The short answer: before this converter detects a single note, it sums both stereo channels together and averages them — every sample becomes (left + right) / 2. That one step is invisible, and it is where a whole class of “missing notes” is decided before pitch detection even starts. Content that is out of phase between the two channels — a stereo-widened pad, a reversed side signal, a bass widened with a phase trick — adds up to zero and is gone for good. A part panned hard to one side loses 6.02 dB of loudness, which on this converter drops its velocity by 11 steps (measured: the same A3 note reads velocity 110 centred, 99 panned left). No sensitivity, mode, or quantise setting can recover any of this, because the cancellation happens on the raw samples before the analysis begins.
1. Where the mix-down actually happens
In dsp.js, the analysis pass opens with this: it walks every channel, adds its samples into a single buffer, and then — if there was more than one channel — divides every sample by the channel count. For a stereo file that is exactly (left + right) / 2.
Two things follow from that one line, and neither is shown anywhere in the interface:
- Mono-compatible stereo is lossless. If left and right carry the same signal, the sum is 2× the signal and the average brings it back to exactly the original. A mono file and a stereo file with identical left and right produce byte-identical MIDI.
- Anything the two channels disagree on is reduced. The average is a mix-down to mono. The more a sound lives in the side (the difference between left and right) rather than the middle (the sum), the more of it the mix-down throws away.
This is not a bug and not a quality choice specific to this tool — every pitch detector that starts from a mono mix-down inherits it. What is unusual is how little warning the interface gives: there is no “stereo” indicator, no channel-count label, and no note saying a file was reduced to mono. The first you hear of it is a missing note.
2. Three stereo layouts, three very different outcomes
To pin the numbers down we synthesised a sustained A3 (220 Hz) and ran the same note through three stereo layouts, then read the velocity the converter wrote out. The table is measured on the site’s own dsp.js, not estimated.
| Stereo layout | What the mix-down gives | Notes | Velocity |
|---|---|---|---|
| Centred (L = R = x) | x | 1 | 110 |
| Hard-panned left (L = x, R = 0) | x / 2 | 1 | 99 |
| Anti-phase (L = x, R = −x) | 0 | 0 | — |
The centred and mono cases are identical, down to the hash of the MIDI bytes — that is the “lossless” case. The hard-panned case is the same note, just quieter by exactly half the amplitude, which is −6.02 dB. On this converter’s velocity rule that is −11 steps, so a note that would read 110 lands at 99. The anti-phase case is the important one: left and right cancel completely and the note simply does not exist on the other side of the mix-down. There is nothing left for the FFT to see.
That last case is not an edge case for audiophiles. A lot of “wide” production is made exactly this way: one side of a pad or a bass carries the signal and the other carries its inversion, so the sound feels huge in headphones but collapses when summed to mono. Run that file through the converter and the wide part disappears while the centred part stays — which is the failure people describe as “some parts just didn’t convert.”
3. The side signal is the first casualty
We made the point directly: the same A4 pad, built once, sent through two layouts.
| Pad layout | Notes | Velocity |
|---|---|---|
| Side-only (L = +y, R = −y) | 0 | — |
| Centred (L = y, R = y) | 1 | 110 |
The pad is the same sound, the same loudness, the same duration. The only difference is whether its energy sits in the middle or the side of the stereo image. The centred one converts; the side-only one vanishes. This is why a track can sound perfectly balanced in your headphones and still come back from the converter with an entire part missing: your headphones keep the side and middle separate, the converter does not.
4. Partial cancellation: the in-between cases scale smoothly
Not every wide part is a perfect inversion. To see how the loss scales, we held left fixed and swept right from silence through −0.5× to −0.8× to full inversion:
| Right channel | Mix-down level | Notes | Velocity |
|---|---|---|---|
| R = 0 | −6.02 dB | 1 | 99 |
| R = −0.5 L | −12.04 dB | 1 | 88 |
| R = −0.8 L | −20.00 dB | 1 | 73 |
| R = −1.0 L | −∞ (silence) | 0 | — |
Two things stand out. First, the loss is continuous: there is no “threshold” below which a side-heavy part suddenly drops out — it just gets quieter in the mix-down until, at full inversion, it is silent. Second, because velocity is a log measure of loudness on this converter, a part that loses only half its amplitude (a −6 dB dip) still lands 11 velocity steps lower, and a −20 dB dip lands 37 steps lower. A wide-sounding pad can therefore survive the conversion as a much quieter version of itself rather than vanishing outright — and a quiet pad reads as a soft note that looks like a tracking error, not like a stereo problem.
5. A real mix: what actually survives
We put the two cases together into one file: a centred lead on the left and right, plus a wide pad whose energy sits only in the side. The converter came back with exactly one note — the lead. The pad was gone. Nothing in the output indicated that a second part had been present and removed; the MIDI simply had fewer notes than the music.
That is the operational takeaway. When a “convert audio to midi” result comes back thinner than the source, the first question is not the sensitivity or the mode — it is whether the missing part was wide. If it was, no control on this page will find it, because it was deleted from the samples before the controls ever saw it.
6. How to check before you convert
- Sum your own file to mono first, by ear. Play the track back through a mono fold-down (most DAWs have a “mono” button on the master). Whatever gets quieter or disappears there is exactly what the converter will lose.
- Watch the hard-panned parts. If a part lives on one side only, expect it to come back up to 6 dB quieter — and on this converter, up to 11 velocity steps lower. Bounce a centred version if you want the full loudness preserved.
- Collapse the stereo width on problem tracks. A gentle mid/side narrowing, or rendering the file to mono-compatible stereo before converting, is the one fix that actually works — because it happens before the mix-down, not after.
- Compare against the CSV, not just the piano roll. Export the CSV and check
velocity. A part that should be loud but came back uniformly soft is more likely a stereo-mix-down loss than a sensitivity issue.
Every one of those checks takes seconds and runs on your own machine, on the same source file, before you commit to a conversion.
Questions people actually ask
Why did some parts of my song not convert to MIDI at all?
Before detecting notes, this converter sums both channels and averages them to mono. Any part whose energy sits in the side of the stereo image — a widened pad, an inverted-side effect, a phase-trick bass — cancels in that sum and never reaches pitch detection. It is not a sensitivity problem and no setting brings it back; the part has to be centred (or the width reduced) before converting.
Why does my hard-panned guitar come back quieter?
A part panned fully to one side is averaged with a silent channel, so it loses half its amplitude in the mix-down — exactly −6.02 dB. On this converter’s velocity scale that is 11 steps: measured, the same note reads velocity 110 centred and 99 panned left. It is the same note, recorded quieter by the stereo fold-down.
Why does a track sound fine in headphones but convert thin?
Headphones keep the middle and side of a stereo image separate, so a wide part sounds full. The converter collapses them together, and the side-only energy cancels. The mix you hear is not the mix the converter sees; a mono fold-down preview shows you the difference in advance.
Can I fix the missing notes with the sensitivity slider?
No. Sensitivity, mode and quantise all operate after the mix-down, on samples that have already been summed. A part that cancelled in the sum is gone from the samples themselves, so no downstream control can recover it. The fix has to happen before conversion — narrow the width or centre the part, then re-run.
Does a stereo file come back as two separate MIDI channels for left and right?
No. The mix-down to mono happens first — the two channels are summed and divided by two — so by the time pitch detection runs there is no left and no right, only one summed signal. Proof it is a true mono fold and not a stereo-aware step: we converted the same A3 tone as a mono-compatible stereo file (left equal to right) and as a plain mono file, and the two exports were byte-for-byte identical (same SHA-256). On a real mix, a centred lead plus a side-only pad came back as a single A3 note — the pad vanished, and the MIDI carries no marker saying a second part was ever there. If you need both parts, convert them as separate stems; before converting, run a mono fold-down preview in your DAW, because anything that goes quiet or disappears there is exactly what the converter will drop.
Can I listen to the MIDI this tool produces?
Yes — but only as a preview, not as a file. After converting, the page shows a Play button that synthesises the notes in your browser with a triangle-wave oscillator and routes them to your speakers, so you can audition what was captured. What it exports, though, is only a .mid and a .csv — the converter has no WAV or MP3 encoder, so it never writes an audio file. If you need an actual audio file, take the .mid into a DAW or a free SoundFont synth and bounce it there. The MIDI this converter gives you is already quantised to a 1/16 grid at 120 BPM (TPQ 480), so it drops straight into any sequencer without re-quantising.
Can I convert audio to MIDI in Cubase?
Yes. The converter is a web page, not a Cubase plugin, so there is nothing to install inside the DAW — you export the part you want as a WAV or MP3, then open this page in any browser and load that file. Detection runs in the browser tab, so the notes come back as a .mid you can drag straight into your Cubase project. Because the analysis is client-side, the result is the same whether the audio came from Cubase, Reaper or any other host, and nothing is uploaded while it is processed.
Can I convert audio to MIDI in Reaper?
Yes. Reaper has its own capture, but this converter is separate: render the track or stem you want as a WAV or MP3, open the page in a browser, and drop the file in. The pitch detection happens in your browser, so you get a .mid without sending the audio anywhere, and you can import that file back into your Reaper project as editable notes. Working from a stem rather than the whole mix gives a cleaner result, because the converter keeps the loudest part in each frame and buries quieter ones.
Can I convert audio to MIDI in Studio One?
Yes. Studio One's built-in extractor follows one pitch at a time, but this converter handles several at once — up to six simultaneous notes. Export the audio you want converted (a clip, a track, or a stem) as WAV or MP3, open this page in a browser, and load it; the notes return as a .mid you can pull into Studio One. Because the conversion runs in the browser, no plug-in is added to your project, and the file itself never leaves your machine during analysis.
Can I convert audio to MIDI in Pro Tools?
Yes. There is no Pro Tools plug-in to install — bounce the audio you want as a WAV or MP3, open this converter in any browser, and load the file. The analysis runs entirely in the browser tab, so you receive a .mid you can import back into Pro Tools as a MIDI track, with nothing uploaded in the process. For the most complete capture, bounce the instrument on its own rather than inside a full mix, since the detector gives each frame's loudest part priority and quieter instruments can get lost.
Drop a file in and watch the piano roll — the Audio to MIDI Converter runs entirely in your browser, so the mix-down behaviour above is yours to reproduce on any stereo file in a few seconds.
Related guides
- WAV to MIDIHow to prepare the file before you convert
- MP3 to MIDI converterHow loud a note has to be before it hits velocity 127
- Song to MIDIWhat a full mix comes back as
- Vocal to MIDIWhy a sung note comes back an octave up
- MP3 to MIDIWhy MP3 loses notes that WAV keeps
- OGG to MIDIWhat the container actually changes (and what it doesn't)
- MP4 to MIDIWhy the AAC audio, not the video, decides your notes
- YouTube to MIDIWhat a full-length soundtrack comes back as
- Video to MIDIWhat happens to the part you cannot hear