PDF conversion and Midi Export from MSCX files.
So recently, I've tried Musescore's import PDF feature. Though it wasn't perfect, I was really amazed by how well it does its job! I've even tested it with unconventional sheet music PDFs like duets and multiple instruments, and it was still able to return satisfying results.
Still, there were certain things that caught my attention during my experimentation with this technology. For some reason, the PDFs that were converted to MSCX files were always set at 120 tempo, I wonder why is that so?
Additionally, when exporting MSCX files to Midi, the Midi file (I'm using FL Studio 20 to open midi files) returns a lot of weird time signature information (7/4, 26/4, 5/4, etc), even though the time signature is consistently 4/4 in the MSCX file.
This is not a complain post for the software. I just want to know more about the certain features I picked up that made me question how does Musescore processes things to generate the things I'm observing.
Thoughts?
Comments
120 BPM is the default tempo, if there is no other set
Regarding the time signatures, probably the PDF import process was misinterpreting multiple voices as extra beats and was adjusting the length of the measures to accommodate this. You probably also see little "+" and "-" signs above some of the measures. So the import process didn't actually go as well as it may have appeared at first; those are all errors that will need correcting most likely.
In reply to Regarding the time… by Marc Sabatella
Actrually the OP states that ...the time signature is consistently 4/4 in the MSCX file.
[i.e., consistently 4/4 in the MuseScore file]
It is ...when exporting MSCX files to Midi, the Midi file (I'm using FL Studio 20 to open midi files) returns a lot of weird time signature information...
The PDF import process, it seems, was satisfactory to the OP. (Except for wondering why the tempo was always being set at 120.)
In reply to Actrually the OP states that… by Jm6stringer
Time signature itself may have been consistently 4/4, but the actual measure durations almost certainly weren't. They seldom are in PDF import because the error I described is extremely common, and it would totally explain the MIDI issue also.
My guess is that while the score looked OK at first glance, further examination - including attempts to playback - would reveal that it wasn't so satisfactory after all, this is just the first sign of that.
But we'd need to see the actual score in order to say more.
In reply to Time signature itself may… by Marc Sabatella
But we'd need to see the actual score in order to say more.
Agreed.
Years ago (before the PDF conversion service was available here) I used Audiveris with good - sometimes perfect - results. Admittedly, it was mostly for lead sheets.
The present conversion service offered by MuseScore is a 'one pass' operation, so results can vary widely depending on the source material.
P.S... By using this existing resource of available pdf score submissions - both successes and failures - to the MS conversion service, I hope that Herve at Audeveris is working on some sort of 'deep learning' AI to produce the ultimate OMR.
@dleeeenv
Music notation software like MuseScore strives to create readable scores to be playable by human musicians.
MIDI files are basically sets of instructions for machine (computer/synthesizer) performance consisting basically of note-on and note-off times, so a term like 'time signature' is not important to a machine. After all, a computer does not possess a foot to tap according to the 'meter'. A machine strives only to play each note at a prescribed time according to its internal clock. Additionally, it needs to know what pitch, what instrument sound, how loud, if any special effects, etc...
Here's part of an 'event list' from a midi file:
1,Min:Sec:Msec=0:00:000,TrackName"tk1"
2,Min:Sec:Msec=0:00:000,MidiPort=0
3,Min:Sec:Msec=0:00:000,ChannelPrefix 1
4,Min:Sec:Msec=0:00:000,ChannelVolume chan: 1 value: 120
5,Min:Sec:Msec=0:00:000,Panpot chan: 1 value: 70
6,Min:Sec:Msec=0:00:000,BankSel LSB chan: 1 value: 0
7,Min:Sec:Msec=0:00:000,EffectsDepth chan: 1 value: 127
8,Min:Sec:Msec=0:00:000,ProgramChange chan: 1, value: 87
9,Min:Sec:Msec=0:00:000,NoteOn chan: 1 note: 65 vol: 100 dur: 240
10,Min:Sec:Msec=0:00:260,NoteOff chan: 1 note: 65
11,Min:Sec:Msec=0:00:521,NoteOn chan: 1 note: 68 vol: 95 dur: 240
12,Min:Sec:Msec=0:00:782,NoteOff chan: 1 note: 68
13,Min:Sec:Msec=0:00:913,NoteOn chan: 1 note: 65 vol: 100 dur: 120
14,Min:Sec:Msec=0:01:043,NoteOff chan: 1 note: 65
15,Min:Sec:Msec=0:01:173,NoteOn chan: 1 note: 65 vol: 100 dur: 120
16,Min:Sec:Msec=0:01:304,NoteOff chan: 1 note: 65
17,Min:Sec:Msec=0:01:304,NoteOn chan: 1 note: 70 vol: 95 dur: 120
18,Min:Sec:Msec=0:01:434,NoteOff chan: 1 note: 70
19,Min:Sec:Msec=0:01:565,NoteOn chan: 1 note: 65 vol: 95 dur: 120
20,Min:Sec:Msec=0:01:695,NoteOff chan: 1 note: 65
21,Min:Sec:Msec=0:01:826,NoteOn chan: 1 note: 63 vol: 95 dur: 120
22,Min:Sec:Msec=0:01:956,NoteOff chan: 1 note: 63...etc.
See?
Note-on times, note-off times, note pitch, volume...
Software like FL Studio 20 makes an attempt, when opening midi files, to present a human readable score and so must occasionally make guesses at (re)creating information that is not present in midi but is needed for proper score notation.
On the other hand...
PDF conversion to music notation is similar to using OCR (optical character reader) for text images. The difference is that the OMR (optical music reader) needs to recognize not only alphanumeric characters, but also all the other symbols that comprise music notation - a more daunting task.
Whenever I try to import a PDF, it always adds extra beats to some bars. Not sure how or why it tries to recompose the music. I have to correct it, although better than starting from scratch. Also, it often adds some mysterious piano pedal indications above the stave? Basically, it doesn't work, which is a great pity. Luxuries aside, it's a great program.