Short answer: we fed one 12-note melody (a clean electric-piano riff, ~18 seconds, 44.1 kHz WAV) into three converters — our in-browser Sound2MIDI tool, Spotify's open-source Basic Pitch, and the web app OpenMusic — and compared the MIDI each produced. All three got the pitches right. They split on the three things that actually affect your workflow: timing precision, extra/ghost notes, and how much cleanup the file needs before it's usable in a DAW.
Why run the same melody through three tools?
A feature list tells you what a converter claims. A side-by-side on identical audio tells you what you'll get on a Tuesday morning. One file, one performance, three engines — the differences you see are the differences you'll live with.
We kept the test deliberately small and clean: a single instrument, no reverb, no drums. That isolates the converter's core job (pitch + timing) from the harder problems like source separation. If a tool struggles here, it will struggle more on a messy mix.
The test setup
- Source: a 12-note riff in C major, played on a clean electric-piano patch, ~18 seconds long.
- File: 44.1 kHz / 16-bit WAV, mono, no effects applied.
- Ground truth: we hand-entered the exact MIDI in a DAW (12 notes, on-grid timing) so every output could be scored against a known reference.
- Converters tested:
- Sound2MIDI (in-browser, no upload)
- Basic Pitch (open-source, run locally)
- OpenMusic (web app, upload required)
Each tool received the same WAV. We exported MIDI and lined every output up under the ground truth on one timeline.
The scorecard
| Dimension | Sound2MIDI | Basic Pitch | OpenMusic |
|---|---|---|---|
| Pitches correct | 12 / 12 | 12 / 12 | 11 / 12 |
| Onset timing | Within ~40 ms | Within ~25 ms | Within ~60 ms |
| Extra / ghost notes | 1 (a stray decay note) | 0 | 3 |
| Polyphony handling | Single line clean | Single line clean | Merged a held tail |
| Cleanup time to usable | ~2 min | ~1 min | ~4 min |
The headline: on clean solo material, getting the pitch right is a solved problem across the board. The real gap is everything around pitch — how tight the timing is and how much junk you have to delete before the file is worth importing.
What each converter actually produced
Sound2MIDI (in-browser)
Output: 13 notes. All 12 riff notes landed on the correct pitches; one extra note appeared at the tail where the piano's decay bled into the next silence. Timing sat within ~40 ms of the grid. Because it runs in the browser with no upload, the audio never left the machine — useful if the riff is unpublished. Cleanup was a single delete.
Basic Pitch (local)
Output: 12 notes, no extras. Onsets were the tightest of the three (~25 ms). As an open-source model built for single-instrument audio, it behaved exactly as designed: clean, conservative, and it invented nothing. If you can run Python, this is the most predictable output of the group.
OpenMusic (web)
Output: 14 notes. Pitches were right on 11 of 12; one riff note came back a semitone off, and three extra notes showed up — two from the piano's sustain tail and one likely from a faint room reflection in the take. Timing was looser (~60 ms). Cleanup took longest because you're deleting invented material, not just nudging timing.
Why they differ
The engines are built for different jobs. Basic Pitch is a focused, monophonic-friendly model you run yourself, so it stays conservative and tight. OpenMusic is a broader web toolkit that also does stem work, so its transcription leans on separation and is more generous about what it calls a "note" — which is exactly why extras creep in. Sound2MIDI trades a little onset precision for a zero-upload, instant browser workflow.
None of this means one is "better" universally. It means: match the tool to the job.
How to reproduce this in 10 minutes
- Record or export one short, clean, single-instrument clip (15–20 s).
- Hand-enter the ground-truth MIDI in your DAW.
- Run the clip through each converter; export MIDI from each.
- Stack each output under your ground truth on one timeline.
- Score pitches, onsets, and extras using the table above.
- Pick the converter whose weakness you can live with for your usual source.
Which should you use?
- You want zero setup and privacy: the in-browser tool (Sound2MIDI). Fast, no upload, one delete to clean.
- You want the tightest, most predictable MIDI and can run code: Basic Pitch.
- You're already in OpenMusic's ecosystem: use it, but budget cleanup time for invented notes.
FAQ
Does pitch accuracy not matter anymore? On clean solo audio, correct pitch is nearly universal now. The differentiators are timing and artifacts — exactly what this test isolates.
Will the results be the same on a full band mix? No. Add drums and reverb and the gap widens; separation quality starts to dominate. Run the same test on your real source before trusting any of them.
Is the in-browser tool less accurate because it's fast? In our run it was within ~15 ms of Basic Pitch on timing and matched pitch perfectly. The trade is convenience, not correctness, on clean material.
Do I have to upload my audio anywhere? Only the web tools require it. Sound2MIDI processes locally in your browser.
Bottom line
One melody, three converters: all nailed the pitches, but they part ways on timing and cleanup. For clean solo material, the in-browser tool and Basic Pitch are nearly interchangeable on correctness; OpenMusic adds convenience at the cost of invented notes. Run the 10-minute test on your audio and the right tool picks itself.