Audio Repair and Restoration

Only have a low-bitrate MP3? How to improve it in REAPER (and what cannot be fixed)

By StudioRescue Editorial

Published October 27, 2026 · Updated October 27, 2026

This usually happens with an interview pulled from a phone app, a voice message, an old podcast episode where the project files are lost, or a recording someone sent through a service that compressed it on the way. The words are all there, but the sound is dull, smeared, and has a watery or metallic shimmer on the "s" sounds and in the background.

Everything below assumes the recording cannot happen again. If it can, record it again. That will beat any repair in this article.

What the encoder actually removed

An MP3 encoder saves space by deciding what you are least likely to hear and discarding it. At a low bitrate it has to discard a lot, and two kinds of damage follow.

1. A ceiling on the top end

The LAME encoder documentation says a lowpass filter "helps reducing the amount of data to encode", and that its default "depends on the target bitrate/quality". In general, the lower the bitrate, the more the encoder has to give up, and LAME also notes it "may sometimes resample automatically when faced with extreme compression conditions". Either way, everything above the cutoff was never stored. That is why the file sounds dull, and why no amount of treble boost makes it bright.

2. Artifacts in what is left

The remaining audio is stored with too little data to describe it accurately, so the errors become audible. Common descriptions are ringing, a metallic or swirling shimmer, and pre-echo, a faint smear just before sharp sounds like a "t" or a hand clap. These are not noise sitting on top of the voice. They are part of how the voice is now encoded, which is why noise reduction tends to make them worse, not better.

First, look for a better copy

Ten minutes of asking around beats an evening of processing. Before you open REAPER, check:

• The device that recorded it. A phone voice memo or a field recorder often keeps the original file even after a compressed copy was shared.
• The other person. If a guest recorded their side locally, ask for that file directly, through a file-sharing link rather than a messaging app, so it is not compressed again in transit.
• The platform. Some call and podcast recording services keep a separate higher-quality track per speaker. The options for that kind of recording are covered in fixing a remote guest's bad audio in REAPER.
• Older exports. If this is an old episode, look for the WAV you rendered before the MP3, in the project folder or a backup drive.

If none of these turns anything up, carry on.

Three checks before you process anything

These tell you how much room you actually have to work in.

Stop it getting worse

Every time you save to MP3 again, the encoder makes a fresh set of guesses about what to throw away, and the losses stack. The Generation loss reference states that re-encoding, switching formats, or changing bitrate all cause it. So:

• Import the MP3 into REAPER once, do all the work on it there, and render your working copies as WAV. The render options are covered in how to export audio from REAPER.
• Encode to MP3 only once, at the very end, for delivery. For a podcast, the format and bitrate choices are covered in podcast export settings in REAPER.
• Never edit, save, and re-edit the MP3 itself in a tool that writes back to MP3 each time.

The repair order in REAPER, with stock plugins

Everything here uses ReaEQ, which comes with REAPER. The REAPER User Guide describes its pass filters as filtering "out frequencies above/below the frequency setting" and its shelves as moving the gain "below/above the frequency setting". The numbers are starting ranges to adjust by ear on your file, not presets. Stop as soon as it sounds acceptable for its purpose.

1. Clean the low end

Add ReaEQ and set the first band to High pass at around 80 to 120 Hz for speech. The bass is rarely where the encoding damage shows, but phone and laptop recordings often carry rumble and handling noise down there, and removing it makes everything above it easier to hear.

2. Pull the swirl down, gently

This is the step that helps most with an artifact-heavy file. Add a High shelf band set a little below your ceiling, often somewhere in 6 to 10 kHz, and cut it by about 2 to 6 dB. Sweep the frequency slowly while listening to sibilants and pauses. You are looking for the point where the shimmer drops back without the voice turning to cardboard.

If the shelf is not enough, try a Low pass band instead, set just under the ceiling you found in the spectrogram. Above that line there was nothing left but artifacts, so filtering there costs little. Bring it lower only while the voice still sounds natural. This is a trade, not a fix: you are choosing slightly duller in exchange for less distracting.

3. Add presence only where content exists

If speech sounds muffled, a small boost of 1 to 2 dB somewhere in 2 to 4 kHz can help intelligibility, provided the spectrogram shows real content there. Do not boost above the ceiling or into the swirly band. You will raise the artifacts, not the detail.

4. Be careful with noise reduction

If there is steady hiss, very light noise reduction can help, using the method in reducing background noise in a podcast in REAPER. Keep it light. Compression artifacts sit right in the band noise reduction works on, so heavy settings tend to turn a mild shimmer into an obvious underwater sound.

5. Set the level last

EQ changes how loud things read, so set the final level after the tone is settled. For spoken word, aim at a published loudness target for your platform rather than at a meter position.

What about AI and "restoration" plugins?

Some paid tools go further than EQ. iZotope's Spectral Recovery, included in RX Advanced, is described by iZotope as helping "restore frequencies above 4k that are removed when the audio is compressed for real-time streaming". Its own product page is clear that those upper frequencies are re-synthesized. In other words, it generates plausible new high-frequency content rather than recovering the original. That can make speech sound brighter and more natural, and it can also add content that was never there. If you try a tool like this, compare it against the plain EQ version at matched loudness before you decide it is better.

StudioRescue has not tested Spectral Recovery or any similar tool for this article, so this is a description of what the vendor claims, not a recommendation.

What this will not fix

• It will not restore the top end. Anything above the ceiling was never saved. EQ can only work with what is left.
• It will not remove artifacts cleanly. Filtering makes them less noticeable by making the region they live in quieter, which also makes the voice a little duller.
• It will not undo other damage that happened before encoding, such as noise suppression or automatic gain control on the recording device. Those have their own limits, covered in what can and cannot be fixed when noise suppression was on.
• It will not make a sung vocal mix-ready. A spoken interview has a forgiving standard. A lead vocal in a song does not, and a low-bitrate copy is a reference, not a source.

Stop it happening again

The fix is upstream. Record in WAV wherever you can, keep the WAV even after you share a compressed copy, and treat MP3 as a delivery format only. For remote interviews, have each person record their own side locally and send the file, which is known as recording a double-ender. And if a recording ever matters, ask for the original file before anyone shares it through an app.