MP3 to MIDI: Mono or Stereo, and What the Fold Costs
A note detector reads one waveform. Hand it a stereo file and the two channels are folded into one waveform first, so the fold decides what can be found. Three ways to fold, three different costs.
Mono or stereo, answered
- Mono is not a downgrade. A MIDI file cannot carry the second channel, so nothing is lost by dropping it before upload.
- The fold is arithmetic. Identical channels added together are 6.02 dB louder than either one; opposite channels add to silence; one channel delayed by a millisecond removes every odd multiple of 500 Hz.
- The MIDI file has no field for a stereo position. Its sixteen channels name parts, not speaker positions.
- If your only copy is stereo, check the level: two channels peaking at -3 dBFS each reach +3.02 dBFS when added.
Three ways to fold two channels into one
Which fold a converter performs is not stated on its page, so the arithmetic is what you need.
| Fold | Level if channels match | What it removes |
|---|---|---|
| Add, unscaled | +6.02 dB | Any content the two channels hold in opposite polarity |
| Add, then scaled back by 6.02 dB | Unchanged | The same opposite-polarity content |
| Keep one channel | Unchanged | Everything that exists only in the other channel |
Only one of the three can be happening to your file, and it is chosen for you.
Two identical channels are 6.02 dB louder than one
Doubling an amplitude is 20 times log10(2) in decibels, which is 6.02 dB. Take a mix whose loudest peak sits at -3 dBFS as the parameter: added without scaling, that peak arrives at +3.02 dBFS. Nothing above full scale can be written to a fixed-point file, so those samples are clamped and the shape above the ceiling is lost. Clipping cannot be repaired later: the discarded shape was never recorded.
Channels that are opposite add to nothing
If one channel is the other inverted, the sum is silence at every frequency. Where the opposition is partial or confined to a band, that band disappears and the rest is untouched. A level check cannot see it: the file looks ordinary.
One millisecond of delay nulls every odd multiple of 500 Hz
Adding a signal to a delayed copy of itself produces a comb, not a roll-off. The sum reaches zero whenever the frequency is an odd multiple of 1 divided by twice the delay. At a delay of 1 ms the nulls land at 500 Hz, 1,500 Hz and 2,500 Hz. One millisecond is 44.1 samples at 44.1 kHz, and in air at 343 m/s it is 34.3 cm of extra path.
Converted to note numbers by n = 69 + 12 times log2(f / 440), those nulls are notes 71.21, 90.23 and 99.08. The first sits between B4 and C5. That separates a comb from a codec ceiling: the 16 kHz cutoff sits above the highest note a MIDI file can hold, so it cannot delete a fundamental, while the lowest comb null is a note.
The file has no place to put a stereo image
A MIDI file carries sixteen channels and note numbers from 0 to 127, and channel 10 always sounds as percussion. The channels name parts — piano, bass, drums — not positions. The stereo field is discarded on the way in and cannot be recovered.
What to do with the file you have
Hand over mono when you can: it removes the level risk and the delay risk in one step. If stereo is all you have, compare the two channels before you upload: near-identical channels carry the level risk, near-opposite channels carry the cancellation risk, and a file whose width came from a delay carries the comb.
One caution about the word: on the converter sites read here, monophonic means one note at a time, not one channel.