I have been exploring a bit the promising AI music XML conversion provided by the Newzik app. As my test project I have Tierkreis by Stockhausen.
I scanned the first part with Newzik, converted to livescore, exported musicXML and imported to dorico. With only one staff everything worked quite nicely, only small corrections were needed here and there. With two staves I ran into problems with voices:
I am not quite sure what is happening, but trying to fix the voices with “change voice” wasn’t very easy. Does someone now how to easily correct the voices? Or could this be somehow avoided already in the musicXML import stage?
Tierkreis.dorico (369.2 KB)
If someone wants to collaborate with me on this project, there are in total 12 similar parts in this composition that I would like to import to Dorico. The goal is to produce transposing parts or parts with different clefs for a performance.
It would be interesting to get a look at the MusicXML file that you shared into Dorico from Newzik, if possible. Can you share the MusicXML file to your computer and then zip it up and attach it here?
Here is the MusicXML -file exported from newzik:
scan_Zesnfj3_all-movements.xml (41.4 KB)
Here is also the PDF for reference:
Tierkreis 1.pdf (610.5 KB)
Dorico has faithfully reproduced the voicing in the xml file.
Thanks for the MusicXML files. Yes, as Janus says, Dorico is rendering the MusicXML file exactly as it is encoded: normally voice 1 is on staff 1, but e.g. in bar 8, this gets swapped, so that voice 2 is now in staff 1.
You can fix this up after the fact in Dorico using the Change Voice commands, but Dorico is taking the MusicXML input reasonably literally. You could try sending this to the team at Newzik and asking them whether it might be something they would like to look at in their MusicXML export routines.
Swap Voice Contents is easier way.
Select all musical contents in bars 8-11 > context menu > Voices > Swap Voice Contents.
Thank you for the help!
I’ll try to contact newzik about the changing voices with two staves, but the swap voice contents -command seems to be the right tool for easily fixing this in Dorico.
I’ll also see if better scan quality improves the results, now the original scan is a bit ”wobbly”
I wanted to raise back this old topic about newzik, I have been using it quite a bit lately it has been a significant improvement to my workflow.
I don’t know if they have been improving the machine learning algorithm, but I can definetely recommend newzik to anyone who needs to bring notation from pdfs or printed scores to dorico. I have tried photoscore, but working with newzik is much nicer since the cleaning up part can be done inside dorico with the tools I am comfortable working with.
There are some common mistakes that the algorithm does:
-interpreting numbers as fingerings
-not understanding if note is tied over system break
But overall it is very reliable in getting the notes and accidentals right. I had to reconstruct a score from printed parts and it even recognised correctly the transpositions of clarinets and french horns and somehow encoded that into the music XML file.
Other job I did was importing a full string quartet score into dorico (the parts were missing and had to be remade) and I was very impressed of how little mistakes I had to correct during that process. It was mostly cleaning up dynamics and playing techniques - occasianally I had to rewrite some of the more complicated passages.
What I like about the algorithm is that it “understands” time signatures and “knows” that 4/4 bar should have a certain amount of notes in it.
All of this just to recommend trying newzik to all fellow doricians!
Does that extend to tuplets? And what about lyrics?
It does, as long at it doesn’t confuse them to be fingerings.
I have just deleted the lyrics and reimported them from a hyphenated text file