How to Digitally Archive and Share Historical Photographs, Documents, and Audio Recordings

Chapter 7. Preparation and Presentation of Historical Audio Recordings

Key Points.

·        Audio recordings in a historical archive usually need to be adapted for particular uses.

·        The steps that may be needed to prepare a recording for presentation are listed in order below, but not all steps are used with all recordings.

o   Make a working copy of the archival recording

o   Split or extract relevant segments from the working copies

o   Convert stereo recording to mono

o   Set the working volume

o   Implement broadband noise reduction

o   Implement discrete noise reduction

o   Make final frequency adjustments

o   Adjust the final volume

o   Set output format

o   Set output file name and location

·        A noise profile can be used to reduce noise during speech or useful sound, and a dynamic processor can reduce or eliminate noise during quiet periods without useful sound.



Audio recordings can be shared in many ways. People enjoy playing historical audio files on a computer or other playing device. The audio files can be incorporated into slideshows that play on a computer or TV. Electronic picture frames are increasingly common and can play audio files while the photographs are displayed. Personal web sites and social networking sites can be used to widely share audio recordings that can be played on an increasing number of devices. This sharing of historical information is the ultimate goal in developing a historical archive.

For most types of presentation, optimization or adaptation of archival recordings is needed. The amount of adaptation depends on the characteristics of the recording, the type of presentation, and the degree of enhancement that is considered appropriate for the project. As with photographs, people expect imperfections in historical recordings and any modification that alters the content or meaning of the recording compromises the historical integrity. Here too, minimal enhancement is not only acceptable; it is the safest practice. At the same time, defects that are distracting and can be reduced without compromising the historical integrity of the recording can be usefully corrected.

High quality recordings may need little more than a minor volume adjustment and perhaps conversion to a different format. On the other hand, informal recordings often have background noise that detracts from the recording and would benefit from significant revision. Removing the background noise can be valuable when the informal recording is the only record of an important story or event. The more advanced methods for noise correction described in this Chapter may be useful in these cases, but cannot be expected to match a high quality recording. These advanced methods will not be needed with good quality recordings.

Steps that may be needed are discussed below in the approximate order that they would be carried out. Many recordings will not need all of these steps. Relevant audio editing software is described first.

7.1 Audio Editing Software

A variety of software programs are available for editing audio files. These programs vary widely in price and intended purpose. Some focus on music creation and production while others focus more on speech and sound effects. Those intended primarily for music production have many effects and processes that are unnecessary for and can complicate the editing of typical historical recordings. Most of the audio editing programs offer free trial periods that are very useful in evaluating the software. The discussion here focuses on Nero WaveEditor and Adobe Audition.

Nero WaveEditor

Nero WaveEditor is part of the Nero suite of multimedia applications. It is the program I first used when working with audio files and my recent investigation of alternatives found that it was the only inexpensive (about $70 for the full suite) application that had basic features that I consider essential. One essential feature is a preview mode that allows any modification to be toggled on and off while the recording is playing. This allows the modification to be evaluated prior to implementing it. Other essential features include a dynamic processor and broadband noise reduction (described later).

Accessing the Nero WaveEditor through the Nero front end can be cumbersome. I set a desktop shortcut that starts the WaveEditor and also have the WaveEditor appear as an “Open with” option when the right mouse button is clicked on a wave file. On my computer with Windows 7 as the operating system the path to the WaveEditor is "C:\Program Files (x86)\Nero\Nero 9\Nero WaveEditor\WaveEdit.exe". This path is used to make the desktop shortcut and used once with the right mouse button “Open with” command to provide “Wave Editor” as an option for opening wave files. The specific version used for the discussion here is Nero WaveEditor 9.

Modifications to audio recordings are initiated using the main dropdown menus. A toolbar below the dropdown menus has buttons to cut, paste, and play the recording. Modifications are previewed by clicking the green play button in the lower left of a dialog box. Changes made to the settings during playback are included in the playback. Clicking Bypass plays the recording without the modification. Bypass can be toggled on and off to evaluate the effects of the modification. Modifications are applied to the recording by clicking an Ok button on the dialog box for the modification. Each dialog box has a Process Offline tab and it is best to click this before clicking Ok. If Process Offline is not clicked, the modification will not be fully applied until the recording is output to a file. This can produce misleading results if other modifications are being implemented.

One useful setting that is not the default for WaveEditor is to display a list of modifications that have been made. This is activated by clicking the dropdown menu item View> Edit History.

Adobe Audition

Adobe Audition is a more advanced audio editing application. The recent version CS5.5 (about $350) has certain features that are very useful and were not available in other similarly priced programs that I evaluated. The workspace has a column with three panels on the left. The top panel lists each audio file or audio track that is open. The active audio track for editing is selected by clicking the relevant row in this panel.

The bottom panel on the left is the History panel that has a row for each modification in sequence. As the recording is played, any history row can be clicked and the state at that point in the history of changes is played. Among other things, this allows the cumulative effect of multiple changes to be easily evaluated. The usual preview for a modification allows evaluation of only one change.

The middle panel on the left is an Effects Rack that allows some modifications to be handled with layers. Clicking the triangle icon on the right of a row brings up a list of modification that can be made as layers. Once applied, a modification can be toggled on or off by clicking the green button on the left side of the row in the Effects Rack and the settings can be revised by double-clicking on the row. The slider for Mix: Dry – Wet on the lower part of the Effects Rack panel controls the degree to which the modifications are blended with the previous unmodified state. Wet means the changes are fully implemented and Dry means the changes are not implemented at all. This blending can be particularly useful with noise reduction methods.

The modifications in the Effects Rack are not applied to the recording for output until the Apply tab is clicked at the bottom of the Effects Rack. The history panel can be used to return to a state before the Apply tab was clicked.

The Effects Rack works well for effects that apply to the entire recording, but effects that apply to selected sections are often better handled without these layers.

One default setting that most users will want to modify is to have playback return to a selected point when playback is stopped rather than continue from where playback stopped. This setting is modified with the dropdown menu Edit> Preferences >Playback. Check the box for “Return CTI to start position on stop.”

Standard Features

Nero WaveEditor and Adobe Audition, like most audio editing software, have certain features in common. The basic display is a waveform that has time on the horizontal axis and sound volume on the vertical axis. The waveform expands both up and down for loud sounds and becomes a narrow line as loudness decreases. Figure 7.1 shows an example of a waveform display of a man talking. The resolution on the time axis can be expanded or reduced with the mouse scroll wheel. Playback can be toggled on and off by tapping the spacebar. Clicking a point on the time axis makes playback start at that point. A section of the recording is selected by dragging the mouse horizontally across time or with the Shift-arrow keys (and Ctrl-Shift-arrow keys for Adobe Audition). When a selection has been made, playback occurs only for the selection and any modifications are applied only in the selection. If the recording is in stereo, a waveform is displayed for each of the two channels.





Audio editing waveform.

Figure 7.1. Waveform display for a man speaking. The horizontal axis is time. The vertical axis is loudness, with loud sounds expanding both up and down. The times when the waveform is narrow are quiet. This waveform is from Adobe Audition.




For both WaveEditor and Audition the settings for a modification can be saved as a preset and reused. In WaveEditor, type a name for the new preset in the field with the pick-list in the lower right of a modification dialog box, then click the red plus button to add the new name to the pick-list. In Adobe Audition, the Presets pick-lists are in the upper left of the dialog boxes and the button to add a present is just to the right of this field. Once the button is clicked, another box comes up asking for the name of the new preset.

The loudness of sound is measured in decibels, which is related to how humans hear sound. Decibels (dB) are a ratio between sounds and therefore compare one sound level to another. A positive number for decibels means a sound is louder than the reference sound and a negative number means the sound is less loud. The basic reference for loudness in audio editing software is the maximum loudness that can be recorded without clipping and distortion. This maximum loudness is set as zero decibels and the loudness of the recording at any point in time is a negative number that indicates how far the current sound is below the maximum. Decibels are also used with other reference points. For example, loudness is often adjusted with a slider that has zero decibels for the current loudness and can be made positive or negative to increase or decrease the loudness.

7.2 Make a working copy of the archival recording

As with images, the best practice for audio recordings is to make working copies and use them for any adjustments or presentations. The safest procedure is to copy an archival recording to a working folder before making modifications. This greatly reduces the possibility that the archival recording will be unintentionally modified. A file naming convention is also a good practice for the different stages of developing presentation recordings. For example, after a recording has been prepared for use in a slide show the recording could be saved to the working folder with _s added to the file name.

7.3 Split or extract relevant segments from the working copies

Often only part of a recording will be used for a presentation. Sections can be selected and deleted using standard commands for Windows (e.g., Edit> Delete). Or, a section can be selected and copied to a new window (File> New).

7.4 Convert stereo recording to mono

A stereo recording of speech may optionally be converted to mono to simplify processing and make the output file smaller. As noted above, mono recording is adequate for human speech. If the original recording was made with a mono microphone, each stereo channel will be the same and one of the stereo channels can be used to create the mono channel. This does not involve conversion or re-sampling the recording. If a stereo microphone was used, then conversion and re-sampling are needed.

Nero WaveEditor converts stereo recordings to mono by combining the right and left channels. Select the dropdown menu Edit> Convert Sample Format. In the dialog box, click the Mono button and set the output sample rate and bit depth (e.g., 44,100 Hz and 16 bits for CD quality). Set the “Anti aliasing filters” to “Accurate anti aliasing filter.” Click Ok to do the conversion. For some recorders, a mono microphone records to only the left or right stereo channel. In these cases it may be slightly preferable to copy the recorded channel to the other channel before making the conversion to mono. For example, the left channel can be copied to the right channel with the dropdown menu Tools> Stereo Processor. For the Right Out group of sliders, set the Left In slider to 100% and the Right In slider to 0%. Then click Process offline and Ok.

Adobe Audition has options to create a mono recording from one stereo channel. This method is optimal when the recording was made with a mono microphone, whether one or both stereo channels were used. Select the dropdown menu Edit> Extract Channels to Mono Files. This makes a separate mono recording for each stereo channel. The new recordings have the same name as the original stereo file, but with _L and _R added for the left and right channel. The new recordings are listed in the Files panel in the upper left of the screen. Open a new recording by double-clicking on it and save it as a file with the usual File> Save As. To close one of the new recordings without saving it, right-click on it in the Files panel and select Close Selected Files.

If the original recording was with a stereo microphone, the stereo channels are combined into a mono recording in Adobe Audition using the dropdown menu Edit> Convert Sample Type. In the Presets pick-list at the top of the dialog box, select “Convert to Mono (Average).” Click Ok to do the conversion.

7.5 Set the working volume

Working with audio recordings is usually easiest if the volume is initially set with the peak or loudest sound at ‑3 decibels. This allows room for increasing volume at certain places if needed. Settings of -6 decibels or -12 decibels would be appropriate if significant adjustments to volume are expected. Any loud artifacts such as from setting up the equipment should be deleted before the volume is set.

In Nero WaveEditor, tap Ctrl-A to select the entire recording and then click the dropdown menu Volume> Normalize. Set the slider to the desired peak level and click Process Offline and then Ok.

In Adobe Audition, tap Ctrl-A to select the entire recording and then click the dropdown menu Effects> Amplitude and Compression> Normalize (process). Check the box for Normalize, set the value to the desired number, and select the button for dB. Checking the box for DC Bias Adjust and setting the value to zero is also useful.

7.6 Broadband Noise Reduction

Broadband noise is created by sources such air conditioners, refrigerators, heating systems, traffic, and wind. The noise is continuous throughout the recording or in significant sections of the recording. This type of noise is difficult to remove because it has a broad range of sound frequencies, including frequencies that overlap with human speech.

Attempts to remove broadband noise can cause artifacts that are described as bubbly or watery, and also cause increased reverberation that sounds like the recording was made in a well. The methods for correcting broadband noise are continually improving and low to moderate amounts of noise can be handled surprisingly well by the more sophisticated programs.

Broadband noise reduction should be one of the first audio processing steps because other adjustments can make the noise fluctuate.  Noise that fluctuates is much more difficult to remove. Broadband noise is most apparent during the quiet periods when there are not useful sounds such as a person speaking.

The basic strategy for reducing broadband noise is to build a noise profile from a section of the recording that has only the background noise with no talking or useful sounds. The noise profile is used to reduce noise throughout the recording or in a selected section. The more sophisticated programs have additional controls to reduce artifacts.

One of the most effective strategies is to use a noise profile to make the noise unnoticeable during useful sounds and use a dynamic processor to reduce the remaining noise during quiet periods.

A dynamic processor has a graph that specifies the output loudness for a given input loudness. The horizontal axis of the graph has the input loudness ranging from low to high and the vertical axis has the corresponding output loudness. A diagonal line indicates no change in loudness. Moving the line down in the region where the input loudness is low causes a reduction of sound only during quiet periods, such as when the only sound is background noise. Complete removal of noise may not be optimal. Keeping some of the noise may provide a more realistic ambiance for a room.

Broadband noise reduction with noise profiles is often effective at making noise unnoticeable during useful sounds, but leaves noticeable noise during quiet periods. Attempting to use a noise profile to reduce noise during quiet periods can produce artifacts. A dynamic processor is very effective at reducing noise during quiet periods without producing artifacts.

Broadband Noise Reduction with Nero WaveEditor

In Nero WaveEditor, select a section that has only noise and is one-half second or longer. Then click the dropdown menu Enhancement> Noise Analysis. This creates the noise profile. Select the area where the noise reduction will be applied. Usually this will be the full recording, which can be selected with Ctrl-A. The reduction is applied with the dropdown menu Enhancement> Noise Reduction. There are two sliders for adjusting noise reduction. For recordings with high noise, a good starting point is to set the top slider for Floor Gain to ‑12 and the bottom slider for Reduction Level to 100%. For lower noise recordings, the settings can be Floor Gain ‑8 and Reduction Level 70%. My experience with high noise recordings has been that reducing the Reduction Level (bottom) slider to less than 100% does not improve sound quality, and the top slider works better for controlling the degree of noise reduction. The top slider should be set as far to the right as possible while providing adequate noise reduction.

In some cases, applying noise reduction two or more times with weak settings such as ‑6 may provide better results than one application at high settings. The best strategy is often to handle the final noise with the dynamic processor as described below. Clicking the Residual Output button during playback plays the sounds that are being removed, which can be used to determine whether useful sound is being removed. Click Process Offline and then Ok to apply the noise reduction.

Nero WaveEditor does not have advanced controls to reduce artifacts and often leaves bubbly or watery sounds during quiet periods. As noted above, these can be eliminated with the dynamic processor. Initiate the dynamic processor with the dropdown menu Tools> Dynamic Processor. In WaveEditor, the horizontal and vertical axes for the dynamic processor range from ‑90 decibels on the lower left to 0 decibels on the upper right, where 0 is the maximum loudness without clipping and ‑90 is extremely low sound. A diagonal line indicates no change in loudness and is the default.

Begin noise reduction with the dynamic processor by selecting a section of the recording that has only noise. Then activate the dynamic processor. If the graph is not a straight diagonal line from lower left to upper right, select Default on the pick-list in the lower right of the dialog box. Click on the green play button, which will start playing the noise in a loop. Click on the diagonal line at a point that is about ‑24 decibels on the horizontal axis. This makes a control point that prevents changes to the upper part of the line. Then click another point that is about 10 decibels lower on the line (e.g., -34 decibels) and slowly drag the line down until the noise just goes away or is at a level that retains the ambiance of the room. This reduces loudness for sounds that are below ‑24 decibels.

Now the dynamic processor needs to be closed in order to select the larger area for the application of the processor settings; however, the current settings also need to be saved. One way to do that is to click the Ok button to apply the processor to the currently selected noise area. Then tap Ctrl-Z to undo that processing. Make a selection for the application of the dynamic processor—usually the entire recording— and then initiate the dynamic processor again. The settings from before will be retained. Click the green play button to listen to the results of the processor and make adjustments as needed. If words sound cut off, setting the slider for Attack Time to the lowest value and the slider for Release Time to 700 ms may be an improvement. Moving the first set point farther down the diagonal line (e.g., ‑30 decibels) may (or may not) improve the noise reduction. After the settings are finished, click Process Offline and then Ok. In theory a noise gate tool can produce similar results, but with WaveEditor the dynamic processor gives better control and results.

Broad Band Noise Reduction with Adobe Audition

To apply noise reduction in Adobe Audition, select a section that has only noise and is one-half second or longer. Then tap Shift-Ctrl-P or click the dropdown menu Effects> Noise Reduction/Restoration> Noise Reduction (process). On the dialog box, click the button for Capture Noise Print. Then click the button for Select Entire File. Alternatively, part of the file can be selected with the mouse without closing the noise reduction dialog box. For recordings with high noise I start with the slider for Noise Reduction set to 98% and the slider for Reduce By set to 12 decibels. For lower noise levels the settings can be Noise Reduction 75% and Reduce By 6 decibels. With high noise, I leave the Noise Reduction slider near 100% and make adjustments to the Reduce By slider.

The Advanced options at the bottom of the dialog box also need to be set. For Spectral Decay Rate, 5% is a good start and values in the range of 0% to 20% are usually best, possibly up to 40% in certain cases. The Smoothing option is claimed to reduce the bubbly artifacts but it has very little or no effect in my experience. The Transition Width can produce reverberation if it is set too high. A low value of 0 to 7 may be best. Precision Factor has little effect, but I use 11, which may give a slightly better result than the default of 7. The default FFT size of 4096 works well and higher values tend to increase reverberation. The setting for Noise Print snapshot is left at 4000.

Settings to Reducing Artifacts

Strong noise reduction with Adobe Audition may cause bubbly artifacts and/or reverberation. My experience has been that increasing the values for Spectral Decay Rate and Transition Width can reduce the bubbly artifacts but also increase reverberation to a distracting degree. In my experience the most effective way to reduce the bubbly artifacts is to keep the Noise Reduction slider at near 100%. Spectral Decay Rate and Transition Width both need to be kept low to reduce reverberation. Reverberation is also increased by FFT size above 4096 and recordings with a low sampling rate such as 11,025 Hz.

At the time this chapter is being written, there is limited information on using the current version of Adobe Audition. Adobe Audition CS5.5 has recently been released and is a major revision from previous versions. There are no books available yet on using it. The help system in the program and two videos by Adobe offer suggestions for the noise reduction settings in Adobe Audition. Unfortunately, these sources provide very different recommendations. For recordings with low noise, a variety of settings can provide adequate noise removal. However, for higher levels of noise such as often occur with informal recordings, the settings need to be set more carefully. I have found the video by Jason Levine (2010) to provide better guidance than the very different recommendations in the help information in the program (Adobe, 2011). The suggestions presented in this chapter for noise reduction may need to be revised as more experience is gained with Audition CS5.5.

Dynamic Processor to Reduce Artifacts

As noted above, applying noise reduction with a noise profile can reduce noise during useful sounds and a dynamic processor can reduce the remaining noise during quiet periods. After the noise reduction with the noise profile, select a section of the recording with only noise. Find the peak amplitude of the selected noise by using the dropdown menu Window> Amplitude Statistics. On the dialog box, click the Scan Selection button on the lower left of the dialog box. The top line of statistics gives the peak amplitude of the noise.

The Dynamic Processor can be run from the Effects Rack if it is being applied to the entire recording or from the dropdown menu if it is being applied to a selected part of the recording. From either the Effects Rack panel or the Effects drop down menu, select Amplitude and Compression> Dynamics Processing. If the graph is not a straight diagonal line from lower left to upper right, select Default on the Presets pick-list. Tap the spacebar to start playing the noise in a loop. Click on the diagonal line at a point that is about 4 or 5 decibels above the maximum noise level found above. The 4 or 5 decibels are relative to the horizontal axis. This makes a control point that prevents changes higher on the line. Then click another point that is about 5 or 6 decibels lower on the line and slowly drag the line down until the noise just goes away or is at a level that retains the ambiance of the room. Without closing the dialog box, tap Ctrl-A to select the entire file. Continue the playback and adjust the settings if needed. If voices seem to be cut off, it may help to move the first point down closer to the maximum noise level and/or increase the Level Detector Release Time under the Settings tab to 200 ms or more. Close the dialog box when the settings are finished.

Graphic Equalizer or FFT Filters to Reduce Artifacts

Reverberation from noise reduction can sometimes be reduced using a Graphic Equalizer or FFT filter option. From either the Effects Rack panel on the left of the screen or the Effects dropdown menu, select Filter and EQ. Then select one of the three Graphic Equalizers or the FFT option. From the default settings of 0 for all frequencies, lower the setting for 500 Hz to ‑3 decibels and the setting for 100 Hz to ‑6 decibels. Set values between 100 Hz and 500 Hz to be a smooth decline. Then set the values below 100 Hz to a sharper reduction. Adjust these setting as needed. Starting at 250 Hz rather than 500 Hz may be appropriate for milder reverberation. Voice can sometimes be enhanced by increasing the values for 5 kHz to 9 kHz to 4 or 5 decibels. The Mix: Wet – Dry adjustment may be useful here also when the Effects Rack is used.

Adaptive Noise Reduction

The Adaptive Noise Reduction effect in Adobe Audition can also be used for effective noise reduction, particularly when the noise level fluctuates. This effect assumes that there is at least about 3 seconds of only noise at the beginning of the recording. If this pure noise is not present, it can easily be created by copying a section of noise from later in the recording and pasting it multiple times at the beginning. Adaptive Noise Reduction takes a couple of seconds to identify and apply noise reduction at the beginning of the recording.

Adaptive Noise Reduction can be run from the Effects Rack with Noise Reduction/Restoration> Adaptive Noise Reduction. This brings up a warning box which notes that adaptive noise reduction is a CPU intensive effect. This warning can be ignored. If computing power is a problem, Adaptive Noise Reduction can be run from the Effects dropdown menu. Adaptive Noise Reduction has many settings that are not easy to understand. Using the presets from the pick-list may be the best strategy. The preset for DeReverb Single Source may be useful in reducing the reverberation from other noise reduction steps. The noise at the beginning of the recording will need to be deleted after noise reduction is complete.

Updated Noise Reduction

My experience has been that Adobe Audition produces better broadband noise reduction than most audio editing applications, but that the Izotope RX 2 noise reduction program currently has less artifacts and reverberation than Adobe Audition. Izotope RX 2 ($350) is a specialty program for noise reduction and is not a general audio editing application like Adobe Audition.

Broadband noise reduction is continually being improved in most audio software and any comparisons here will quickly be out of date. Those dealing with noisy recordings can make comparisons of different software options at the time of purchase. Trail versions of the relevant software are available and can be used for comparisons.

7. 7 Discrete Noise Reduction

Discrete noise is generated by events such as ringing phones, doorbells, car horns, closing doors, dropping objects, and bumping the microphone. If there is no useful sound occurring at the time these events happen, the noise usually can simply be deleted from the recording. However, if the noise occurs during speech or other useful sounds, removing the noise can be very difficult unless the audio editing software has certain features.

The best way to work with discrete noise is a spectral display or spectrogram display. The horizontal axis is time as in the waveform display, but the vertical axis is sound frequency, ranging from low frequencies on the bottom to high frequencies at the top. The brightness or color at a given point is the loudness or intensity of the particular frequency at the particular point in time. Figure 7.2 shows the spectral display of a phone ringing while a man is speaking. The bright areas toward the bottom on the left side of the display are the primary frequencies in the man’s voice. The less bright areas in the higher frequencies indicate the role of higher frequencies in human speech. The banded area on the right of the display is the phone ringing and the middle of the display shows the voice and phone overlapping. The purple specks across many frequencies when the man is not speaking and the phone is not ringing indicate broadband noise. The display would be perfectly black for a region that had no sound. Broadband noise reduction would eliminate most of the background specks. The corresponding waveform display is shown above the spectral display.





Audio editing spectral display.

Figure 7.2. Spectral display and waveform display for a phone ringing while a man is speaking. The spectral display is below the waveform display. The horizontal axis is time and the vertical axis of the spectral display is increasing sound frequency. The brighter colors are louder sounds at the frequency and time.  The left part of the display is the man speaking. The right part is the phone ringing. The middle of the display is the phone ringing while the man is speaking. The purple flakes or streaks throughout the recording indicate broadband noise. This display is from Adobe Audition.




There are two basic strategies for removing discrete noise. One strategy is to develop a noise profile and apply that to the areas where the noise and useful sound overlap. This is straightforward for cases as shown in Figure 7.2 when there is a section with isolated noise. The other strategy requires basically manual reductions to certain sound frequencies at certain points in time. This strategy can be implemented with different methods that require widely varying amounts of effort.

With either strategy, broadband noise should be reduced before working on the discrete noise. The process of reducing discrete noise may make broadband noise vary over time, which is much more difficult to handle.

Discrete Noise Reduction using Nero WaveEditor

Nero WaveEditor, like many audio editing applications, has limited capabilities for handling discrete noise. It has a crude spectrogram display that is activated with the dropdown menu View> Spectrogram Display. At best this spectrogram display can give an approximate idea of the location and frequencies of the noise.

The noise reduction methods used for broadband noise can be adapted for some discrete noises such as the ringing phone shown in Figure 7.2. Make a noise profile from a selection with the phone ringing with no useful sound. Then carefully select a segment with the phone ringing during useful sound and apply noise reduction using the noise profile. While listening to the preview, move the slider for Gain Floor to the left until the phone ringing is reduced to an acceptable level. This may result in significant distortion of the voice or other useful sounds. The Equalizer under the Tools drop down menu might be used to enhance some of the lost frequencies. However, I have rarely been successful in using Nero WaveEditor to restore significant discrete noise to an acceptable degree.

Another option for removing discrete noise, and the only option when a noise profile cannot be obtained, is to apply a frequency filter to the section of the recording with the noise. The filter needs to remove only certain frequencies. First select the section (or one section) of the recording with the noise. Play the selected section to verify that it does not contain any parts without the noise. Then click the dropdown menu Enhancement> Filter Toolbox. This displays a graph with frequency on the horizontal axis and amplification of the frequencies on the vertical axis. From the pick-list of preset options in the lower part of the dialog box, select Default. This will make the graph of amplification by frequency have a straight line across the top. Click the checkbox for User Drawn Filter Response and make sure the checkboxes for Bandpass Filter and Notch Filters are cleared.

Click on the line at the top of the graph at a point for a frequency that is below the frequency of the noise. This can be a guess. Then click on a point that is above the frequency of the noise. Then click a point between these two points and drag the line down to form a deep V. Drag the bottom of the V to the left as far as it will go, which will be directly below the top point on the left. Click a point on the right side of the V and drag it down as far to the right as it will go, which will be directly below the point on the right. A deep rectangular well is formed. Click the green preview button to play the noise in a loop and adjust the various points for the rectangular well until the noise is cut out or at an acceptable level. Move the points to make the well as narrow and shallow as possible but still controlling the noise. This trial and error strategy usually can be done relatively quickly. If the noise occurred multiple times, the same filter can be applied to other occurrences. Unfortunately, after the noise is removed, the remaining sound may have some distortion.

Discrete Noise Reduction using Adobe Audition

Adobe Audition has more sophisticated and useful methods for handling discrete noise. The spectral display is very useful. If it is not displayed, the bottom of the waveform display can be dragged up with the mouse to reveal the spectral display. Displaying both the waveform and spectral displays is useful for discrete noise. Discrete noise can be easily seen and selected. The frequencies of sound are displayed with useful precision.

The noise reduction methods used for broadband noise can be adapted for some discrete noises such as the ringing phone shown in Figure 7.2. Make a noise profile from a selection with the phone ringing with no useful sound. Then carefully select a segment with the phone ringing during useful sound and apply noise reduction using the noise profile. Start with the Spectral Decay Rate set to zero. While listening to the preview, increase the Reduce By slider until the noise is eliminated. Values of 40 to 100 decibels may be needed. After noise reduction the sound may be reasonably good with some mild reverberation. The reverberation may (or may not) be reduced with the methods described in the section above for broadband noise.

Adobe Audition allows editing with the spectral display, which can be very useful for certain discrete noises. Tapping E or clicking the rectangular Marquee Selection Tool icon on the toolbar at the top of the screen allows selection of rectangular areas on the spectral display. This selects certain frequencies at certain times. Any effects or playback during a selection on the spectral display applies to only the selected frequencies and times. For example, a rectangular selection could be used to define a noise reduction profile using only certain frequencies. This might be useful in some cases, although my experience has been that the results are not better than a profile from the usual time selection using all frequencies. There is also a Paint Brush Selection Tool that can be used to select an odd shaped area.

The primary use of selections on the spectral display is to select the discrete noise and then reduce the loudness of the noise. An easy way to reduce loudness is with the small floating sound volume panel. After selecting the noise, reduce the volume as needed. The sound from the selected area can be played in a loop while the loudness is adjusted. If the noise fluctuates or pulsates in a way that is difficult to precisely select, another strategy is to select an area with the fluctuating noise and use the Dynamic Processor to reduce noise in certain ranges of loudness. This is easiest when the noise is louder than the useful sounds. In such cases, a deep square well of noise reduction is applied narrowly on the high end of the Dynamic Processor.

Adobe Audition also has a Healing Brush Tool that attempts to replace or blend noise with adjacent sounds. As with the Healing Brush Tool in Photoshop, the results are different for different directions of movement of the tool. The Healing Brush works best for noise that manifests as a narrow line on the spectral display with other sounds constant before, during, and after the noise. For broader noise and when other sounds are changing, my experience has been that the Healing Brush tends to produce artifacts. The strategies above that reduce rather than replace noise are usually better in these situations.

In some cases with severe discrete noise, complete correction of the noise may not be possible with the methods described here. As yet, I have not found any better alternatives.

Pop, Crackle, and Hiss Noise Reduction

Recordings digitized from audio tapes or from vinyl records often contain noise in the form of pops, crackles, or hiss. Audio editing software generally has special tools for handling these types of noise. These tools usually have effective default settings and are straightforward to use. For cases that are not handled well by these tools, the broadband and discrete noise reduction methods described above can be used.

7.8 Make Final Frequency Adjustment

Recordings may benefit from enhancement of certain frequencies.  This is particularly true for recordings of music digitized from tapes. These adjustments are done with an equalizer. The specific adjustments depend on the nature of the music and the quality of the recording. Voice recordings can sometimes be enhanced by increasing frequencies in the range of 5 kHz to 9 kHz by 4 or 5 decibels. Slight increases in reverberation may also improve some recordings. In Nero WaveEditor reverberation adjustments are made under the Enhancements dropdown menu. In Adobe Audition, reverberation is adjusted under the dropdown menu Effects> Reverb.

7.9 Set the Final Volume

The final volume for a recording is set based on consistency with other sources of audio for a specific use. This is best done by a person listening to sequential audio files and manually adjusting the volume. Certain audio editing concepts and tools are useful in making these final adjustments.

RMS (root mean square) amplitude is a measure of the average amplitude or loudness of sound. It corresponds to human hearing much better than peak volume that is often used by audio normalization tools.

Perceived loudness is another measure that tracks human hearing even more closely because it incorporates the increased sensitivity that humans have for frequencies in the range of the human voice. When working with RMS or perceived loudness, care must be taken to avoid sound spikes that exceed the maximum that can be digitally processed.

A compressor reduces the volume of the loudest sounds. This makes the sound volume more uniform. A compressor has a setting for threshold that determines how loud the sound must be for the reduction to be applied, and a setting for ratio that determines how much reduction is applied. Many compressors have an option for gain that amplifies the sound after the peaks have been compressed.

A limiter is a more extreme compressor that prevents sound from exceeding a specified maximum value. Many limiters also have an option for gain or amplification after compression.

A leveler creates more uniform sound levels by increasing volume for quiet sounds and reducing volume for loud sounds. A leveler usually has settings for the overall loudness and degree of uniformity of the output sound.

Compressors, limiters, and levelers are all forms of dynamic processing and can be achieved with a dynamic processor.

Nero WaveEditor

Setting the final volume with Nero WaveEditor is a relatively manual process. If the sound level is different for different persons speaking, one option is to select certain areas of the recording and manually set the volume using the dropdown menu Volume> Volume Change. Another option is to use a compressor. WaveEditor implements a compressor as a preset option in the dynamic processor (Tools> Dynamic Processor). After compression is applied to obtain acceptably uniform loudness, the overall loudness can be increased to the final level using the volume change control.

Adobe Audition

Adobe Audition provides information about RMS amplitude, perceived loudness, and peak volumes that can be useful in setting the final volume. This information is obtained with the dropdown menu Window> Amplitude Statistics. Click the Scan Selection button to generate the statistics.

Audition also has a specific tool for leveling speech volume. It is activated with the dropdown menu Effects> Amplitude and Compression> Speech Volume Leveler and can be used from the Effects Rack. The preset settings for Soft and Medium may be useful. Under the advanced options, setting the Noise Offset to 25 decibels rather than the default of 15 decibels often improves sound quality. If the recording includes sounds other than human voice, a limiter can be applied rather than the speech volume leveler. The limiter is initiated with the dropdown menu Effects> Amplitude and Compression> Hard Limiter, or it can be used from the Effects Rack. Here too the preset settings work well, particularly the ones for Light and Medium.

Adobe Audition can also set overall volume based on RMS amplitude, perceived loudness, or peak volume and can process several files together to obtain consistency. Start the process with the dropdown menu Effects> Match Volume. Then drag one or more recordings from the Files panel on the left into the Match Volume dialog box. These recordings will be processed together. Click the Settings button on the lower left to specify the criteria for the sound volume. The criteria include numeric value for RMS, perceived loudness, or peak volume. Alternatively, the criteria can be to match the RMS amplitude of a specified file. Application of a limiter can also be specified. Click the button for Match Volume to apply the effect.

7.10 Set Output Format

The output format for playing audio will typically be either a wave file of CD quality or MP3. A CD quality wave file has 44,100 Hz and a bit depth of 16 bits.

The MP3 format has several variable settings. The simplest form of MP3 encoding is a constant bit rate (CBR) that is set for a particular type of use. A bit rate of 192 kbps (kilobits per second) is considered CD quality. A bit rate of 128 kbps produces smaller files and is often used for music on the internet. A bit rate of 64 kbps is often used for long recordings of people speaking. Other bit rates that are sometimes used include 160 kbps for music, 96 kbps for speech, and 320 kbps as the highest bit rate. The alternative to CBR is variable bit rate (VBR), which produces better sound quality for a given file size, but is more complex and slightly more prone to problems. VBR usually has several options for quality but the software often does not indicate the associated bit rate or intended uses. The MP3 format incorporates a sampling rate and bit depth that are usually set to 44,100 Hz and 16 bits and may not be adjustable.

If the bit depth is being reduced for output (e.g., from 24-bit to 16-bit), it is best to use the dither option. If the bit depth does not change, the dither option is not needed. This applies for both wave and MP3 files.

Nero WaveEditor

For Nero WaveEditor the format of the output file can be set with the usual dropdown menu File> Save As. The format is selected with the pick-list for “Save as file type.”

When “PCM wav file” is selected, the output file has the same sampling frequency and bit depth as the working recording being saved. If these need to be changed, they must be changed with Edit> Convert Sample Format.

Nero includes the Lame MP3 Encoder, which is one of the best MP3 encoders. It is configured prior to saving the file. The MP3 settings are set with the dropdown menu Options> Audio Format Settings. Click the tab for Encoders and then double-click Lame MP3 Encoder. The CBR settings are straightforward. For VBR, the quality settings are not obvious. VBR quality setting 0 has a target bit rate of 320 kbps, setting 1 has a target of 245 kbps, 2 has 190 kbps, 4 has 165 kbps, 5 has 130 kbps, 7 has 100 kbps, and 9 has 65 kbps. After the MP3 settings are complete, selecting “Lame MP3 Encoder” as the file type for the File> Save As command will output a file with these settings.

In WaveEditor, dithering is turned on with the dropdown menu Options> Editor Options. Click the tab for Save/Output Settings and then click the checkbox for “Use dithering when converting to lower bit depths.”

Adobe Audition

In Adobe Audition, the output format is set using the usual dropdown menu File> Save As. In the Save As dialog box the format can be set to either Wave PCM or MP3 Audio. The pick-lists for Sample Type and Format Settings are activated by clicking the corresponding Change button. The options are straightforward. The presets for Sample Type can be useful. For MP3 VBR quality, “Highest Quality” corresponds to 224 kbps, “High Quality” to 192 kbps, “Medium Quality” to 160 kbps, “Medium Bitrate” to 128 kbps, and “Low Bitrate” to 96 kbps.

7.11 Set Output File Name and Location

Working file and files that are created for presentation should have file names and locations that clearly indicate the purpose of the files. As noted earlier, file names can have codes added such as _s to indicate a file for slideshows. Placing the files for different purposes in separate folders can be convenient if they are often copied or moved in groups. Adding codes to the file names is valuable for reducing confusion even if the files are kept in separate folders.


Adobe, 2011, “Using Adobe Audition CS5.5.” Accessed August 29, 2011 at

Fries, Bruce; & Fries, Marty, 2005. Digital Audio Essentials. Published by O’Reilly Media in Sebastopol, CA.

Levine, Jason, 2010. “Spectral Editing Techniques in Adobe Audition for the MAC.” Video viewed September 19, 2011 at

Nero AG, 2010, Nero 9 Manual: WaveEditor. Accessed April 22, 2010 at



[Version of 5/28/2012]

Historical photos and documents

Chapter Contents

Chapter 7. Preparation and Presentation of Historical Audio Recordings

7.1 Audio Editing Software

7.2 Make a working copy of the archival recording

7.3 Split or extract relevant segments from the working copies

7.4 Convert stereo recording to mono

7.5 Set the working volume

7.6 Broadband Noise Reduction

7. 7 Discrete Noise Reduction

7.8 Make Final Frequency Adjustment

7.9 Set the Final Volume

7.10 Set Output Format

7.11 Set Output File Name and Location


Chapters in the Book

Introduction: Background, Purpose, and Example Archive Projects

1. Basic Principles of Archiving Photographs and Documents

2. Good and Best Practices for Making Digital Images

3. Setting Tone and Color in Master Images

4. Documentation for Historical Images

5. Preparation and Display of Historical Photographs and Documents

6. Good and Best Practices for Making Audio Recordings

7. Preparation and Presentation of Historical Audio Recordings

8. Historical Slide Shows

9. Copyrights and Other Legal Rights


A. Practical Color Management

B. Overview of Scanning Software

C. Effective Resolution of Scanners

D. Using Copy Stands with Cameras

E. Batch Processing of Images

Other Resources

Guidance on Using Scanning Software

Guidance on Using Slide Show Software

Downloads and Links

This website and book were developed by Jim Kennedy.


© 2012 James E. Kennedy