Cubase audio to MIDI: does export level change which notes come back

The short answer: no — turning the bounce level down in Cubase does not delete a single note, because every threshold this converter applies is a ratio, not a volume. We took one four-note phrase (C4 D4 E4 G4) and scaled it in fine steps: at −40 dB it returned 4 of 4 notes at velocity 39, at −50 dB, 4 of 4 at velocity 20, and even at −140 dB it still returned the same four pitches in both modes, with velocity clamped to its floor of 1. What level changes is loudness, not pitch: the only line in the whole pipeline that reads absolute amplitude is the velocity formula, and in the exported file that difference is exactly 8 bytes out of 89. Notes do eventually start dropping, but not because of the fader — they drop when a real fixed-point bounce runs out of bits. On a 16-bit export the four notes stay clean down to −57 dB and first break at −58 dB; monophonic mode reaches zero notes by −70 dB, while a 24-bit export held all four at −80 dB. So do not re-bounce to chase level: convert the quiet file here, then fix velocity in Cubase; if the part is very quiet, export 24-bit instead of 16-bit. Everything runs in your browser — no upload, no account, nothing to install.

1. Should you convert here, install something, or stay in Cubase?

Before the numbers, the honest decision. Three situations, three answers, each pinned to a value we measured on this converter rather than to an opinion about a host application.

Everything after this section assumes you are in the first case. That is the one where a browser converter is genuinely the fastest path.

2. The route, in three sentences

Export the region from Cubase as one audio file, at any sensible level and preferably 24-bit. Drop it on this converter, download the .mid and the .csv, then drag the .mid back into your Cubase project and load an instrument on the track. The rest of this page is about the one export choice — how far you turn the level down — and the single setting that actually decides whether notes are lost.

3. What we measured, and how

Every number below comes from running this site’s real dsp.js and midi.js in Node on synthesised tones whose correct answer we already knew. No third-party tool was measured, and no figure is estimated. The reference clip is four notes (C4 261.63 Hz, D4 293.66 Hz, E4 329.63 Hz, G4 392.00 Hz), 0.5 s each, with a mild harmonic series and 10 ms ramps, rendered at 22 050 Hz with a peak amplitude of 0.35. Unless stated: sensitivity 60, shortest note 60 ms, 120 BPM, quantize off; both monophonic and polyphonic mode are reported because they fail differently. Scaling the clip is a single multiplication of every sample by 10^(dB/20).

The analysis grid is fixed by the converter, not by your Cubase project: audio is resampled to 22 050 Hz, read with a 2048-point FFT stepped 256 samples at a time, so one frame is 11.610 ms (86.13 frames per second) and one bin is 10.7666 Hz. Monophonic search runs 55–1800 Hz; polyphonic peaks are read up to 4000 Hz with at most 6 peaks per frame. Your project’s 48 kHz session, its 96 kHz render and its 3/4 time signature do not enter that arithmetic — we ran the same melody at 44.1k, 48k and 22.05k and the note list was identical every time.

Our four-DAW export overview already measured this phrase at 0, −6.02, −12.04, −20.00, −26.02 and −33.98 dB and found four of four notes at every step, with velocity sliding from 113 to 50. That table stopped at −33.98 dB. This page picks up there and measures the lower levels people actually ask about, then goes looking for the level at which notes do start to disappear.

4. The low levels: −40 dB, −50 dB, and every step past them

Here is the direct answer to the export-level question. We scaled the identical four-note phrase down and re-ran the detector at each level. Nothing else changed — not the mode, the sensitivity, the shortest-note value or the tempo.

At −40 dB and −50 dB — the two levels this page was written to answer for — both modes returned exactly 60,62,64,67. We then pushed a fine sweep in 2 dB steps from −100 dB down to −200 dB: every single level still returned all four notes in both modes. There is no float-gain threshold at which this detector gives up. A level of −200 dB means the loudest sample is around 3.2 × 10−11, far below anything a DAW fader can produce, and the analysis is unchanged.

So if your Cubase bounce came back with missing notes, the fader is almost certainly not the cause. The usual suspects are a region trimmed too tight (why a short region returns no notes) or a mode that cannot see the material — the four-step troubleshooting order starts with format and mode for exactly this reason.

5. Why the notes survive: the gate is a ratio, not a volume

This is the structural reason, read straight from the converter’s own source. In segment() the loudness gate is

gate = floor + (rmsMax − floor) × (0.32 − 0.30 × sens)

where floor is the 10th percentile of RMS across this file’s frames and rmsMax is its loudest frame. At sensitivity 60 the multiplier is 0.14. Multiply every sample in the file by a factor k and every one of those quantities — each frame’s RMS, the percentile, the maximum — is multiplied by k too, so the gate becomes k × gate. The test rms[f] ≥ gate is therefore identical for any positive k. Measured across the entire sweep, gate / rmsMax stayed at 0.93213 from 0 dB to −140 dB, while rmsMax itself fell from 3.212e−1 to 3.212e−8.

The same holds for the other two gates. Polyphonic mode keeps a peak only if its magnitude reaches relThr × framePeak, where relThr = 0.14 − 0.11 × sens; that is a ratio of two magnitudes in the same spectrum. Monophonic mode compares a harmonic-salience score bestScore / total against 0.35 − 0.25 × sens; again a ratio. There is no absolute amplitude constant anywhere in the decision. Scaling the whole signal changes the FFT magnitudes on both sides of every comparison by the same factor, so the note boundaries do not move.

One consequence is worth stating plainly: a normalised bounce is not more accurate, and a quiet bounce is not less accurate. They produce the same note list. Normalising before export is a habit carried over from recording, where the noise floor is fixed by the hardware; here the floor is computed from the file itself.

6. The one thing level does change: velocity, down to a floor of 1

There is exactly one place in dsp.js where absolute amplitude is used. When a note is pushed, its velocity is computed as

db = 20 × log10(rms + 1e−9);  vel = round(20 + (db + 60) × 100 / 54);  clamp to 1..127

That is why a quiet bounce gives you a quiet MIDI part rather than a wrong one. We reconstructed the RMS the detector averaged for the first note and checked the formula against the actual exported velocity:

The slope is linear: 100/54 = 1.8519 velocity steps per dB, which is about 11–12 steps for every 6.02 dB of level — the same value the four-DAW overview measured over its first six steps. From −65 dB downward every note is velocity 1, the quietest value the format allows, so further attenuation is invisible in the file’s dynamics. Velocity here is derived from loudness, not from how the part was played, which is also why it is nearly flat across the four notes: why converted MIDI sounds robotic covers the feel, and how loud a note has to be before it hits velocity 127 gives the top of the curve.

The practical fix is not at export. Convert the quiet file, then scale the velocities in Cubase’s MIDI editor or with the logical editor. What the browser controls can and cannot change explains why this page cannot do that for you: its five controls re-run the segmentation, they do not edit the note events.

7. Where notes really start dropping: the bit depth of the bounce

Everything above is a pure gain change on floating-point samples. A real Cubase bounce is not floating point unless you ask for it: a 16-bit or 24-bit WAV or AIF stores the signal as integers, and the rounding error is a fixed absolute size that does not shrink with the signal. To measure that, we quantised each scaled clip to integer PCM referenced to full scale, clamped it, and decoded it back to floats exactly as the browser decoder would.

A finer sweep at 1 dB steps puts the first break at −58 dB: clean down to −57 dB, and at −58 dB monophonic mode drops from four notes to three. By −70 dB the signal spans only six integer steps, monophonic mode reports nothing, and polyphonic mode is still holding all four pitches. At −80 dB, with a two-step peak, even polyphonic mode breaks — it returns seven objects including octave ghosts (102, 104, 106) that were never played. The quantisation noise has become a spectrum of its own, and the harmonic-salience gate in monophonic mode is the first thing to fail. Why the two modes fail differently explains the mechanism.

Now the useful half. We repeated the identical sweep as a 24-bit integer bounce:

Not one note was lost at any level, in either mode. At −80 dB a 24-bit bounce still has 524 integer steps of peak where a 16-bit bounce has two. The variable that decides whether a quiet part survives is bit depth, not fader level. If you are bouncing a very quiet part in Cubase, choosing 24-bit (or 32-bit float) instead of 16-bit buys far more than turning the level back up. The file format itself is a separate question — one .m4a can be lossless ALAC or 32 kbps AAC, and that choice has its own measured effect, while how to prepare the input file covers the container side.

8. Can sensitivity rescue a very quiet 16-bit bounce?

Partly, and with a catch. Sensitivity moves the gate interpolation and the salience threshold, so on a quantisation-damaged file it can admit frames that the default setting rejects. We swept it on three low-level 16-bit bounces:

At −70 dB, sensitivity 80 finally gets monophonic mode back to four notes — but polyphonic mode grows to five, and at sensitivity 100 it returns 10 note objects, six of them ghosts created by the quantisation noise. That is the standard failure mode of raising sensitivity on damaged audio: it keeps more noise, not more real notes. The honest order is to fix the export first (24-bit, or a healthier level), then leave the sliders alone. If you have already let ghost notes into the file, the cleanup passes are what remove them, and the troubleshooting order puts sensitivity third for this reason.

9. What the .mid actually contains at every level

The cleanest proof that level does not change the notes is the file itself. We built the Standard MIDI File at each level and parsed the bytes back with a running-status parser:

Every file is 89 bytes, format 1, 2 tracks, 480 ticks per quarter note, one channel, with 4 note-on and 4 note-off events, zero program-change and zero pitch-bend. A byte-for-byte diff between the 0 dB and −70 dB exports finds exactly 8 differing bytes out of 89, at offsets 52, 57, 61, 66, 70, 75, 79, 84 — all eight are velocity bytes going from 113 to 1. Not one pitch, start time, end time or duration byte changes. The .csv tells the same story: with the velocity column stripped, the note list is identical at every level; the file size only shrinks from 197 to 193 to 189 bytes because “113” takes three characters and “1” takes one.

That 89-byte shape is the same file every host receives, and what a MIDI file actually is explains why it contains instructions rather than sound. For the byte-level anatomy, three routes to a MIDI file compared byte by byte shows a recording, a hand-drawn part and a score export all ending in this same shape, and how big these files can get explains why size tracks note count rather than audio level. If you want the seconds rather than ticks, the CSV note list is the closer record of what the analyser saw.

10. Export-level checklist for Cubase

1. Do not re-bounce to chase level. From 0 dB to −140 dB the note list was 60,62,64,67 in both modes. A normalised bounce is not more accurate.
2. Choose 24-bit or 32-bit float for a quiet part. 16-bit first broke at −58 dB; 24-bit held four of four notes at −80 dB. This is the only export-level decision that changes which notes come back.
3. Expect low velocity, and plan to fix it. Velocity falls about 11–12 steps per 6.02 dB and clamps at 1 below −65 dB. Scale velocities in Cubase after import, or leave them if the part is meant to be soft.
4. Keep the region long enough. Level does not delete notes, but a tightly trimmed region does — a region under 200 ms returns none. Give each note room.
5. Export mono for a single line. The analyser averages channels before it measures anything, so a hard-panned part costs 11 velocity steps; see the four-DAW export overview for the channel measurement.
6. Do not raise sensitivity to rescue a damaged bounce. At −70 dB and 16-bit, sensitivity 100 returned 10 polyphonic objects, six of them ghosts. Fix the bit depth, not the slider.
7. Check the CSV before the piano roll. With the velocity column removed, a quiet export and a loud one should diff clean. If the pitches differ, the cause is elsewhere: check format and mode first.

The BPM box is not a level fix either: it is written verbatim into the file and moves no notes — how accurate the written tempo is measures that exactly. If you are bouncing several parts, do them one file at a time; why there is no folder-drop batch button explains the workflow. On a phone the same page works with a file from the share sheet; mobile audio to MIDI covers the input differences.

Convert an exported file now — runs in your browser, nothing is uploaded.

Related guides

• Other DAWs audio to MIDIWhich export settings change the notes: length, level, channels, silence and fades
• Reaper audio to MIDIWhy a region under 200 ms returns no notes: the 92.88 ms window tax and the 155 ms knife edge
• Mono vs polyWhy the two detection modes fail differently on the same clip
• Fix dropped notesThe order to fix it: format, mode, sensitivity, shortest note
• Humanize robotic MIDIWhy the export is grid-locked and velocity-flat, and the moves that fix the feel
• MP3 to MIDI converterHow loud a note has to be before it hits velocity 127
• MIDI editorWhat the browser controls can and cannot change after conversion
• What is a MIDI fileInstructions not audio: format 0/1/2, tracks, channels and what each costs in bytes
• How to make a MIDI fileThree routes, one 2-track file, compared byte by byte
• MIDI file sizeWhy size tracks note count rather than audio length or level
• Audio to notesThe 7-column CSV and the exact seconds behind every onset
• M4A and FLAC to MIDIOne .m4a can be lossless ALAC or 32 kbps AAC
• WAV to MIDIHow to prepare the file before you convert
• Sample rate & MIDIWhy 44.1k, 48k and 22.05k return the same note list
• Clean up a converted MIDIThe passes that remove ghost notes and split objects
• Bass line extractionWhy the converter reads the whole signal and cannot isolate one voice
• Drum transcription limitsThe 55 Hz floor, the 4000 Hz ceiling and the 11.61 ms grid
• Pitch bend detectionWhat a glide adds to your MIDI, and when it matters
• Batch convert audioWhy there is no folder-drop button, and what one file at a time costs