A 9.1-second solo vocal from the vocadito dataset was converted with mp3tomidiflow.com in Chrome. The MIDI downloaded successfully and contains 40 note events. Successful download does not mean note-perfect transcription.
This is one monophonic vocal example, not a test of piano, chords, drums or full songs. Both human annotations and the unedited output are shown, including differences. No claim about overall accuracy or time saved.
MIDI stores notes and timing. Open it in a DAW with an instrument to hear it; it does not contain the original voice.
Comparison with two human annotations
Match rules, set before conversion: pitch difference ≤50 cents; onset difference ≤50 ms (strict) or ≤200 ms (diagnostic). Maximum one-to-one matching. Duration accuracy is not scored. The two annotators label different note boundaries, so neither table is a product-wide accuracy score.
Onset tolerance
Annotator
Reference notes
Matched
Unmatched reference
Unmatched output (of 40)
50 ms
A1
35
14
21
26
50 ms
A2
27
10
17
30
200 ms
A1
35
22
13
18
200 ms
A2
27
22
5
18
“Unmatched” can reflect pitch errors, timing differences or note splitting; it does not classify every unmatched event as a wholly wrong musical note. Review the output rather than deleting notes solely from these counts.
Note timeline
Horizontal position is time; higher notes are higher on each chart. Hover a bar for pitch and duration. The output also contains a MIDI pitch 82 event absent from both reference pitch ranges.
Human annotation A1
Human annotation A2
Actual converter output
How to check and edit the result
Import the downloaded MIDI into your DAW and select an instrument sound.
Listen to the source recording, then check pitch and note starts on the piano roll.
Inspect the isolated high note and split events in context before editing. MIDI contains detected notes, not the original voice.
Treat the download as an editable draft. Check the pitches, starts and split notes in your DAW. This test does not establish that conversion is faster than manual transcription, or that paid/free alternatives are better or worse.
Source and reproducibility
Audio and annotations: Rachel Bittner, Katherine Pasalo, Juan José Bosch, Gabriel Meseguer Brocal and David Rubinstein, vocadito, version v3 (2021), CC BY 4.0. Attribution retained; recording unchanged. The MIDI is machine-generated from that recording. This excerpt also appears in upstream developer fixtures; it is not represented as a held-out benchmark.
Production converter test: statistics off; native file selection, Convert once, Download once. No code/model settings changed. The original A1 note annotations and original A2 note annotations are available as CSV. Their columns are start time in seconds, pitch in Hz and duration in seconds. Download the unchanged audio and unedited MIDI above to inspect the same inputs and output. No endorsement by the vocadito authors or Spotify is implied.