The short answer: no. Changing your file's sample rate between 44.1 kHz, 48 kHz and 22.05 kHz does not change the notes this converter writes to MIDI. Every input is collapsed onto the same internal 22 050 Hz analysis grid before a single note is read, so the pitch detector sees the same signal no matter which rate you started from. We ran the converter's own detector on the same melody at all three rates and it returned an identical 10-note sequence every time. Pick the rate your DAW or recorder already uses; it is not something you need to "fix" before converting. If the question behind yours is really about codecs, the codec matters more than the rate — see MP3 vs WAV for note loss.
What the converter does to sample rate on the way in
The detector does not read your file at its native rate. It resamples to a fixed working rate first, then measures pitch against that grid. Two consequences follow directly:
- Any input above 22 050 Hz is pulled down to 22 050 Hz. That covers 44.1 kHz and 48 kHz, which are the two rates you actually meet in exported audio.
- The analysis window is then fixed: a 2048-sample FFT stepped forward 256 samples. At 22 050 Hz that is one frame every 11.610 ms and one frequency bin every 10.77 Hz — the same grid whether you arrived from 44.1k or 48k.
The pitch search itself only looks between 55 Hz and 1800 Hz. The Nyquist limit at the 22 050 Hz working rate is 11 025 Hz, so the entire band the detector cares about sits far below the point where any resampling artefact could appear. That is the structural reason the rate stops mattering.
The integer case vs the non-integer case
Two of the rates resample differently, and it is worth being precise about why neither one changes your notes:
- 44.1 kHz → 22.05 kHz is an exact 2× decimation. Halving the rate is a clean integer step, so the downsample is about as gentle as resampling gets.
- 48 kHz → 22.05 kHz is a non-integer ratio (about 2.177×). This needs interpolation between samples rather than simple dropping. It is the harder of the two, so if sample rate were going to shift a note, 48 kHz is where you would expect to see it first.
Because the content the detector reads tops out at 1800 Hz and the working Nyquist is 11 025 Hz, even the non-integer 48 kHz resample has no energy left to alias into the search band. The interpolation can nudge the tiniest details of a waveform; it cannot move a 110 Hz note to a different semitone. The measurement below shows that difference staying below the threshold of detection.
What we actually measured: one melody, three rates
Claiming a number you cannot reproduce is worthless, so we ran it. We synthesised one ascending melody — A2, E3, A3, C4, E4, G4, A4, C5, E5, A5 — as a clean multi-harmonic tone, rendered it to PCM at 44.1 kHz, 48 kHz and 22.05 kHz, and pushed each file through this site's own pitch detector (dsp.js), monophonic mode, sensitivity 60, shortest note 60 ms:
| Source rate | Resample to 22 050 Hz | Bin width | Frames analysed | Notes detected | Note sequence |
|---|---|---|---|---|---|
| 22.05 kHz | none (already at grid) | 10.7666 Hz | 397 | 10 | A2 E3 A3 C4 E4 G4 A4 C5 E5 A5 |
| 44.1 kHz | 2× integer decimation | 10.7666 Hz | 397 | 10 | A2 E3 A3 C4 E4 G4 A4 C5 E5 A5 |
| 48 kHz | 2.177× non-integer | 10.7666 Hz | 397 | 10 | A2 E3 A3 C4 E4 G4 A4 C5 E5 A5 |
All three came back with the same 10 notes in the same order, and even the internal grid was identical — matching bin width (10.7666 Hz) and matching frame count (397) tell you the three files had already become the same signal before the detector looked at them. The one rate that is hardest to resample, 48 kHz, did not drop or shift a single note. For this converter, sample rate is not a variable that reaches the MIDI.
Why this is the honest answer, not a lucky take
A single clean line is the easy case, so the fair question is whether the rate would matter on busier material. It would not, for the same structural reason: every input above 22 050 Hz is forced onto that one grid, and the search band (55–1800 Hz) lives inside a working Nyquist of 11 025 Hz with huge headroom. On a dense mix the things that actually change your notes are the mix itself — overlapping parts, reverb, noise, how many voices are sounding at once — not whether the file was 44.1k or 48k. If you want to read about what does move the result, monophonic vs polyphonic covers where notes get lost, and how to check a conversion shows how to read the note list back.
The one case that does change things: files below 22.05 kHz
The resample step only pulls down. A file that is already below 22 050 Hz — 16 kHz or 11.025 kHz, which you sometimes get from old voice memos or telephony audio — is analysed at its own rate, with no upsampling. That genuinely changes the grid: at 16 kHz one bin is 7.81 Hz wide instead of 10.77 Hz, so the low end reads a little differently. That is outside the three rates this page addresses, but it is the real exception, and it is why a low, muffled recording is worth upscaling or re-exporting before you convert. Above 22.05 kHz the rule holds: the rate does not reach your MIDI.
How to test it on your own file
You should not trust any published accuracy figure you cannot reproduce. This takes about two minutes:
- Take one clip and export it at 44.1 kHz and again at 48 kHz from the same source.
- Convert both here with identical settings — same mode, same sensitivity, same shortest-note value.
- Download the .csv for each. The columns are
index,midi_note,note_name,start_sec,end_sec,duration_sec,velocity. - Diff the
note_namecolumns row by row. On a normal take they will match exactly, which is the finding: the rates agree because they become the same grid.
If you ever see a mismatch, check the pitch of the differing note. A difference concentrated in the bass is the analysis resolution acting below roughly 181 Hz, shared by both files — not the sample rate. Either way the cause is the window, not the rate you exported at.
Questions
Should I export at 48 kHz to get better MIDI?
No. Above 22 050 Hz the converter resamples everything to one grid, so a 48 kHz export and a 44.1 kHz export of the same take return the same notes. Spend the effort on a cleaner source and a quieter room instead.
Does a higher rate help with low bass notes?
Not through this converter. The bass limit comes from the 10.77 Hz bin width at the fixed 22 050 Hz working rate, which is identical for every input above 22 050 Hz. More rate above the working rate cannot buy finer low-end resolution than the window already sets.
What if my file is 96 kHz or 192 kHz?
Same rule: it is pulled down to 22 050 Hz before analysis, so it lands on the same grid as 44.1 kHz. You gain nothing in the MIDI and only wait longer for the resample. Export at 44.1 kHz and you skip the wait.
Is 22.05 kHz the best rate to convert from, then?
It is the one that needs no resample, so it is marginally the most direct path — but in our measurement it returned the same notes as 44.1k and 48k. Convert from whatever rate your recorder already produced; resampling it down yourself adds nothing.
Why does the converter resample to 22 050 Hz at all?
It keeps the analysis cheap enough to run entirely in your browser, and it still leaves a Nyquist limit of 11 025 Hz — well above the 1800 Hz ceiling the pitch search uses. The resample is what makes the source rate irrelevant, not what costs you notes.
Convert a file now — runs in your browser, nothing is uploaded.
Related guides
- MP3 to MIDIWhy MP3 loses notes that WAV keeps
- WAV to MIDIHow to prepare the file before you convert
- Mono vs polyWhat each detection mode misses
- Audio to MIDI converterHow accurate a conversion really is
- Pitch bend detectionWhat a glide adds to your MIDI, and when it matters
- Song to MIDIWhat a full mix comes back as
- MIDI converterWhat is actually inside a MIDI file
- Clean up a converted MIDIThe 5 passes that turn a raw export into a usable part