After writing about frequency domain analysis recently, I thought a brief example of the utility of this technique in a practical audio application would be useful. I digitized an approximately 10 second piece of acoustic guitar performance from a recording dubbed to a Type I cassette tape about 40 years ago from a quarter inch reel to reel tape recording. The source reel to reel recording was made on two tracks of a TEAC 2340 using high quality quarter inch tape and normal bias at 7.5 ips (and quarter track semi-pro format, i.e., either two or four tracks, as desired, running in the same direction along the length of the quarter inch wide magnetic tape). You can hear the tape hiss noise on the old recording here (use headphones or turn up your volume loud or you may not hear the noise):

littlemarthafreqdomaudarticlewnoise.mp3 |
I used Audacity (the fantastic open source audio editor) to generate a frequency analysis of that recording:
That spike on the right side of the plot is at about 15 kHz (15000 Hz) and is at -60 dB. I suspect that this is a residual harmonic (subharmonic) of the reel to reel 150 kHz tape bias signal. In order to record in the linear area of magnetic tape it is common to combine a bias signal with the signal being recorded in order to assure there is sufficient magnetic flux to the tape. You can’t hear much over 20 kHz, so this is not a problem in playback; however, since we have a lot of tape hiss, which is higher frequencies, we want to reduce that frequency along with other higher frequencies. You could use a notch filter to target specifically the 15 kHz frequency, but because the tape hiss occupies a much larger range of frequencies, I decided to use a multi-band equalizer to roll off (decrease the level of) all the frequencies in the recording higher than approximately 3 kHz. I used the Equalization Effect in Audacity to decrease the levels of frequencies in the recording, decreasing them by a larger and larger factor as the frequency increased (you can see that I moved the attenuation sliders farther and farther down as the frequency they effected increased in the screen below). I did this because most of the legitimate sound content, that is, the recorded guitar sound, is lower frequency than the tape hiss, but I wanted to keep some of the brightness of the guitar while reducing the perception of the hiss noise. Here’s how I setup the equalizer:
After reducing those high frequencies by applying the above equalization settings, this is how the recording sounds:

littlemarthanoiseoutfreqdomaudarticle.mp3 |
You can see the effect of the equalization on the frequency analysis of the edited recording:
See how the 15 kHz noise spike is now about -79 dB? That is 19 dB lower than originally and you can see that all those frequencies higher than 3 kHz or so are now ramping down at a pretty good slope to a much lower level. That is why the edited recording has almost no tape hiss (yet the guitar still sounds almost as bright as it did originally). I should note that another reason eq’ing out the high frequencies a bit took the hissing noise down so well is that the cassette tape was apparently recorded (copied from the reel to reel) in Dolby (B) noise reduction mode on the cassette deck used (the cassette tape Dolby box was checked by the person making the recording, so I believe they did indeed use Dolby).
Magnetic tape has an inherent noise floor (a tape hiss) deriving from the magnetization of the ferrite particles on the tape emulsion even in the absence of a recorded signal, i.e., if you take a blank cassette tape and play it back at full volume, you will hear some hiss. Dolby noise reduction was developed in the 60’s to try to reduce that inherent magnetic tape noise. Basically, when you record a tape with Dolby noise reduction the high frequencies in the audio you are recording are boosted as they are transferred to the magnetic tape, the higher level (loudest) signals less, the lower level (quietist) signals more. When you play the tape back, you want to use the same Dolby noise reduction circuitry to reduce those boosted high frequencies by the same amounts to restore the original dynamic range i.e., reduce the higher level high frequency signals a little, the lower level high frequency signals a lot. Since the tape hiss comes from the playback tape at a constant low level (as a property of the magnetic tape itself rather than the recorded audio), it gets lowered a lot as it enters the Dolby circuit prior to the playback amplifier (since it is a low level high frequency signal) and drops by as much as 10 dB below the actual audio you recorded (since the low level high frequency audio you recorded was boosted when you recorded it). This reduces the level of tape hiss and its perception by the hearer significantly.
In our test case here, the cassette recording was apparently recorded with Dolby noise reduction, so the high frequencies were boosted. However, when I played them back, I did not have access to Dolby circuitry. When I cut the high frequencies (as I described earlier), I probably reduced the tape hiss a lot, but only reduced the guitar recording high frequencies back to their original levels (since they would have been previously boosted during the Dolby recording process years ago when the tape was dubbed from the reel to reel master). In effect, I manually created a Dolby noise reduction filter.
So, I think you can see some immediate applications for these kinds of techniques in audio recording and mixing. For example, if your acoustic guitar track sounds muddy, look at the frequency plot of the recording and see where the lower frequency levels are, then do some equalization in that area. As a matter of fact, before I began the noise reduction process on this test recording, I did reduce the frequency content below 400 Hz (using the Equalization Effect) in order to clarify the sound of the guitar (and compressed and amplified the track, which was recorded at too low a volume). There are some general rules of thumb for eq of common rock instruments, for example, bass guitar tracks muddy up a multi-track mix unless you eq out the general neighborhood of 250 Hz in the bass guitar track. You can, of course, skip the frequency analysis and go entirely by ear. In that case, just increase the slider levels on particular frequencies of your equalizer until the objectionable sound quality becomes even worse, and then lower those same frequency sliders on the equalizer to eq out the undesired frequencies for that track. Or, conversely, adjust the sliders on the equalizer for a particular track at different frequencies until the track punches through the surrounding mix or acquires the type of sound you are looking for. It is helpful to loop a few seconds of the track (set it up to keep repeating automatically) while you are making this kind of analysis by ear, so you don’t have to keep restarting the track.
Magnetic tape has an inherent noise floor (a tape hiss) deriving from the magnetization of the ferrite particles on the tape emulsion even in the absence of a recorded signal, i.e., if you take a blank cassette tape and play it back at full volume, you will hear some hiss. Dolby noise reduction was developed in the 60’s to try to reduce that inherent magnetic tape noise. Basically, when you record a tape with Dolby noise reduction the high frequencies in the audio you are recording are boosted as they are transferred to the magnetic tape, the higher level (loudest) signals less, the lower level (quietist) signals more. When you play the tape back, you want to use the same Dolby noise reduction circuitry to reduce those boosted high frequencies by the same amounts to restore the original dynamic range i.e., reduce the higher level high frequency signals a little, the lower level high frequency signals a lot. Since the tape hiss comes from the playback tape at a constant low level (as a property of the magnetic tape itself rather than the recorded audio), it gets lowered a lot as it enters the Dolby circuit prior to the playback amplifier (since it is a low level high frequency signal) and drops by as much as 10 dB below the actual audio you recorded (since the low level high frequency audio you recorded was boosted when you recorded it). This reduces the level of tape hiss and its perception by the hearer significantly.
In our test case here, the cassette recording was apparently recorded with Dolby noise reduction, so the high frequencies were boosted. However, when I played them back, I did not have access to Dolby circuitry. When I cut the high frequencies (as I described earlier), I probably reduced the tape hiss a lot, but only reduced the guitar recording high frequencies back to their original levels (since they would have been previously boosted during the Dolby recording process years ago when the tape was dubbed from the reel to reel master). In effect, I manually created a Dolby noise reduction filter.
So, I think you can see some immediate applications for these kinds of techniques in audio recording and mixing. For example, if your acoustic guitar track sounds muddy, look at the frequency plot of the recording and see where the lower frequency levels are, then do some equalization in that area. As a matter of fact, before I began the noise reduction process on this test recording, I did reduce the frequency content below 400 Hz (using the Equalization Effect) in order to clarify the sound of the guitar (and compressed and amplified the track, which was recorded at too low a volume). There are some general rules of thumb for eq of common rock instruments, for example, bass guitar tracks muddy up a multi-track mix unless you eq out the general neighborhood of 250 Hz in the bass guitar track. You can, of course, skip the frequency analysis and go entirely by ear. In that case, just increase the slider levels on particular frequencies of your equalizer until the objectionable sound quality becomes even worse, and then lower those same frequency sliders on the equalizer to eq out the undesired frequencies for that track. Or, conversely, adjust the sliders on the equalizer for a particular track at different frequencies until the track punches through the surrounding mix or acquires the type of sound you are looking for. It is helpful to loop a few seconds of the track (set it up to keep repeating automatically) while you are making this kind of analysis by ear, so you don’t have to keep restarting the track.