MP3 to MIDI Settings Not Applied? Why the Download Ignores Them

Everything below was measured on this page's own converter running in a real browser at easyaudiotomidi.com. Real files were pushed into the file input, real mouse events were sent to the real Convert, Copy and Download buttons, and the page's own outputs were captured: the text it passed to the clipboard, the bytes it handed to the download, the status line with its colour class, and the four readouts. Where a number comes from the page's source rather than from a run, it says so.

Where each button gets its data

All three handlers are short, and the difference between them is the whole story. First, Convert. Every control is read inside the conversion, and the results are stored:

// inside convert(), after decoding
var threshold = 1 - parseFloat(sensEl.value);   // Confidence
var minMs     = parseInt(minmsEl.value, 10) || 0;
var notes     = framesToNotes(mono, TARGET_RATE, threshold, minMs);
var transpose = parseInt(transEl.value, 10) || 0;
var bpm       = parseInt(bpmEl.value, 10) || 120;

lastNotes    = notes;
var midi     = buildMidi(notes, bpm, transpose);
lastMidiBlob = new Blob([midi], { type: 'audio/midi' });

Then Download. It reads no control at all. It reuses the stored blob, takes the file name from the last file you loaded, and then overwrites the status line:

btnDownload.addEventListener('click', function () {
  if (!lastMidiBlob) { return; }
  var base = currentName.replace(/\.[^.]+$/, '') || 'audio';
  var url = URL.createObjectURL(lastMidiBlob);
  var a = document.createElement('a');
  a.href = url;
  a.download = base + '.mid';
  document.body.appendChild(a);
  a.click();
  a.remove();
  setTimeout(function () { URL.revokeObjectURL(url); }, 1000);
  setStatus('Saved ' + base + '.mid to your downloads.', 'ok');
});

And Copy, which is the odd one out. It reads Transpose at the moment of the click - and only Transpose:

btnCopy.addEventListener('click', function () {
  if (!lastNotes.length) { return; }
  var transpose = parseInt(transEl.value, 10) || 0;
  var text = noteListText(lastNotes, transpose);
  ...
});

So the page has three different freshness rules, not one. That is the source of every surprise in this article.

What I measured

One 4-second test file (an 8-note scale, C4 to C5) was converted with every control at its default: Tempo 120, Confidence 0.85, Transpose 0, Shortest note 60. The conversion reported 8 notes, C4 - C5, and produced a 105-byte .mid. The copied note list was 294 bytes across 9 lines, MD5 407ccb3701358c3906e1f179db0679f9.

Then, for each control in turn, I reset everything to the default, converted again, moved that one control without pressing Convert, and pressed Copy and then Download with real mouse events. Finally I pressed Convert and captured the same outputs again, so that every row has a "what the page would have given you" comparison next to "what it actually gave you".

The four-control matrix

Control moved after converting	Copied text changed?	Downloaded .mid changed?	Readouts changed?	Would Convert have changed the .mid?
Tempo 120 to 300	no	no	no	yes, 2ed25dfd...
Confidence 0.85 to 0.50	no	no	no	yes, 0fcb90dd...
Transpose 0 to +12	yes	no	no	yes, 1bd5932c...
Shortest note 60 to 400	no	no	no	no, identical

Read the third column across: no, no, no, no. All four downloads were the same 105 bytes and the same MD5 hash, which is the file from the original conversion. Whatever you moved, Download handed back the old file.

The last column is what makes that a bug rather than a design choice. For Tempo, Confidence and Transpose the setting you moved would have changed the file - pressing Convert produced different bytes in all three cases. Shortest note is the only one where the row ends in a tie, and that is a property of this particular file: its shortest note lasts 0.432 s, which is already longer than the 400 ms threshold, so raising the threshold could not remove anything. On a file with short notes the same change is dramatic, as the next section shows.

Three buttons, three freshness rules

Pulling the source and the measurements together, this is what each part of the page actually listens to:

What you look at or click	When it reads the controls
Notes found / Duration / Pitch range	Only during Convert. Moving a control does not touch them.
Piano roll preview	Only during Convert. It is a picture of the notes from that run.
Download .mid	Never. It reuses the stored file. Only the file name is read live, and it comes from the last file you loaded, not from a control.
Copy note list	Transpose only, at the moment you click. Every other column is from the stored notes.
Status line	Rewritten by Convert, by Copy and by Download. Download's message is always green.

The Copy row is why text and file can describe different music. That behaviour, taken apart byte by byte, is in what the copied note list actually contains, and the full behaviour of each control when you do press Convert is in what each control changes.

The green message that stops being true

This is the part that will cost you a bad file if you are not watching. The status line is the only summary the page gives you, and after a conversion it is a green success message with a note count in it. Moving a control leaves that message untouched, and pressing Download replaces it with another green message. Neither of them is re-checked against the settings on screen.

I used a 3-second test file made of notes of different lengths (20, 40, 60, 80, 100, 160, 320 and 640 ms, separated by silence). At the default Shortest note of 60 ms it converts to 6 notes. Here is the same file with Shortest note raised to 1000 ms - first without re-converting, then with:

Step	What the page says	File you get
Converted at the default 60 ms	green: Converted 6 note(s) from 3.0 s of audio in 0.1 s. Range A4 - A4.	90 bytes
Shortest note set to 1000 ms, no Convert	still green, same sentence, unchanged	-
Then Download	green: Saved D1.mid to your downloads.	90 bytes, the old 6-note file
Then Convert	red: Decoded 3.0 s of audio, but no steady pitch was found.	33 bytes, zero notes

The last two rows are the same settings. Without a re-conversion the page told me, in green, that it had 6 notes and saved a file - and that file was a real 6-note MIDI file. One click later the same settings produced a red failure and a 33-byte file containing no notes at all. The 90-byte file has MD5 67dbe6e84e7719a86b2b733afe6fd8b2; the 33-byte one is 6281fae2ae4d48a4a53e1e6cd062fa70.

A gentler version of the same thing: raising Shortest note to 400 ms on that file gave me a stale 90-byte file claiming 6 notes, where a real conversion gives 1 note and 43 bytes. And with a different test file - a 2-second tone whose fundamental is deliberately weaker than its second harmonic - moving Confidence from 0.85 down to 0.50 without re-converting left the readout saying A3 - A3 and handed me a 42-byte file at 8680179f..., while the real conversion of those settings reported A4 - A4 and wrote a different 42-byte file at 5f668c94.... Same length, same note count, one octave apart. Why a looser confidence can flip the octave like that is measured in the control breakdown.

What does refresh the page

• Convert to MIDI. This is the only action that re-reads all four controls, and it rebuilds everything together: the readings, the piano roll, the stored notes and the stored .mid file. Nothing else on the page does that.
• Clear. It empties the stored notes and the stored file, so after Clear the Download button does nothing whatsoever - no file, no message change. Clear is the only action that produces an empty state; a failed conversion does not. Those states are listed one by one in what the page leaves behind when a conversion does not work.
• Loading a different file. This changes the name the download will use, because the name comes from the last file you loaded. It does not change the contents, because the contents come from the last successful conversion. So you can end up with a file called after one track that contains another.
• Nothing else. Before the first conversion, Download is inert: I loaded a file, clicked Download, and the status line did not move from Ready: M1.wav. Press Convert to MIDI., and no file was produced.

The practical rule

1. Treat Convert as part of the setting change. If you touch any of the four boxes, press Convert before you copy or download. There is no partial update and no live preview.
2. Use the readouts as your proof. Notes found and Pitch range are the only on-screen numbers that tell you which conversion you are holding. If they do not match what the current settings should give you, you are holding an older one.
3. Copy and Download back to back. If the text is your record of the file, click both without touching anything in between - or re-convert first. Otherwise the Transpose box is the one control that can make the two disagree.
4. Do not trust the green line on its own. It reports the last conversion, and Download rewrites it to say "saved" regardless of whether the file still matches your settings.

Honest boundaries

• One browser, one machine. Every number comes from Chrome 153 on Windows, against the live page at easyaudiotomidi.com.
• The test audio is synthetic. Sines with a little harmonic content, no noise floor, no reverb, no real instrument. The staleness behaviour does not depend on the audio, but the note counts and file sizes in my examples do.
• I measured one value per control. Tempo 300, Confidence 0.50, Transpose +12 and Shortest note 400 / 1000 are the values I moved them to. I did not sweep the whole range of each control for this article - the complete range behaviour is in the control breakdown.
• The captures come from the page's own calls. The clipboard text is the string the page passed to navigator.clipboard.writeText, and the .mid bytes are the blob it passed to URL.createObjectURL when the download anchor was clicked. The download itself was blocked so nothing was written to disk.
• This is a description of what the page does today, not a promise about tomorrow. It is a behaviour I measured, and it may be changed at any time.

Open the converter and try the settings yourself