Microphone to MIDI: What This Converter Does With a Recording

The short answer: this converter does not listen to a live microphone. There is no real-time, in-browser microphone capture here — whether you call it mic to MIDI or MIDI mic, the answer is the same: record the take first, then convert the file. You record your instrument or voice to a file first — with your phone, a DAW, or your operating system's own sound recorder — and then drop that file in and it turns into a MIDI file. That boundary is worth stating up front, because "microphone to MIDI" can mean two different things and this tool does one of them. Once you have a recording, though, the noise behaviour is well measured: in our own tests, polyphonic mode recovered a fixed 32-note melody at 32/32 across clean audio and at 30 dB (a quiet room), 17 dB (a living room) and 6 dB (a busy café) signal-to-noise ratios, while monophonic mode held 32/32 only down to 30 dB and then collapsed to 0/32 at 17 dB.

Why there is no live microphone input

This is a file converter, not a live capture engine. The front page accepts an audio file — WAV, MP3, OGG, M4A or FLAC — decodes it in your browser, reads it frame by frame, and writes out a MIDI file. There is no code path that opens getUserMedia, holds a microphone stream, or streams audio into the detector as you play. So if what you want is "play now, see notes now", this tool will not do that half. What it will do is take the recording you already have and turn it into editable MIDI.

That is a deliberate boundary, not a missing feature we are hiding. A live microphone brings echo cancellation, device permissions, and a moving noise floor that a file-based detector is not built to handle. Keeping the input as a finished file is what lets the whole thing run locally with nothing uploaded.

How to get from microphone to MIDI

Record your playing or singing to a file. A phone voice-memo app, a DAW armed on one track, or your OS sound recorder all work.
Export or copy that recording as one of the supported formats: WAV, MP3, OGG, M4A or FLAC.
Open the converter and drop the file in (or click to choose it).
Pick a detection mode and sensitivity, watch the piano roll, and download the .mid or .csv.

The file step is the whole trick. Whatever you point a microphone at — an amp, a piano, your own voice — becomes a static recording first, and then the conversion runs on that recording exactly as it would on any other audio file.

What room noise does once your take is a file

Claiming a number you cannot reproduce is worthless, so we measured it with this site's own pitch detector (dsp.js). We synthesised a fixed 32-note melody and mixed it with background noise at four levels — clean, a quiet room (30 dB signal-to-noise), a living room (17 dB) and a busy café (6 dB) — then ran each through the converter at its default settings:

A fixed 32-note melody recovered out of 32, at this converter's default settings, as background noise rises. Measured with this site's own detector.
Recording	SNR	Polyphonic recall	Monophonic recall
Clean	—	32 / 32	32 / 32
Quiet room	30 dB	32 / 32	32 / 32
Living room	17 dB	32 / 32	0 / 32
Busy café	6 dB	32 / 32	0 / 32

The pattern is clear and it is not symmetric. Polyphonic mode rides the melody's strong spectral peaks above the noise, so a continuous line stays solid even in a café. Monophonic mode tracks a single spectral peak and has no harmonics to lean on, so once noise pushes its salience below the 0.20 gate it drops the whole line at once — clean and quiet-room recordings are fine, but at 17 dB it returns nothing.

Why noise on its own is worse than noise in a song

One result surprises people: play only the background noise, with no melody on top, and polyphonic mode returns a pile of spurious notes — 473 of them across the three noisy levels. That is far worse than the same noise mixed into a melody, where it produced zero false notes. The reason is the voice gate. The detector sets its noise floor from the quietest 10% of frames in the file, so a melody's own silent gaps between notes lift that floor and quietly suppress the noise sitting in the gaps. A file of pure noise has no silent gaps, so nothing lifts the floor and the detector keeps scoring noise as if it were music.

The practical reading: a real recording — anything with notes and the silences between them — is far safer than a continuous wash of room tone. The gaps are doing useful work.

Mono or poly for a noisy live take

If you recorded a single line — one voice, one horn, one guitar — in a less-than-silent room, reach for polyphonic mode rather than monophonic. The table above is the reason: polyphonic held the line at 32/32 even in a café, while monophonic fell to zero at 17 dB. Polyphonic mode looks for several spectral peaks at once, and a clear melody's peak sits well above the noise; monophonic mode's single-peak salience simply does not have that margin. The trade-off is that polyphonic mode on busy material can return a few extra notes that were not played, so check the piano roll before you trust the result. The room hurts in a second way too — reverb reflections smear note boundaries and can swallow fast notes in monophonic mode — which is the measured difference in live recording vs studio take: which transcribes cleaner.

How to record for the cleanest conversion

Get close to the source. A close microphone gives you a stronger signal over the room, which is exactly the SNR the detector needs.
Cut the reverb. A live room adds a long tail that blurs note boundaries. Recording in a smaller, drier space, or closer in, keeps notes separate.
Do not clip. A distorted peak flattens the harmonic structure the detector scores, which hurts both modes.
Mono is fine. The converter averages every input to mono before analysis anyway, so a stereo recording buys you nothing here.
Keep one line clear. For a single-instrument take, isolate it. The detector follows one strong line far better than a tangle of overlapping ones.

How to convert your recording

Drop the file in and let the piano roll draw. Move the sensitivity slider and watch what appears and disappears — on a noisy take, raising sensitivity keeps more of the quiet notes but also more of the noise, so the piano roll is your judge, not the note count. When you are happy, download the .csv and open it in a spreadsheet: the columns are index,midi_note,note_name,start_sec,end_sec,duration_sec,velocity. Diff that against what you played and you have a check you can actually reproduce.

Questions

Can this tool listen to my microphone in real time?

No. There is no live microphone capture in the browser. You record to a file first — phone, DAW or sound recorder — then convert that file. The conversion runs on the finished recording, not on a live stream.

Will a noisy room ruin the MIDI?

It depends on the mode. In our measurements a continuous melody came back at 32/32 in polyphonic mode even in a busy café (6 dB SNR), while monophonic mode collapsed to 0/32 at 17 dB. For a single live line in a noisy space, polyphonic mode is the robust choice; check the piano roll for a few stray notes.

Can I just hold my phone up to an amp and convert?

That is exactly the workflow this tool supports. The phone recording becomes the file; you then drop it in and convert. Record a clean take — close to the amp, not clipping — and the result is the same as any other audio file you would feed it. And once it is a file, you can run the whole conversion right on the phone — Can you do audio to MIDI on iPhone or Android? covers the input quirks and the on-device time cost.

Does recording to MP3 instead of WAV hurt the conversion?

MP3 adds a little coding noise on top of the room noise, but in our tests it did not measurably drop real notes on a buried line — the bigger lever is the recording itself. If you want the details, why MP3 to MIDI loses notes that WAV keeps works through the exact FFT reason.

Convert a file now — runs in your browser, nothing is uploaded.

Microphone to MIDI: what this converter actually does with a recording