a career in radio broadcasting and sound recording after joining the Concert Network as an announcer and music programmer.
APPENDIX A: GETTING SET UP TO REPAIR AND RESTORE AUDIO EQUIPMENT Software iZotope RX 4 installs both the standalone audio editing application, and the separate plug-ins for use with a DAW. If you prefer working inside a DAW, you can load many of the RX 4 modules as plug-ins, provided it supports one or more of the following formats in 32- or 64-bit: VST, VST3, AU, and AAX. Spectral Repair only functions in DAWs that support offline processing, such as Avid’s Pro Tools 11.
TIP • Dual monitor setups allow you to place a spectrogram on one screen, and your audio processing modules on another. • RAM. If you work on large files, and have several audio or video editing programs running at once, a large amount of RAM is needed to cope with the load. • A backup hard drive.
that will help improve the sound of your room. Companies that make acoustic treatment typically have a large number of resources available, and these resources are designed to help you determine what your space requires. TRANSFERRING MEDIA TO THE COMPUTER When you sit down and start restoring an audio project, you might be lucky enough to have a digital audio file that has either already been digitized, or that began life as a digital format.
• The vast majority of turntables don’t output a line level signal. They usually require a phono preamp. Phono preamps not only boost the level of audio coming in from the turntable, but they also apply special equalization—a part of the RIAA standard for creating vinyl records. Without this gain and EQ stage, your transferred vinyl will likely sound bad. • Most audio amplifiers designed to work with turntables have phono preamps built in.
While we could write an entire book on these formats, it’s simply out of the scope of this guide. If this is an area you’re interested in, you’ll find numerous websites offering useful information and good tips on 78 RPM history, availability and playback equipment, as well as storage and general care advice. Magnetic Tape There are endless recordings out there made on cassette recorders and reel-to-reel machines.
APPENDIX B: GENERAL RX 4 TOOLS The following is a list of general RX 4 tools that are extremely useful for the workflow of the audio repair and restoration engineer. COMPARING SETTINGS The ability to easily, repeatedly and accurately compare settings—visually and aurally—with the RX 4 modules is a valuable tool and timesaver for every project you undertake. You’ll find the Compare Settings feature to be an extremely useful reference when contrasting a wide range of choices with the original.
BATCH PROCESSING RX 4 offers numerous time savers, and Batch Processing is one of its best. Simply defined, Batch Processing enables the automation of processing on file groups. If you have several files that need to be processed in the same way, you can use Batch Processing to define the set of steps (e.g. denoise, then declick, then normalize), as well as the output format and naming convention for each. RX 4 will then automatically process all of the files in the background.
Running a Batch Process Once you’re satisfied with the batch processing jobs, click Process to run them all. You’ll see a progress dialogue box while RX 4 runs each job. To cancel the current job and all subsequent jobs, click Cancel. WAVEFORM STATISTICS Waveform Statistics supplies the audio engineer with useful information about the peak, RMS, and loudness levels of the audio, as well as potential warnings about DC offsets and clipping. It’s accessible via View in the upper menu bar.
PRESETS Groups of settings in each of RX 4’s modules and plug-ins can be saved and recalled as presets. Once you’ve made modifications to the settings in a module, you can name the preset and save it with a custom filename. Moreover, you can easily export your presets as .XMLs and store them on your computer— perfect for backup, or to share with other RX 4 users. KEYBOARD SHORTCUTS Similar to presets, RX 4 includes default keyboard shortcuts.
APPENDIX C: REPAIRING THE INCLUDED AUDIO FILES Let’s take a look at some real world examples of audio problems and some tips and tricks to help go about fixing these common problems. To try out these fixes for yourself, download the example files. EXAMPLE 1: REMOVING BROADBAND NOISE FROM A CONCERT RECORDING Details: Player-organ recording in Boston: Vierne Organ Symphony No. 1- Allegro (STEREO). Originally recorded on Ampex 440-B, ¼” analog tape, zero noise reduction used during session.
and reducing them would perhaps affect the low-level organ harmonics as they fall into the noise floor. • The Reduction Curve has been enabled, and it reducing the high and low ends more harshly, since this is where the tape hiss is occurring. The curve is also being used to back off the mid range, where the organ is centered. • In RX 4 Advanced, you can see the Enhancement slider. Enhancement is being used to enhance signal harmonics that fall below the noise floor.
EXAMPLE 2: RESTORING AN HISTORICAL SPEECH: MAKING VOICE MORE INTELLIGIBLE Details: This file contains the voice of famed philanthropist Andrew Carnegie, recorded in 1914 (Internet source) (MONO). Comments: Some of the words in this speech, recorded by Carnegie in 1914, are somewhat difficult to decipher—owing to a pronounced Scottish accent, made even more muffled by someone’s rudimentary attempt at noise reduction. In addition, there’s a steady background noise present.
• Use Deconstruct to boost the tonal elements of the signal (voice) and reduce the variable noise/crackle. • Use the Spectral Denoiser to reduce steady-state background noise. TIP • Using an approach that incorporates both EQ and Denoiser can make buried dialogue much more intelligible.
EXAMPLE 3: CLEANING UP A PHONE INTERVIEW WITH DECLICK AND SPECTRAL REPAIR Details: Interview with Francis Ford Coppola, recorded directly to DAT from phone patch (MONO). Comments: The problems were caused mainly by mechanical contact with the telephone handset. For instance, at 11.62 seconds, where Coppola says the words “motion picture business,” a loud click is audible. Goals: • Use Declick’s Interpolate mode to remove the full bandwidth handset noises.
• Use Spectral Repair’s Replace mode to remove the limited bandwidth handset noises. • Use Spectral Repair’s Attenuate to remove the low frequency thumps.
• Use the Dialogue Denoiser to reduce the level of background noise in the dialogue. RX TIP • Each of the phone handset noises is comprised of a vertical spike and a horizontal thump. Removing the vertical spikes with Declick’s Interpolate mode still results in an audible thump. A quick pass with Attenuate mode will remove the low frequency sound.
EXAMPLE 4: REMOVING CLICKS AND POPS FROM A CONCERT ON RECORD Details: Kings College Choir, Cambridge, England. Source: 33 1/3 RPM LP (1965, STEREO). Comments: This vinyl disc was transferred for playback as a CD. Fortunately, the only problems encountered were mostly light vinyl clicks and heavy room ambience—a result of the massive chapel at Cambridge University being the recording venue. Declick was employed using the M-band (random clicks) algorithm and one or two spots in manual mode.
• Use the Deconstruct module to reduce the variable noise components while adding a slight lift to the tonal noise components. • Use the Spectral Denoiser to reduce background noise. Try to retain the musical character of the performance while removing as much of the offending noise as possible. • Use Spectral Repair to reduce the low traffic hum that creeps in occasionally.
EXAMPLE 5: REMOVING CLIPPING FROM A PHONE INTERVIEW DETAILS: This phone interview, from an iZotope Podcast with producer Morgan Page, was recorded with a phone line breakout box and a USB audio interface. Comments: This is a classic example of input gain being set too high, resulting in some nasty clipping. In this example, the distorted sections can be clearly heard, but viewing them with the spectrogram display and waveform overlay highlights them in clearer detail.
EXAMPLE 6: REMOVING GUITAR STRING SQUEAKS WITH SPECTRAL REPAIR Details: This recording by guitarist Jamie Robertson contains a few loud squeaks—the result of sliding one’s hand up and down the fretboard. Comments: These situations are always tricky.
APPENDIX D: TIPS FROM THE PROS We asked a few of our trusted friends and “RXperts” to contribute their advice with regards to audio repair, restoration, and editing. BILL JACKSON One of the tricks I use with RX is to carefully Denoise using the D (best) algorithm, then resample and Denoise using the C algorithm. This helps when there is broadband noise as well as a high frequency hiss that you want to reduce.
Another hidden gem in RX is the 24/16 bit conversion. This must be one of the best available—the BBC require 16 bit masters—so I use this. Sometimes in our radio show we have to use 78 shellac discs. The declicking in RX can make unusable discs perfect for the program. CHRIS SHAW I find that Spectral Repair is really good at taking the edge off vocals that are recorded a bit too hot or when a singer’s voice is a bit too shrill as he/she hits the top of their range.
The Declick tool is not just a one trick pony for cleaning up vinyl recording. Its great for removing digital clicks caused by clocking problems, mouth clicks on voiceovers, and some kinds of distortion can be cleaned up with the Declick tool. I have just taken a recording made in a bedroom with very little soft furnishings, and actually had a ping to it. Ran it through Dereverb and was able to get a result close to as if it had been recorded in a radio studio.
Copyright © 2014 iZotope, Inc. All rights reserved. iZotope®, RX®, Ozone®, Insight®, Nectar®, Alloy®, MBIT+™ and the iZotope logo are trademarks or registered trademarks of iZotope, Inc. All other trademarks contained herein are the property of their respective owners. Other products and company names mentioned herein may be trademarks of their respective companies. This material is provided for information purposes only; iZotope, Inc. assumes no liability related to its use.
