Adobe recommends |
| ||
 | Introduction to Noise Reduction and Advanced Noise ReductionColin SmithAutomatically remove clicks, restore clipped audio, and heal individual artifacts. Or clean up complex noise with two powerful effects. | ||
Techniques for restoring audio
You can fix a wide array of audio problems by combining two powerful features. First, use Spectral Display to visually identify and select ranges of noise or individual artifacts. (See Select spectral ranges and Select artifacts and repair them automatically.) Then, use either Diagnostic or Noise Reduction effects to fix problems like the following:
Crackle from wireless microphones or old vinyl records. (See Automatic Click Remover effect.)
Background noise like wind rumble, tape hiss, or power-line hum. (See Adaptive Noise Reduction effect and DeHummer effect.)
Phase cancellation from poorly placed stereo microphones or misaligned tape machines. (See Automatic Phase Correction effect.)
The real-time restoration effects above, which
are available in both the Waveform and Multitrack editors, quickly
address common audio problems. For unusually noisy audio, however,
consider using offline, process effects unique to the Waveform Editor,
such as Hiss Reduction and Noise Reduction.
- A.
- Hiss
- B.
- Crackle
- C.
- Rumble
Noise Reduction effect (Waveform Editor only)
The Noise Reduction/Restoration > Noise Reduction effect dramatically reduces background and broadband noise with a minimal reduction in signal quality. This effect can remove a combination of noise, including tape hiss, microphone background noise, power-line hum, or any noise that is constant throughout a waveform.
The proper amount of noise reduction depends upon the type of background noise and the acceptable loss in quality for the remaining signal. In general, you can increase the signal‑to‑noise ratio by 5 to 20 dB and retain high audio quality.
To achieve the best results with the Noise Reduction effect, apply it to audio with no DC offset. With a DC offset, this effect may introduce clicks in quiet passages. (To remove a DC offset, choose Favorites > Repair DC Offset.)
- A.
- Drag control points to vary reduction in different frequency ranges
- B.
- Low amplitude noise.
- C.
- High amplitude noise
- D.
- Threshold below which noise reduction occurs.
Apply the Noise Reduction effect
- In the Waveform Editor, select a range
that contains only noise and is at least half a second long. To select noise in a specific frequency
range, use the Marquee Selection tool. (See Select spectral ranges.)
When recording in noisy environments,
record a few seconds of representative background noise that can
be used as a noise print later on.Noise Reduction options
- Capture Noise Print
- Extracts a noise profile from a selected range, indicating only
background noise. Adobe Audition gathers statistical information
about the background noise so it can remove it from the remainder
of the waveform.If the selected range is
too short, Capture Noise Print is disabled. Reduce the FFT Size
or select a longer range of noise. If you can’t find a longer range,
copy and paste the currently selected range to create one. (You
can later remove the pasted noise by using the Edit >
Delete command.)
- Save the Current Noise Print
- Saves the noise print as an .fft file, which contains information about sample type, FFT (Fast Fourier Transform) size, and three sets of FFT coefficients: one for the lowest amount of noise found, one for the highest amount, and one for the power average.
- Load a Noise Print from Disk
- Opens any noise print previously saved from Adobe Audition
in FFT format. However, you can apply noise prints only to identical
sample types. (For example, you can’t apply a 22 kHz mono profile
to 44kHz stereo samples.)Note: Because noise prints are so specific, a print for one type of noise won’t produce good results with other types. If you regularly remove similar noise, however, a saved profile can greatly increase efficiency.
- Graph
- Depicts frequency along the x‑axis (horizontal)
and the amount of noise reduction along the y‑axis
(vertical).
The blue control curve sets the amount of noise reduction in different frequency ranges. For example, if you need noise reduction only in the higher frequencies, adjust the control curve downward to the right of the graph.
If you click the Reset button
to
flatten the control curve, the amount of noise reduction is based
entirely on the noise print.To better focus
on the noise floor, click the menu button 
to
the upper right of the graph, and deselect Show Control Curve and
Show Tooltip Over Graph. - Noise Floor
- High shows the highest amplitude of detected noise at each frequency;
Low shows the lowest amplitude. Threshold shows the amplitude below
which noise reduction occurs. The three elements
of the noise floor can overlap in the graph. To better distinguish
them, click the menu button , and
select options from the Show Noise Floor menu.
- Scale
- Determines how frequencies are arranged along the horizontal x‑axis:
For finer control over low frequencies, select Logarithmic. A logarithmic scale more closely resembles how people hear sound.
For detailed, high‑frequency work with evenly spaced intervals in frequency, select Linear.
- Channel
- Displays the selected channel in the graph. The amount of noise reduction is always the same for all channels.
- Select Entire File
- Lets you apply a captured noise print to the entire file.
- Noise Reduction
- Controls the percentage of noise reduction in the output signal. Fine-tune this setting while previewing audio to achieve maximum noise reduction with minimum artifacts. (Excessively high noise reduction levels can sometimes cause audio to sound flanged or out-of-phase.)
- Reduce By
- Determines the amplitude reduction of detected noise. Values between 6 and 30 dB work well. To reduce bubbly artifacts, enter lower values.
- Output Noise Only
- Previews only noise so you determine if the effect is removing any desirable audio.
- Advanced settings
- Click the triangle to display the following options:
- Spectral Decay Rate
- Specifies the percentage of frequencies processed when audio falls below the noise floor. Fine‑tuning this percentage allows greater noise reduction with fewer artifacts. Values of 40% to 75% work best. Below those values, bubbly‑sounding artifacts are often heard; above those values, excessive noise typically remains.
- Smoothing
- Takes into account the variance of the noise signal in each frequency band. Bands that vary greatly when analyzed (such as white noise) will be smoothed differently than constant bands (like 60-Hz hum). In general, increasing the smoothing amount (up to 2 or so) reduces burbly background artifacts at the expense of raising the overall background broadband noise level.
- Precision Factor
- Controls changes in amplitude. Values of 5-10 work best, and odd numbers are ideal for symmetrical processing. With values of 3 or less, the Fast Fourier transform is performed in giant blocks, and between them drops or spikes in volume can occur. Values beyond 10 cause no noticeable change in quality, but they increase processing time.
- Transition Width
- Determines the amplitude range between noise and desirable audio. For example, a width of zero applies a sharp, noise gate to each frequency band. Audio just above the threshold remains; audio just below is truncated to silence. Alternatively, you can specify a range over which the audio fades to silence based upon the input level. For example, if the transition width is 10 dB, and the noise level for the band is ‑60 dB, audio at ‑60 dB stays the same, audio at ‑62 dB is reduced slightly, and audio at ‑70 dB is removed entirely.
- FFT Size
- Determines how many individual frequency bands are analyzed.
This option causes the most drastic changes in quality. The noise
in each frequency band is treated separately, so with more bands,
noise is removed with finer frequency detail. Good settings range
from 4096 to 8192.
Fast Fourier Transform size determines the tradeoff between frequency- and time-accuracy. Higher FFT sizes might cause swooshing or reverberant artifacts, but they very accurately remove noise frequencies. Lower FFT sizes result in better time response (less swooshing before cymbal hits, for example), but they can produce poorer frequency resolution, creating hollow or flanged sounds.
- Noise Print Snapshots
- Determines how many snapshots of noise to include in the
captured profile. A value of 4000 is optimal for producing accurate
data.
Very small values greatly affect the quality of the various noise reduction levels. With more snapshots, a noise reduction level of 100 will likely cut out more noise, but also cut out more original signal. However, a low noise reduction level with more snapshots will also cut out more noise, but likely retain the intended signal.
Adaptive Noise Reduction effect
The Noise Reduction/Restoration > Adaptive Noise Reduction effect quickly removes variable broadband noise such as background sounds, rumble, and wind. Because this effect operates in real time, you can combine it with other effects in the Effects Rack and apply it in the Multitrack Editor. By contrast, the standard Noise Reduction effect is available only as an offline process in the Waveform Editor. That effect, however, is sometimes more effective at removing constant noise, such as hiss or hum.
For best results, apply Adaptive Noise Reduction to selections that begin with noise followed by desirable audio. The effect identifies noise based on the first few seconds of audio.
- Reduce Noise By
- Determines the level of noise reduction. Values between 6 and 30 dB work well. To reduce bubbly background effects, enter lower values.
- Noisiness
- Indicates the percentage of original audio that contains noise.
- Fine Tune Noise Floor
- Manually adjusts the noise floor above or below the automatically calculated floor.
- Signal Threshold
- Manually adjusts the threshold of desirable audio above or below the automatically calculated threshold.
- Spectral Decay Rate
- Determines how quickly noise processing drops by 60 decibels. Fine‑tuning this setting allows greater noise reduction with fewer artifacts. Values that are too short create bubbly sounds; values that are too long create a reverb effect.
- Broadband Preservation
- Retains desirable audio in specified frequency bands between found artifacts. A setting of 100 Hz, for example, ensures that no audio is removed 100 Hz above or below found artifacts. Lower settings remove more noise but may introduce audible processing.
- FFT Size
- Determines how many individual frequency bands are analyzed. Choose a high setting to increase frequency resolution; choose a low setting to increase time resolution. High settings work well for artifacts of long duration (like squeaks or power-line hum), while low settings better address transient artifacts (like clicks and pops).
- High Quality Mode
- Performs slower processing but achieves superior results.
Automatic Click Remover effect
To quickly remove crackle and static from vinyl recordings, use the Noise Reduction/Restoration > Automatic Click Remover effect. You can correct a large area of audio or a single click or pop.
This effect provides the same options as the DeClicker effect, which lets you choose which detected clicks to address (see DeClicker options). However, because the Automatic Click Remover operates in real time, you can combine it with other effects in the Effects Rack and apply it in the Multitrack Editor. The Automatic Click Remover effect also applies multiple scan and repair passes automatically; to achieve the same level of click reduction with the DeClicker, you must manually apply it multiple times.
- Threshold
- Determines sensitivity to noise. Lower settings detect more clicks and pops but may include audio you wish to retain. Settings range from 1 to 100; the default is 30.
- Complexity
- Indicates the complexity of noise. Higher settings apply more processing but can degrade audio quality. Settings range from 1 to 100; the default is 16.
Automatic Phase Correction effect
The Noise Reduction/Restoration > Automatic Phase Correction effect addresses azimuth errors from misaligned tape heads, stereo smearing from incorrect microphone placement, and many other phase-related problems.
- Global Time Shift
- Activates the Left and Right Channel Shift sliders, which let you apply a uniform phase shift to all selected audio.
- Auto Align Channels and Auto Center Panning
- Align phase and panning for a series of discrete time intervals,
which you specify using the following options:
- Time Resolution
- Specifies the number of milliseconds in each processed interval. Smaller values increase accuracy; larger ones increase performance.
- Responsiveness
- Determines overall processing speed. Slow settings increase accuracy; fast settings increase performance.
- Channel
- Specifies the channels phase correction will be applied to.
- Analysis Size
- Specifies the number of samples in each analyzed unit of audio.
For
the most precise, effective phase correction, use the Auto Align
Channels option. Enable the Global Time Shift sliders only if you
are confident that a uniform adjustment is necessary, or if you
want to manually animate phase correction in the Multitrack Editor.DeHummer effect
The Noise Reduction/Restoration > DeHummer effect removes narrow frequency bands and their harmonics. The most common application addresses power line hum from lighting and electronics. But the DeHummer can also apply a notch filter that removes an overly resonant frequency from source audio.
To quickly address typical audio problems, choose
an option from the Presets menu.- Frequency
- Sets the root frequency of the hum. If you’re unsure of the
precise frequency, drag this setting back and forth while previewing
audio.To visually adjust root frequency and
gain, drag directly in the graph.
- Q
- Sets the width of the root frequency and harmonics above. Higher values affect a narrower range of frequencies, and lower values affect a wider range.
- Gain
- Determines the amount of hum attenuation.
- Number of Harmonics
- Specifies how many harmonic frequencies to affect.
- Harmonic Slope
- Changes the attenuation ratio for harmonic frequencies.
- Output Hum Only
- Lets you preview removed hum to determine if it contains any desirable audio.
Hiss Reduction effect (Waveform Editor only)
The Noise Reduction/Restoration > Hiss Reduction effect reduces hiss from sources such as audio cassettes, vinyl records, or microphone preamps. This effect greatly lowers the amplitude of a frequency range if it falls below an amplitude threshold called the noise floor. Audio in frequency ranges that are louder than the threshold remain untouched. If audio has a consistent level of background hiss, that hiss can be removed completely.
To reduce other types of noise that have a wide
frequency range, try the Noise Reduction effect. (See Noise Reduction effect (Waveform Editor only).)
- Capture Noise Floor
- Graphs an estimate of the noise floor. The estimate is used by
the Hiss Reduction effect to more effectively remove only hiss while
leaving regular audio untouched. This option is the most powerful
feature of Hiss Reduction.
To create a graph that most accurately reflects the noise floor, click Get Noise Floor with a selection of audio that contains only hiss. Or, select an area that has the least amount of desirable audio, in addition to the least amount of high frequency information. (In the spectral display, look for an area without any activity in the top 75% of the display.)
After you capture the noise floor, you might need to lower the control points on the left (representing the lower frequencies) to make the graph as flat as possible. If music is present at any frequency, the control points around that frequency will be higher than they should be.
- Graph
- Represents the estimated noise floor for each frequency in
the source audio, with frequency along the horizontal ruler (x‑axis)
and the amplitude of the noise floor along the vertical ruler (y‑axis).
This information helps you distinguish hiss from desirable audio
data.
The actual value used to perform hiss reduction is a combination of the graph and the Noise Floor slider, which shifts the estimated noise floor reading up or down for fine tuning.
To disable tooltips for frequency and amplitude,
click the menu button to the
upper right of the graph, and deselect Show Tooltip Over Graph. - Scale
- Determines how frequencies are arranged along the horizontal x‑axis:
For finer control over low frequencies, select Logarithmic. A logarithmic scale more closely resembles how people hear sound.
For detailed, high‑frequency work with evenly spaced intervals in frequency, select Linear.
- Channel
- Displays the selected audio channel in the graph.
- Reset
- Resets the estimated noise floor. To reset the floor higher
or lower, click the menu button to
the upper right of the graph, and choose an option from the Reset
Control Curve menu.For quick, general‑purpose
hiss reduction, a complete noise floor graph isn’t always necessary.
In many cases, you can simply reset the graph to an even level and manipulate
the Noise Floor slider.
- Noise Floor
- Fine‑tunes the noise floor until the appropriate level of hiss reduction and quality is achieved.
- Reduce By
- Sets the level of hiss reduction for audio below the noise floor. With higher values (especially above 20 dB) dramatic hiss reduction can be achieved, but the remaining audio might become distorted. With lower values, not as much noise is removed, and the original audio signal stays relatively undisturbed.
- Output Hiss Only
- Lets you preview only hiss to determine if the effect is removing any desirable audio.
- Advanced settings
- Click the triangle to display these options:
- Spectral Decay Rate
- When audio is encountered above the estimated noise floor,
determines how much audio in surrounding frequencies is assumed
to follow. With low values, less audio is assumed to follow, and
hiss reduction will cut more closely to the frequencies being kept.
Values of 40% to 75% work best. If the value is too high (above 90%), unnaturally long tails and reverbs might be heard. If the value is too low, background bubbly effects might be heard, and music might sound artificial.
- Precision Factor
- Determines the time-accuracy of hiss reduction. Typical values range from 7 to 14. Lower values might result in a few milliseconds of hiss before and after louder parts of audio. Larger values generally produce better results and slower processing speeds. Values over 20 don’t ordinarily improve quality any further.
- Transition Width
- Produces a slow transition in hiss reduction instead of an abrupt change. Values from 5 to 10 usually achieve good results. If the value is too high, some hiss may remain after processing. If the value is too low, background artifacts might be heard.
- FFT Size
- Specifies a Fast Fourier Transform size, which determines
the tradeoff between frequency- and time-accuracy. In general, sizes
from 2048 to 8192 work best.
Lower FFT sizes (2048 and below) result in better time response (less swooshing before cymbal hits, for example), but they can produce poorer frequency resolution, creating hollow or flanged sounds.
Higher FFT sizes (8192 and above) might cause swooshing, reverb, and drawn out background tones, but they produce very accurate frequency resolution.
- Control Points
- Specifies the number of points added to the graph when you click Capture Noise Floor.
Edit audio clips from Adobe Premiere Pro CS5.5 or After Effects
Control effect settings with graphs
