What we tested in the current converter
We generated a 4.55-second mono source at 44,100 Hz with isolated C4, E4, G4, and C5 tones. Each tone lasted 0.8 seconds, with 0.25 seconds of silence between tones. We encoded the same samples as constant-bitrate MP3 files at 32, 64, 128, and 320 kbps. No published recording, performer, title, or channel identifier was used.
| MP3 bitrate | Input size | Intended pitches found | MIDI result |
|---|---|---|---|
| 32 kbps | 18,495 bytes | C4, E4, G4, C5 | 6 events; 105 bytes |
| 64 kbps | 36,780 bytes | C4, E4, G4, C5 | 6 events; 105 bytes |
| 128 kbps | 73,560 bytes | C4, E4, G4, C5 | 6 events; 105 bytes |
| 320 kbps | 183,902 bytes | C4, E4, G4, C5 | 6 events; 105 bytes |
All four browser runs returned HTTP 200, created one musical track, and reported no console errors. The files had the same note count, but they were not byte-for-byte equivalent note performances: some starts, durations, and velocities changed after encoding and decoding.
What changed at 32 kbps
The first C4 began at 0.267 seconds in the 32 kbps result, compared with 0.232 seconds at 64, 128, and 320 kbps. The 32 kbps output also placed its second short fragment after C5, while the three higher-bitrate outputs placed a short fragment after E4. Those observations show that bitrate can affect boundaries even when the main pitch set and total event count stay unchanged.
Why 64 and 128 kbps matched here
The 64 and 128 kbps outputs contained the same six notes with the same parsed starts, durations, and velocities in this run. That is a result for this simple synthetic source, not proof that the two bitrates are interchangeable for music. Dense mixes, percussion, room sound, and overlapping instruments give a transcription model different evidence.
Did 320 kbps win?
No measured outcome on this file makes 320 kbps the winner. It was the largest input, but it still produced six events. Its timing was close to the 64 and 128 kbps results, while a few durations and velocities differed. Increasing the bitrate cannot restore detail that was already removed from an earlier low-bitrate source.
Which MP3 should you use?
- Keep the cleanest original you legitimately have. Avoid exporting through another lossy encode before transcription.
- Use an existing MP3 directly. Converting it to a higher bitrate or to WAV does not recreate discarded audio information.
- Review note edges. Compression may change short fragments and boundaries even when the intended pitches remain.
- Test a short representative section. Results from a simple melody do not predict a dense mix.
What this test does and does not prove
It verifies the current converter's output for four MP3 encodes made from one self-generated source. It does not establish an accuracy percentage, recommend a universal minimum bitrate, or promise that every 32 kbps file will preserve the correct pitches.
For the container comparison, see WAV versus 128 kbps MP3. If compression creates short boundary notes, use the extra-note cleanup test.