Many times, you’ll come across recordings that have digital clipping distortion or missing upper frequencies due to lossy audio compression or low sample rates. While none of these processes are fully reversible, they are possible to mitigate to some extent using declipping (in the former case) and a process that I’ve come to “upmastering” (credit to Solra Bizna for the name). I described these processes in a PDF I included with the Eternal X 1.2 OST, found on my Marathon soundtracks page. This is an updated version of those notes.
Note that, while I wrote this guide with iZotope RX in mind (the gold standard of audio repair software, in my book), it’s perfectly possible to adapt its principles in other audio processing software. The quality of your results may vary, depending on how good your chosen software is at declipping and pitch-shifting audio.
Please contact me if you have any questions.
My first step for around 75% of [the Eternal X 1.2 OST] was declipping. Many people seem to pay little attention to the levels when mixing and/or mastering songs; as a result, the peaks and troughs of waveforms get clipped off (it’s essentially the result of a variable overflow), creating a form of distortion informally referred to as ‘black noise’. It results in a muffled sound and can cause listener fatigue. (If you want to find yourself annoyed at 90% of modern music, look up the loudness war.)
Thus, I employed iZotope’s declipper for any song that clipped. I must emphasize that the declipper cannot perfectly reconstruct the data lost from clipping, but it usually comes up with reasonable approximations that will suffice for most people’s listening. The declipper scans the waveform and tells you where it clips. You can usually set the filter to start at 0.25 or 0.5 dB below that and get good results.
(I’d usually recommend running this process on an entire album at once to avoid messing up gapless transitions and preserve its songs’ relative volumes, which can have artistic reasoning behind them. However, here, I could reasonably infer that they didn’t, so I did this process song-by-song. Additionally, on some albums, some songs may clip at different levels; for that matter, different segments of individual songs may also clip at different levels. Listen, and use your best judgement.)
(Notably, iZotope themselves recommend declipping before performing any other audio repair.)
Next, I ran most songs 90° through a phase rotation filter; this mostly affects subsonic (≤20 Hz) frequencies, but also usually increases dynamic range. I have no hypothesis for why this reduces listener fatigue; it shouldn’t be perceptibly different at all. However, this is the only palatable way I’ve found to listen to various albums that are too distorted to salvage with declipping (i.e., Merzbow’s Venereology and Pulse Demon⁽¹⁾; the Psychic Paramount’s Gamelan into the Mink Supernatural). Since it performed wonders for those recordings, it’s now a standard part of my remastering process.
[Warning: You should generally only use phase rotation on a finished mix, unless there are phase issues with a recording you’re mixing; phase rotation can be destructive. In the most extreme case, phase rotating a recording by 180° and mix pasting it with its original produces silence!]
The above two steps, along with normalising to -0.1 dB, should suffice for any recording for which you have a lossless source. Unfortunately, this was the case for none of the Eternal tracks; they were all MP3s. Craig Hardgrove and Nicholas Singer didn’t have higher-quality versions of their recordings; Dr Hardgrove said he might someday be able to make lossless versions, but it didn’t sound like it’d be anytime soon.⁽²⁾ Thus, I approximated the missing upper frequencies as well, using a process we might call “upmastering”. (I was unable to contact Thomas Livingston or Eike Steffen at all.)
In each case, I copied my results to a new file. Then I ran the ‘pitch & time’ filter to raise the pitch by 12 half-steps, or an octave, using the highest possible quality algorithm, Radius. This is a time-consuming, processor-intensive task that may take several times a song or album’s running time on many machines. Feel free to step away to get a drink or something while it runs.
Once it finishes, roll off the lower frequencies of your pitch-raised copy with the Corrective EQ filter. You’ll want to discard everything below about 14 kHz, depending upon the quality of MP3 compression. I used the analogue filter ‘low shelf’ for most of the bands, but the exact slope and gain I used depended on the track. Lastly, mix paste the results with your original track. You may have to experiment with several different EQ settings to find the best-sounding version; unfortunately, there’s no one-size-fits-all solution here, and simply eyeballing the results won’t suffice. Use your best judgement. Once you’re satisfied, normalise to -0.1 dB, export as FLAC, and you’re done.
[Note: Modern versions of iZotope call this filter simply ‘EQ’, and rather than using the low shelf, you can just use a high pass filter with it. I’ve also taken to adding a much subtler low pass filter as well.]
This process may be able to salvage many modern recordings that you may otherwise find fatiguing. To be honest, I find myself putting about 90% of the music I download these days through at least some of these steps, but I admit to being obsessive about this sort of thing. Regardless, I hope some readers find it useful for some of the worst offenders.
Upmastering works much better for some instruments than for others. Essentially, if an instrument is primarily based on even harmonics (meaning that if you a play a 440 Hz note, it will have resonant frequencies at 880 Hz, 1,760 Hz, 3,520 Hz, 7,040 Hz, 14,080 Hz, etc.), it will work almost flawlessly; if it is primarily based on odd harmonics, it won’t work nearly as well, though many listeners still probably won’t be able to notice since the above process results in most frequencies below 16 kHz being rolled off.
As it so happens, nearly every instrument prominent in rock music – drums, bass, guitar, organ, and piano – has fairly substantial even harmonics. This is less reliable for jazz and classical instruments – closed-end wind instruments tend to be based almost entirely on odd harmonics, so a clarinet will have almost no resonant frequencies an octave apart, while a trumpet will only have some. Synthesizers are a complete mixed bag; some synth voices have no even harmonics at all, while others have plenty. This Steve Hoffman Forums thread may be enlightening on this count.
So, in summation, your steps are as follows:
A final warning: The spectrogram isn’t an EQ tool. It’s more important that your remasters sound good than that their spectrogram looks good. At most, the spectrogram can help you guess what levels and EQ to use for your first attempt, but you should rely on your ears for sounds, not your eyes.