There are some old theory textbooks that I’ve long been very interested in making translations of. Just for my own enjoyment and musical life. But my French is too basic.
They’re available from ISMLP as photocopied pdfs. The formatting is in that old French style, similar to what Messiaen used for his “technique…” book.
I would like to use AI to do the translation of the old-style French. That is no problem. The hundreds, if not thousands, of musical examples pose the challenge. Google AI says that it can spit the musical examples out as xml, and then I can get them into Dorico. I see that as giving headaches as well, when there’s 20 similar examples on one page, all dealing with the same subject, and the AI misinterprets something etc etc.
Anyway, has anyone done anything like this? Is there a workflow that minimizes the headache? Or do I have to spend countless hours in Dorico readjusting all the musical examples? Cheers. Merry Christmas and Happy Hanukkah.
Thanks so much for your reply. Yes, the xml output, is 80% accurate, with what it grabs, but it doesn’t grab everything from the page. Any recommendations?
Of course I’m not sure about it, that’s why I’m posting my situation on an Internet forum. I’m not sure what you’re trying to say/imply. Please explain. Translation is translation, no? Are you saying there will be some inaccuracies? Of course there will. The 10 pages of the text I’ve tested, are not perfect, but my rusty French, or that of my comrades, can remedy that without too many problems.
I can’t speak for Hugo, but I’m saying that AI speaks with authority and talks a good enough game to fool you into trusting it. Yet on a subject I actually know myself, I find its output often horrifying. It has no problem telling you contradictory impossibilities, and obviously never has any actual experience of its own to draw on - it seems to have no self check or filter to say, “wait that can’t be right.”
I am not anti-AI, but the reality is that it’s absolutely not dependable for this kind of thing. Perhaps it can be a starting point for translation. XML, absolutely not. I look forward to the day when it is.
Dan, I’m wondering what you mean by this? Do you mean “absolutely not dependable”, or “absolutely not possible”? Because, I just did a few pages. It’s not perfect by any means, and yes, it might be quicker to just do in Dorico myself. But it is clearly doing it. It still needs touching-up to be sure, but I can say for certain, that I am uploading a photocopied pdf, and Gemini Pro, is extracting musical examples, and giving me the xml code, which I’ve then put in Dorico and got semi-usable results. I guess my initial post, was wondering how to make it more dependable.
Even assuming that you’ve got perfect musical examples and an accurate translation, the layout work, even in a DTP app, is going to take some time and thought. And for something that is mainly text with in-line music examples, I would do it in a DTP app, not in Dorico directly.
It sounds like a great project, and I wish you the best of luck with it.
This was the essence of what I meant. You may get something, but at present, it’s some combination of gibberish and errors. In the end, you’ll spend far less time (and have far greater confidence) just doing it yourself. That may well change soon as AI continues to improve.
Thanks for the reply Ben, yes, for sure, if I find an actual non-terrifying workflow to accomplish this, I’ll do it in a DTP. Definitely. Thanks for the encourgagment. Yes, the books are the educational materials by Charles Koechlin, who I’ve always loved and found his compositonal explorations mirrored things I was interested in too. It was only a couple of years ago that I realized he had written all these educational texts.
It is certainly not an ‘easy’ book to convert. The ‘scattered’ layout makes it ‘challenging’ for most software, but this might be a ‘layout of the time’, reducing the page count and making it economical to print. With today’s digital books, the page count is less important. 200 or 800 pages cost the same, i.e. nothing.
I cannot speak to book design in Dorico, but I would personally avoid it for this book. @dan_kreider knows best what can be done in InDesign. The ‘inline’ images might be a ‘challenge’.
I tried a small portion in LaTeX (skipping the engraving in Dorico, and some ‘slashed numbers’, using scanned images only);
Replacing the scanned images with new Dorico engraved samples would be easy.**
The line distances are made automatically (not sure how that works in InDesign). One could argue that the 2nd last and last line should be more vertically compressed, like
InDesign is built for this sort of thing. You can set fixed line leading (space between lines). And images can be placed inline, with many options for anchoring and padding.
The key to a fast workflow is taking time to set up paragraph styles, character styles, object styles, and parent pages. There is a lot of automation that can be done to make things go quickly. Object styles can be applied using key commands. Also, GREP is your friend.
When it comes to managing slices, I typically do one project per chapter (depending on how big the chapters are) with multiple flows, and name each slice carefully. Make your Dorico pages thin and long, so there’s no need to trim slices above and below. Just export one flow at a time as a PDF.
This I find interesting. When you have time some day, could you please make a topic with detailed instructions? It would be interesting to read and learn.
