Hummingtomidi

Why Your MP3 to MIDI Sounds Bad

Five types of audio that break browser-based conversion — and what to do about each one.

Short answer: If your MIDI output sounds wrong, the audio you fed in probably falls into one of five categories that break monophonic pitch detection: distorted audio, heavy reverb, drum-heavy mixes, multiple instruments playing at once, and low-bitrate MP3. None of these are bugs in the converter — they are signal conditions where pitch tracking physically cannot work.

The converter runs in your browser, analyzing one pitch at a time using autocorrelation. When the input signal is clean and monophonic, it works well. When the signal is noisy, blended, or compressed, the detected pitch drifts, jumps, or produces garbage notes. The fix is almost always on the input side, not the converter side.

1. Distorted audio

Distortion — whether from a fuzz pedal, an overdriven preamp, or clipping during recording — adds harmonics that are louder than the fundamental pitch. The autocorrelation detector locks onto the strongest periodic signal, and in heavily distorted audio, that is often a harmonic, not the root note.

In testing, a guitar riff recorded through a fuzz pedal came out transposed up by a perfect fifth — the detector found the second harmonic (3:1 frequency ratio) and treated it as the fundamental. A clean recording of the same riff produced the correct pitches.

How to fix it

Use the cleanest version of the audio you can find. If you have the DI (direct input) signal, use that instead of the amped tone. If you only have the distorted version, expect notes to be off — sometimes by a consistent interval, which you can correct after the fact in your DAW by transposing the MIDI.

2. Heavy reverb

Reverb smears each note into a wash of decaying reflections. The detector sees the original note and its reflections simultaneously, and the pitch becomes ambiguous during the decay tail. You get wandering pitches, extra notes where none were played, and notes that refuse to end cleanly.

A vocal recorded in a bathroom (extreme reverb) produced MIDI with 40% more notes than the singer actually sang — the detector kept finding "pitches" in the reverb tail long after each phrase ended. The same vocal recorded in a dry room produced clean, accurate notes that matched the melody.

How to fix it

If you can choose between a dry and a wet recording, always pick dry. If you only have the reverby version, try trimming the decay tails — the converter detects false notes most often during sustained reverb, not during the attack. A shorter, punchier audio clip converts more cleanly than a long, ambient one.

3. Drum-heavy mixes

Drums are not pitched — or rather, they are pitched in ways that have nothing to do with the melody. Kick drums, snares, hi-hats, and cymbals produce broadband energy, transient spikes, and noise. The detector tries to find a pitch in that noise and fails in unpredictable ways: random notes, octave jumps, or complete silence during drum-heavy sections.

A pop song with a loud drum track produced MIDI that alternated between brief correct melodic fragments (during vocal-only sections) and garbage notes (every time the drums hit). The drums did not just add errors — they actively drowned out the melody the detector was trying to track.

How to fix it

Remove the drums before converting. Use a stem splitter or an instrumental/vocal isolator to strip the percussion, then convert the isolated melody. If you do not have a stem splitter, try converting just the intro or a breakdown section where the drums drop out — those sections often produce usable MIDI.

4. Multiple instruments playing at once

This is the same root cause as the chord problem — monophonic detection tracks one pitch. When a bass, guitar, and vocal play simultaneously, the converter picks whichever is loudest at each moment and outputs that as the melody. The result is a Frankenstein line that jumps between instruments mid-phrase.

A rock track with bass, rhythm guitar, and lead vocal produced MIDI that started on the bass line, jumped to the guitar during the guitar's louder passage, then latched onto the vocal during the chorus — three different melodies stitched into one unusable sequence.

How to fix it

Split the track into stems and convert each one separately. A single isolated instrument produces a clean, usable MIDI file. Five stems converted individually give you five MIDI tracks you can layer back together in your DAW. One full mix converted at once gives you one messy, useless track.

5. Low-bitrate MP3

MP3 compression at low bitrates (below 96 kbps) introduces artifacts: metallic ringing, smeared transients, and missing high frequencies. These artifacts confuse the pitch detector because the compression has altered the very waveform the detector relies on. At 64 kbps, a clean piano recording started producing notes that were a quarter-tone sharp — not enough to sound like a wrong note, but enough to sound out of tune against the original.

The converter decodes the MP3 in the browser before analyzing it. By the time the signal reaches the detector, the compression damage is already baked in — there is no way to recover the original waveform from a heavily compressed file.

How to fix it

Use the highest-quality version of the audio you have. A 320 kbps MP3, a WAV, or a FLAC file will always convert more accurately than a 64 kbps MP3. If the only available copy is a low-bitrate MP3, expect pitch errors in the 10–50 cent range (fractional semitones) that make the MIDI sound slightly out of tune.

What good input looks like

After testing these failure cases, the pattern is clear. The converter produces its best results when the audio is:

If your input meets all five criteria, the converter will produce a MIDI file that closely matches the melody you hear. If it violates even one, the output degrades — and the more criteria you violate, the worse it gets.

The honest bottom line

No browser-based MP3 to MIDI converter can rescue a bad input. The conversion happens in your browser — no upload, no signup, no server-side processing — and that means the analysis is limited to what a single-page web app can do with an audio buffer. Polyphonic separation, stem extraction, and artifact repair are all beyond that scope.

If your output sounds bad, check the input first. Nine times out of ten, the audio file is the problem, not the converter. You can also flip the direction with the MIDI to WAV converter to hear how a clean MIDI renders back into audio — useful for comparing what the detector found versus what you expected.

Want to test it on a clean file? Drop a solo melody and see the difference.

Try the MP3 to MIDI converter →