Tweet
Spectral Filter
(Forensics Version Only – Advanced Filters)
This filter is essentially a very high resolution Graphic Equalizer that uses FFT techniques. It includes four EQ modes of operation from which to choose. In manual mode, it allows you to create a very high-resolution frequency response contour containing up to 32,000 bands of equalization (essentially a 32,000 Band Graphic Equalizer). The user interface system is intuitive and allows you to zoom-in on a particular portion of the audio spectrum that needs accurate and specific frequency response contouring. By using the right mouse button, you can add bands or inflection points, or delete them simply by pointing and clicking on the graph. This is very useful in Forensic audio applications for removing in-band and out-of-band extraneous noises because of its high degree of frequency selectivity and its very steep slope characteristic.
The spectral filter also includes a spectral inverse filter mode which can be used either manually or automatically. This feature measures the signal amplitude per frequency bin in a .wav file and then applies an inverse response curve in order to normalize it to a reference contoured shape. Essentially, it reverse normalizes to constant signal amplitude per unit bin against a user selected (by preset selection) curve, or via a user customized curve. In other words, frequency bins with lower signal levels than the reference curve are amplified until those bands equal the reference level and those frequency bins with greater amplitude than the reference curve are reduced in amplitude until they match the reference curve. One could view this as an automatic equalizer or an Auto EQ. Spectral inverse mode is very effective for creating an automatic equalization of poorly recorded forensics audio files. This can result in dramatically improved intelligibility of distorted or muffled sound files. It can be viewed as a speech clarifier. File sampling for this feature can be performed manually by highlighting the area of interest or automatically by way of the use of the Auto Sample (checkbox) feature. In Auto Sample mode, the spectral inverse filter becomes essentially an adaptive equalizer and continuously re-samples your .wav file (or system input signal in the case of Live Preview mode) on the fly. Live Preview mode is accomplished via the Multifilter feature in your Diamond Cut suite of filters.
The Spectral Copy mode of operation of the Spectral Filter allows a.wav file to be normalized in terms of frequency content per unit bin to another file or to a certain section of a .wav file. To use the Spectral Filter in this manner, simply highlight a portion of a .wav file while in Spectral Copy mode and then click on the Sample Spectrum button. Next, bring up another .wav file and either Preview or Run the filter. The spectral response of one file will then become imposed on the second file. This mode may be found to be useful when an exemplar of a particular sound exists in one file, but is suspected to be buried in the noise of a second file. An examiner could then sample on the exemplar file containing only the sound of interest (its sound-print) and use that response to help amplify signals in that area of the spectrum on a second file in order to makes certain signals more discernable or intelligible.
A fourth Spectral Filter mode of operation is called “Spectral Difference” (sometimes referred to as Spectral Matching or EQ Matching). In this mode, you can impose the frequency response of one file onto another. This is useful if you want two independently recorded sound tracks to sound similar in terms of frequency response. Simply put, you can cause one .wav file to sound sonically similar to another. To use this mode, you must perform the following operations after switching the Spectral Filter EQ mode to “Spectral Difference”. You will note when this operation is performed, that a new button appears labeled “Sample Source”. Before proceeding, normalize the gain of both .wav files to the same value (this feature is found under the CD Prep Menu).
1. Bring up your reference .wav file (the file which you want
your other file to sound like). This first file is your “Source”
file which should be the sonically higher quality of the two
files that you will be working with.
2. Highlight a sector (around 5 to 30 seconds) of this source
reference file. The sample should be taken
somewhere in the middle of the file. It should not be taken at
the noisy lead-in sector of the file. Better signal averaging results from longer sampling time values.
3. Click on the “Sample Source” button and the system will
then perform some calculations culminating in the creation of
a green colored graph which represents the frequency
response of the sample.
4. Next, bring up your target .wav file (the file requiring
frequency response modification).
5. Click on the Spectral Filter Preview Button and you will
notice that the system will draw two more graphs.
6. You will then hear the corrected response of the Target
.wav file.
7. Optionally, you can highlight an area of the Target
.wav file and press the “Sample Spectrum” button. This will
take a sample of the target file and compute a difference
spectrum.
8. The frequency response of the Target .wav file is shown in
Blue on the graphic display.
9. The red spectral curve is the difference spectrum and is
also the equalization curve that will be applied to the Target
.wav file.
The upper and lower crossover frequencies are settable in the “Spectral Difference” mode of operation by way of the “twin green goalposts” which are visible on the graphical display. These vertical green lines can be “dragged” horizontally along the frequency axis using the left mouse button. The left-most “goalpost” controls the lower crossover frequency while the right-most “goalpost” controls the upper crossover frequency. The frequencies in-between the “goalposts” are the values for which the spectral difference is calculated and is reflected by the shape of the red graph. Those signals above and below the “goalpost” crossover frequencies revert to the native target file values. Adjust these “goalposts” until you achieve the most natural sound with the least amount of high frequency hiss and low frequency rumble.
Another mode of operation which may be found to be useful in some applications (having widely varying target frequency responses) is the Auto Sample button option. This will update the calculation of the spectral difference on-the-fly providing you with a variable spectral difference compensation system.
Note 1: Forensics Audio Work may benefit from high values
of FFT size while studio audio work (or “Hi-Fi” work) may
benefit from substantially lower values of FFT size.
Note 2: The Spectral Difference filter can be used in a variety
of ways. You can use just one file and sample different
portions to generate a difference spectrum or use two different
files and apply the spectral difference to a third file.
Note 3: When using Spectral Copy and/or Spectral Difference
modes of operation, both .wav files must have the same
sampling rate and bit depth attributes.
Note 4: When using the Spectral Difference mode to match the response of a noisy file to a clean file (Cassette and CD for example), it is important to do all of the restoration work on the noisy file before applying the spectral difference filter otherwise the noise components with be detected as signal and distort the desired response.
Note 5: When using Spectral Copy and/or Spectral Difference
modes of operation, both .wav files must use the same FFT
Size setting.
The Spectral Filter
The following controls are provided in the Spectral Filter:
*Note 1: Adding or deleting frequency bands can also be accomplished by clicking the right mouse button on the graphical display. A dialog box will pop up giving you some optional actions.
Note 2: It is important to convert all forensics .wav files (8 kHz, 11.025 kHz and 22.05 kHz) up to a 44.1 kHz sampling rate before using the Spectral Filter. To accomplish this, use the Change Sample Rate feature found under the Edit menu.
Spectral Filter Application Example #1
You have a Forensics Audio file that is extremely muffled, meaning that its intelligibility is extremely poor because of a lack of the sibilant sounds due to significant high frequency loss. To improve the intelligibility of this file, place the Spectral Filter into Spectral Inverse Mode and check the Auto Sample checkbox. Choose the Auto EQ, Normalize to White Noise or the Auto EQ, Normalize to Human Voice preset. Preview the file. If the sound is improved to your satisfaction, then “Run” the Filter. If the improvement is not adequate, experiment with some of the other Auto EQ based factory presets until you find the best one for your purposes.
Spectral Filter Application Example #2
You have a recording of a male and a female chatting with one another in an automobile. The recorder microphone is closest to the female and it is very hard to discern the male end of the discussion. The background noise is due to the high speed at which the automobile is traveling. However, there is one small sector of the recording where the car was stopped at a traffic light and both the male and female ends of the discussion are clear and discernable. To improve the clarity of the male voice during the times at which the car is traveling at high speeds, place the Spectral Filter in Spectral Copy mode. Highlight the male voice (only) at the point where the car was stopped at the traffic light and then click on the Sample Spectrum button in order to obtain a sound-print. Then, using this sample, “Preview” or “Run” the Spectral filter in the areas of the recording during which the male voice was not discernable.
Spectral Filter Application Example #3
You have a cockpit voice recording containing a lot of airplane noise. You desire to listen to hear if the flaps-down switch has been flipped at a certain point in time, but at the point in time in question, all you hear is random noise. You desire to enhance the playback of the recording to hear whether or not the sound of that switch being flipped by the pilot of the aircraft is present on the recording. You can go to a flight simulator of the aircraft in question and record the sound of that switch being flipped with a quiet ambient noise environment. Bring up that sound file and highlight the actual sound of the switch being flipped. Next place the Spectral Filter into Spectral Copy mode of operation and then click on the Sample Spectrum button. Bring up the cockpit voice recording .wav file and “Preview” it around the time period in question. This action should provide more gain of the sound-print spectrum occupied by the flipping of the switch potentially making it more audible. Obviously, you can’t prove the negative by this technique, but if the switch is heard, you can demonstrate that something took place at the point in time in question. Of course, that positive result may represent the wrong switch being thrown, but that is beyond the scope of this application example.
Spectral Filter Application Example #4
You have recordings of an old television variety show from the 1950s (created before the advent of video tape recording). Your job is to assemble a new digital edit of this television series. The dialog and music are both recorded on a cine optical track. However, a much higher quality copy of the musical portion of the show exists on magnetic tape (sans dialog). In the editing process, you need to use the optical track to assemble the edit, but you insert the magnetic track for the musical portion of the job because of its higher sonic quality. However, you observe that there is an unnatural transition between the dialog (optical) track and the musical (magnetic) track. You can create more natural transitions between the dialog and the musical interludes by employing the Spectral Difference mode of the Spectral Filter to the project. Use the Magnetic track as the “Source” Track and then apply that to the Target optical portions of the final edit using the Spectral Difference feature.
Spectral Filter Application Example #5
You have transferred an old acoustical recording of a famous opera singer to your hard drive. You have removed the noise, but the recording sounds honky, hollow, resonant and unnatural. But, you have a more modern recording of this singer created during the electrical recording period of time. You can use the Spectral Difference feature to create a more natural sounding result. Just use the more modern recording as the Source .wav file and then apply it to the acoustical recording as the Target.
(Forensics Version Only – Advanced Filters)
This filter is essentially a very high resolution Graphic Equalizer that uses FFT techniques. It includes four EQ modes of operation from which to choose. In manual mode, it allows you to create a very high-resolution frequency response contour containing up to 32,000 bands of equalization (essentially a 32,000 Band Graphic Equalizer). The user interface system is intuitive and allows you to zoom-in on a particular portion of the audio spectrum that needs accurate and specific frequency response contouring. By using the right mouse button, you can add bands or inflection points, or delete them simply by pointing and clicking on the graph. This is very useful in Forensic audio applications for removing in-band and out-of-band extraneous noises because of its high degree of frequency selectivity and its very steep slope characteristic.
The spectral filter also includes a spectral inverse filter mode which can be used either manually or automatically. This feature measures the signal amplitude per frequency bin in a .wav file and then applies an inverse response curve in order to normalize it to a reference contoured shape. Essentially, it reverse normalizes to constant signal amplitude per unit bin against a user selected (by preset selection) curve, or via a user customized curve. In other words, frequency bins with lower signal levels than the reference curve are amplified until those bands equal the reference level and those frequency bins with greater amplitude than the reference curve are reduced in amplitude until they match the reference curve. One could view this as an automatic equalizer or an Auto EQ. Spectral inverse mode is very effective for creating an automatic equalization of poorly recorded forensics audio files. This can result in dramatically improved intelligibility of distorted or muffled sound files. It can be viewed as a speech clarifier. File sampling for this feature can be performed manually by highlighting the area of interest or automatically by way of the use of the Auto Sample (checkbox) feature. In Auto Sample mode, the spectral inverse filter becomes essentially an adaptive equalizer and continuously re-samples your .wav file (or system input signal in the case of Live Preview mode) on the fly. Live Preview mode is accomplished via the Multifilter feature in your Diamond Cut suite of filters.
The Spectral Copy mode of operation of the Spectral Filter allows a.wav file to be normalized in terms of frequency content per unit bin to another file or to a certain section of a .wav file. To use the Spectral Filter in this manner, simply highlight a portion of a .wav file while in Spectral Copy mode and then click on the Sample Spectrum button. Next, bring up another .wav file and either Preview or Run the filter. The spectral response of one file will then become imposed on the second file. This mode may be found to be useful when an exemplar of a particular sound exists in one file, but is suspected to be buried in the noise of a second file. An examiner could then sample on the exemplar file containing only the sound of interest (its sound-print) and use that response to help amplify signals in that area of the spectrum on a second file in order to makes certain signals more discernable or intelligible.
A fourth Spectral Filter mode of operation is called “Spectral Difference” (sometimes referred to as Spectral Matching or EQ Matching). In this mode, you can impose the frequency response of one file onto another. This is useful if you want two independently recorded sound tracks to sound similar in terms of frequency response. Simply put, you can cause one .wav file to sound sonically similar to another. To use this mode, you must perform the following operations after switching the Spectral Filter EQ mode to “Spectral Difference”. You will note when this operation is performed, that a new button appears labeled “Sample Source”. Before proceeding, normalize the gain of both .wav files to the same value (this feature is found under the CD Prep Menu).
1. Bring up your reference .wav file (the file which you want
your other file to sound like). This first file is your “Source”
file which should be the sonically higher quality of the two
files that you will be working with.
2. Highlight a sector (around 5 to 30 seconds) of this source
reference file. The sample should be taken
somewhere in the middle of the file. It should not be taken at
the noisy lead-in sector of the file. Better signal averaging results from longer sampling time values.
3. Click on the “Sample Source” button and the system will
then perform some calculations culminating in the creation of
a green colored graph which represents the frequency
response of the sample.
4. Next, bring up your target .wav file (the file requiring
frequency response modification).
5. Click on the Spectral Filter Preview Button and you will
notice that the system will draw two more graphs.
6. You will then hear the corrected response of the Target
.wav file.
7. Optionally, you can highlight an area of the Target
.wav file and press the “Sample Spectrum” button. This will
take a sample of the target file and compute a difference
spectrum.
8. The frequency response of the Target .wav file is shown in
Blue on the graphic display.
9. The red spectral curve is the difference spectrum and is
also the equalization curve that will be applied to the Target
.wav file.
The upper and lower crossover frequencies are settable in the “Spectral Difference” mode of operation by way of the “twin green goalposts” which are visible on the graphical display. These vertical green lines can be “dragged” horizontally along the frequency axis using the left mouse button. The left-most “goalpost” controls the lower crossover frequency while the right-most “goalpost” controls the upper crossover frequency. The frequencies in-between the “goalposts” are the values for which the spectral difference is calculated and is reflected by the shape of the red graph. Those signals above and below the “goalpost” crossover frequencies revert to the native target file values. Adjust these “goalposts” until you achieve the most natural sound with the least amount of high frequency hiss and low frequency rumble.
Another mode of operation which may be found to be useful in some applications (having widely varying target frequency responses) is the Auto Sample button option. This will update the calculation of the spectral difference on-the-fly providing you with a variable spectral difference compensation system.
Note 1: Forensics Audio Work may benefit from high values
of FFT size while studio audio work (or “Hi-Fi” work) may
benefit from substantially lower values of FFT size.
Note 2: The Spectral Difference filter can be used in a variety
of ways. You can use just one file and sample different
portions to generate a difference spectrum or use two different
files and apply the spectral difference to a third file.
Note 3: When using Spectral Copy and/or Spectral Difference
modes of operation, both .wav files must have the same
sampling rate and bit depth attributes.
Note 4: When using the Spectral Difference mode to match the response of a noisy file to a clean file (Cassette and CD for example), it is important to do all of the restoration work on the noisy file before applying the spectral difference filter otherwise the noise components with be detected as signal and distort the desired response.
Note 5: When using Spectral Copy and/or Spectral Difference
modes of operation, both .wav files must use the same FFT
Size setting.
The Spectral Filter
The following controls are provided in the Spectral Filter:
- Amplitude vs. Frequency Graph: Up to three graphical lines are shown on the Spectral Filter graph. In Manual mode, a black line with square inflection points (or touch points) are shown. In Spectral Inverse Filter mode, the blue line represents the frequency response of the .wav file (or a portion thereof). The Red line represents the calculated Inverse Spectrum response required to correct the response to the black line curve.
- Graphic Representation of the user defined Frequency Response: Drag the blue inflection points (dots) with your mouse in order to establish the desired frequency response.
- Range (dB): This parameter determines the vertical axis range for the Spectral Filter. The following ranges are provided:
- + 40 / - 100 dB
- + 20 / - 40 dB
- + / - 40 dB
- + / - 20 dB
- + / - 10 dB
- + / - 3 dB
- FFT Size (Number of Frequency Bands): Higher Values of FFT Size improves frequency resolution / frequency discrimination but with the tradeoff of higher levels of digital artifacting. Choose the best balance between those two requirements based on the needs of the file that you are working on. The following FFT size selections are available in the Spectral Filter:
- 256 (128 Bands)
- 512 (256 Bands)
- 1,024 (512 Bands)
- 2,048 (1,024 Bands)
- 4,096 (2,048 Bands)
- 8,192 (4,096 Bands)
- 16,384 (8,192 Bands)
- 32,768 (16,384 Bands)
- 65,635 (32,768 Bands)
- Zoom In: Click on the Magnifying Glass Icon with the “+” sign or drag the mouse over the area of the spectral graph on which you desire to focus.
- Zoom Out: Click on the Magnifying Glass Icon with the “-” sign.
- EQ Mode: Allows you to choose between the four main modes of the spectral filter. They are Manual Adjustment, Spectral Copy, Spectral Inverse and Spectral Difference. Manual Adjust is a direct manipulation of the frequency response curve similar to a standard Equalizer. Spectral Copy creates a response curve that is the shape of the average response of the sampled file. Spectral Inverse creates an EQ response that seeks to normalize the response of a signal to a desired frequency response curve (or contour). Spectral Difference creates an EQ response that represents the difference is frequency response between to samples.
- Auto Sample Checkbox: This feature enables an adaptive mode of operation in which the system automatically samples and applies a portion of the .wav file to the Spectral Inverse or Spectral Difference Filter for normalization / correction on an ongoing basis. This feature is particularly useful in “moving mic” situations when the acoustical environmental is changing throughout the recording. This mode of operation will help optimize the intelligibility of the .wav file in these variable acoustical environment situations. You can freeze the sampling any time desired by un-checking the checkbox while previewing the file. If you freeze the system using the Auto Sample feature, the last sound-print will be held in memory thereafter and will be used as the reference response when the filter is “Run”.
- Gain Control Slider: Adjusts the Spectral filter overall output level over the range of – 40 dB to + 40 dB.
- Bypass Check box: Allows you to hear the source material with the Spectral Filter disconnected from the system.
- Keep Residue: Allows you to hear what you are removing when using the Spectral Filter. Mathematically, the keep residue signal is the Input Source Signal minus the Spectral Filtered Signal.
- Reset: Returns all off the blue inflection points (bands) on the graphical display to a “flat” (white noise related) response.
- Sample Spectrum Button: This control is used in conjunction with the Spectral Inverse Mode of operation. It allows you to take a sound-print (manually) of a highlighted sector of your .wav file for use as the signal to be used to create the desired spectral inverse filter response.
- Sample Source Button: This button is used to sample a source reference file’s frequency response and is only appears when the system is operating in Spectral Difference Mode.
- Presets: (Save or Delete buttons) - The factory presets are especially useful when using the Spectral Inverse Filter Mode in that Normalization curves are provided to White, Pink, Inverse Pink, Brown and Inverse Brown noise contours with ease. Other useful curves are also provided. For definitions of the mentioned noise types, please refer to the Glossary of Terms section of this Users Guide.
- Right Mouse Button: To use this feature, select the function you desire with the right mouse button and move the band with the left mouse button. Here are the functions available for the Spectral Filter by using the Right Mouse Button:
- Add Point - (This feature adds a frequency band on the graph where you are pointing with the mouse.)*
- Delete Point - (This feature deletes a frequency band on the graph where you are pointing with the mouse.)
- Reset Point Count - (Use this feature to reset the system to its default value - - - factory default = 0 dB, flat-line response.)
*Note 1: Adding or deleting frequency bands can also be accomplished by clicking the right mouse button on the graphical display. A dialog box will pop up giving you some optional actions.
Note 2: It is important to convert all forensics .wav files (8 kHz, 11.025 kHz and 22.05 kHz) up to a 44.1 kHz sampling rate before using the Spectral Filter. To accomplish this, use the Change Sample Rate feature found under the Edit menu.
Spectral Filter Application Example #1
You have a Forensics Audio file that is extremely muffled, meaning that its intelligibility is extremely poor because of a lack of the sibilant sounds due to significant high frequency loss. To improve the intelligibility of this file, place the Spectral Filter into Spectral Inverse Mode and check the Auto Sample checkbox. Choose the Auto EQ, Normalize to White Noise or the Auto EQ, Normalize to Human Voice preset. Preview the file. If the sound is improved to your satisfaction, then “Run” the Filter. If the improvement is not adequate, experiment with some of the other Auto EQ based factory presets until you find the best one for your purposes.
Spectral Filter Application Example #2
You have a recording of a male and a female chatting with one another in an automobile. The recorder microphone is closest to the female and it is very hard to discern the male end of the discussion. The background noise is due to the high speed at which the automobile is traveling. However, there is one small sector of the recording where the car was stopped at a traffic light and both the male and female ends of the discussion are clear and discernable. To improve the clarity of the male voice during the times at which the car is traveling at high speeds, place the Spectral Filter in Spectral Copy mode. Highlight the male voice (only) at the point where the car was stopped at the traffic light and then click on the Sample Spectrum button in order to obtain a sound-print. Then, using this sample, “Preview” or “Run” the Spectral filter in the areas of the recording during which the male voice was not discernable.
Spectral Filter Application Example #3
You have a cockpit voice recording containing a lot of airplane noise. You desire to listen to hear if the flaps-down switch has been flipped at a certain point in time, but at the point in time in question, all you hear is random noise. You desire to enhance the playback of the recording to hear whether or not the sound of that switch being flipped by the pilot of the aircraft is present on the recording. You can go to a flight simulator of the aircraft in question and record the sound of that switch being flipped with a quiet ambient noise environment. Bring up that sound file and highlight the actual sound of the switch being flipped. Next place the Spectral Filter into Spectral Copy mode of operation and then click on the Sample Spectrum button. Bring up the cockpit voice recording .wav file and “Preview” it around the time period in question. This action should provide more gain of the sound-print spectrum occupied by the flipping of the switch potentially making it more audible. Obviously, you can’t prove the negative by this technique, but if the switch is heard, you can demonstrate that something took place at the point in time in question. Of course, that positive result may represent the wrong switch being thrown, but that is beyond the scope of this application example.
Spectral Filter Application Example #4
You have recordings of an old television variety show from the 1950s (created before the advent of video tape recording). Your job is to assemble a new digital edit of this television series. The dialog and music are both recorded on a cine optical track. However, a much higher quality copy of the musical portion of the show exists on magnetic tape (sans dialog). In the editing process, you need to use the optical track to assemble the edit, but you insert the magnetic track for the musical portion of the job because of its higher sonic quality. However, you observe that there is an unnatural transition between the dialog (optical) track and the musical (magnetic) track. You can create more natural transitions between the dialog and the musical interludes by employing the Spectral Difference mode of the Spectral Filter to the project. Use the Magnetic track as the “Source” Track and then apply that to the Target optical portions of the final edit using the Spectral Difference feature.
Spectral Filter Application Example #5
You have transferred an old acoustical recording of a famous opera singer to your hard drive. You have removed the noise, but the recording sounds honky, hollow, resonant and unnatural. But, you have a more modern recording of this singer created during the electrical recording period of time. You can use the Spectral Difference feature to create a more natural sounding result. Just use the more modern recording as the Source .wav file and then apply it to the acoustical recording as the Target.