Convert an MP3 File to MIDI: What If One Note Is Too Quiet?

If a quiet note vanishes when you convert an MP3 file to MIDI, try “More notes” and compare the new draft with the recording. In a controlled four-note MP3 on this site, Cleaner and Balanced each returned three notes; More notes recovered the deliberately quiet second note. It also uses a looser detection threshold, so inspect the whole result for extras before downloading.

Choose your MP3 in the local browser converter, start with Balanced, inspect the piano roll, then change just the sensitivity if a note is missing. This page tests that specific missing-note decision. The separate format test asks whether different MP3 bitrates changed timing for a different, even-volume clip; it does not answer what happens to a weak note.

What I tested on this machine

On September 25, 2026, I generated a 2.4-second mono WAV at 44.1 kHz with four separate sine-wave pitches: MIDI 60, 64, 67, and 72. Each began at 0.00, 0.60, 1.20, and 1.80 seconds and lasted 0.45 seconds. The second note's amplitude was 0.012 while the other three were 0.35—about 29 times smaller in sample amplitude. I encoded the single WAV as a 192 kb/s MP3 using FFmpeg/libmp3lame. I decoded that MP3 to float PCM with FFmpeg, then called the site's unchanged downsampling, detection, and MIDI-writing functions once for each sensitivity setting. This is a test of the real detector code, but not a test of the browser's AudioContext decoder or file-picker interface.

In More notes, the recovered E4 began at 0.557 seconds, rather than its generated 0.600-second onset, and lasted 0.418 seconds rather than the generated 0.450 seconds. So restoring the pitch did not make its timing exact. Cleaner and Balanced produced byte-identical 61-byte MIDI files in this test; More notes produced a different 69-byte file. No extra pitches appeared in this deliberately clean clip. That does not mean the loose setting is safe on a noisy recording.

A three-pass check for your own MP3

1. Listen to a short, isolated phrase first. Identify one quiet note by ear and mark roughly where it starts. If several instruments overlap, isolate the one you want before asking a single-track detector to follow it.
2. Convert once in Balanced. Compare the preview with your source. If a quiet note is absent but louder notes are correct, keep that draft and run the same file in More notes. Do not change source, bitrate, and sensitivity all at once; then you cannot tell what mattered.
3. Compare both drafts note by note. Keep the recovered note only if its pitch and placement match the audio. Delete any new fragments that do not belong. If both settings fail, a clearer recording or manual note entry may be quicker than repeated automatic passes.

For a successful conversion, the question is not simply whether a .mid downloads. The file must contain the notes you needed, at usable times. The MIDI export check covers file structure and post-download inspection; the main guide covers broader source and cleanup choices.

Inspect and repeat the experiment

Download the generated WAV, the test MP3, and each result: Cleaner MIDI, Balanced MIDI, and More notes MIDI. The observation file lists every detected pitch, onset, duration, file size, and SHA-256 hash. The local reproduction script is tests/quiet-note-mp3-research.mjs in the site project. It creates the tone sequence, encodes and decodes the MP3 with FFmpeg, and runs the same detector functions extracted from dist/script.js.

This is one synthetic, monophonic example. It cannot establish an accuracy percentage for songs, vocals, bends, background noise, or other browser decoders. A real quiet note might still be missed in More notes, and a noise transient might become a false note. Use the side-by-side audio and MIDI, not the setting name, to decide what to keep.