Recovering High Frequencies in Compressed Audio: A Harmonic Transposition Technique

When an audio file goes through lossy compression — a low-bitrate MP3, or audio pulled from a messaging app — something obvious shows up on the spectrum: part of the high-frequency information is simply gone.

It’s not just less brightness. Harmonics that are part of the original timbre of instruments and vocals disappear too.

An equalizer can boost what’s still there, but it can’t bring back information that no longer exists.

Over the past few years I developed a workflow that tries to approximate that lost information by generating new content that’s harmonically related to the original signal. It doesn’t attempt to restore the original audio exactly — that’s physically impossible once a lossy codec has removed the information — but rather to reconstruct a much more natural sense of spectral extension than simply boosting the highs.

The problem

Open a heavily compressed MP3 in a spectrum analyzer (Waves PAZ Analyzer, for example) and you’ll usually see a very defined cutoff around a certain frequency.

Depending on the bitrate, content can disappear completely above 14, 16, or even 18 kHz.

The result usually feels dull, with less air, less definition, and less sense of depth.

The concept

My idea starts from a very simple observation.

If I shift a full copy of the audio exactly one octave up (+12 semitones), every frequency in the file doubles.

In other words:

• 2 kHz moves to roughly 4 kHz.
• 4 kHz shows up near 8 kHz.
• 8 kHz generates content around 16 kHz.

This doesn’t recover the original lost information.

What it does is generate new content whose harmonic relationship stays coherent with the existing signal.

That content can then be carefully blended with the original signal to reconstruct part of the sense of brightness.

The procedure

1. Duplicate the audio

I import the file into Ableton Live and create an identical copy of the channel.

2. Transpose one octave up

On the duplicated channel I use Ableton Live’s Complex algorithm, then raise it exactly +12 semitones. I don’t use Complex Pro or other algorithms — Complex tends to give me the character I’m looking for in this process.

3. Isolate only the useful information

On the transposed track I place a linear-phase EQ (usually FabFilter Pro-Q 3). Using the spectrum analyzer, I find exactly where the original file’s information starts falling off, and from that point I apply a high-pass filter to keep only the generated content that will occupy the region where the original file went missing. This avoids unnecessarily reinforcing frequencies that are already present.

4. Balance using spectral analysis

I don’t set the level by ear alone. I compare the spectral response in Peak mode and raise or lower the transposed channel’s gain until the visual and audible transition feels even. The goal isn’t to exaggerate the highs, but to extend the spectrum’s energy naturally.

5. Reintegrate both signals

Both channels are routed to the same bus, where I do a final shaping pass:

• A gentle high-frequency filter to avoid an artificial top end.
• Very light compression to glue both signals together.

The result usually feels more open, present, and detailed — especially on heavily compressed material.

What’s actually happening?

It’s important to understand that this technique doesn’t recover information removed by the MP3 codec. That’s not possible through conventional processing.

What it does is generate new content, harmonically related to the original signal, and place it precisely in the region where the spectrum was left incomplete.

Perceptually, the ear interprets that new information as a more natural extension of the existing content. That’s why the result tends to feel less artificial than simply applying a high-shelf EQ.

When is this useful?

This method tends to work especially well on:

• Low-bitrate MP3s.
• Audio downloaded from WhatsApp or Telegram.
• Old material with high-frequency loss.
• Overly compressed stems.
• Low-quality mix references.
• Quick restoration when the original file isn’t available.

Limitations

Not every audio file responds the same way. If the material already has enough high-frequency content, this technique can produce excessive brightness or an artificial character.

It’s also not a substitute for AI-based restoration, and it can’t recreate exactly the information removed during compression. Think of it as a perceptual reconstruction tool, not a mathematical restoration.

Conclusion

We often talk about «recovering the highs,» when in reality we’re just boosting whatever little survived compression.

This technique takes a different path: using the audio’s own harmonic structure to generate new content, coherent with the original material, and integrating it carefully through spectral analysis and minimal processing.

It’s not about tricking the ear with excess brightness, but about giving back some of the sense of openness and detail that tends to get lost in aggressively compressed formats.

As far as I could research, I couldn’t find a description of this exact workflow anywhere. It’s a methodology I developed during my restoration work, and it’s given me good results on material with high-frequency loss.

Luigi Manzoni is a Paraguay-based music producer, mixing and mastering engineer, and multi-instrumentalist. He runs Billow Ma, his recording studio in Asunción, and has worked on 500+ music credits for artists on Universal Music, IDS Music, and ONErpm. Get in touch to discuss a mixing, mastering, or production project.

Also available in Spanish.