Add Link to Github pages version in README (if exists ...)

[designzwang]

Hi @VehpuS ,

trying to find a way to automate Voice synthesizing from MusicXML for choir practice files, I tried similar steps as you did - synsi and oddvoices, and now I discovered your github repo - very promising and I would like to try it. However, there seems to be no online demo for a quick test, is there? Your readme states "Either use the Github pages / netlify hosted version ... " which could mean there is a github pages version. If so, it would be gread to put a link to it in the README. Or did you mean users should clone the repo and then create a github pages themselves in their own github account?
Thank you,
Uli

[VehpuS]

There is a Github pages hosted version here: https://vehpus.github.io/singing-synthesis/

I have already used it to synthesize a few guides for my arrangements and while I'd love to iterate more to simplify the process (applying automatic fixes to lyrics based on known quirks of the synthesis library, mainly) I can personally attest that it's viable, albeit with some iteration.

Let me know if I can help / if you have ideas for clarification and improvement

[VehpuS]

The link appears in Github itself but I'll definitely take your note and add it to the readme as well

[designzwang]

@VehpuS Thanks for your reply! So i must have missed the link, sorry.
Meanwhile I fiddled around with docker, and along the way , took notes how to get it up and running. Polished them up a little bit so others could follow along - feel free to use them for the docs or not ;-) However, after getting it up and accessing it via http://localhost:5173/, i could generate WAV files of the correct length , but they were empty. i.e. they were mute.
Using your github pages version, the files had sound, and they were about double the size ( 22 vs 11 MB for about 3:50 min)

So while your Wrapper and GUI to process and split musicXML works fine ( well done!) , the final result that oddvoices produces is worse than what I got out of the sinsy web version. So currently I would rather do the split the voices manually ( single xml export from musescore) and process one-by-one with sinsy.

docker-windows-11.md

[VehpuS]

That's why I keep both versions up - glad you were able to get sinsy to run, since I know it's not that easy! Sinsy does have nicer sounding voices but can't change tempo and was difficult to set up + requires a backend, and is not built to support being used as a virtual instrument which I thought could be interesting to attempt to setup later - so I personally focused on polishing that pipeline since my goal isn't prettiness (that's the singers' job ;) ) but to create easy to use guides and ideally a tool to test vocal musical notation with lyrics as easily and accessibly

[VehpuS]

I'd love to add your notes - want to create a Pr so you get the proper credit? ;)

[designzwang]

Actually I have only used the japanese web gui at https://www.sinsy.jp/ , it looks outdated but works at least for some mxml files, and some others that sinsy rejected could be processed with your github pages version.

Being the tech guy in our choir I first created a pipeline to automatically render weighted (piano) audio mixes for all voices from a single Midi. Now I was ( like you) looking for a way to additionally and automatically generate acceptable "sung" versions.

In the short time that I spent on this, I have seen many improvements in the paid tools arrangers use - i.e. over time it will be lesser work to correctly fix up the lyrics for synthesis. Actually I would expect useful export plugins for tools like Musescore to appear very soon. Also, it seems to become more common for arrangers to deliver "sung" versions where they can control the quality. So even though it is (currently) a lot of additional work, the last three arrangemens we bought from different arrangers already came with AI generated tracks, all of them created with Dreamtonics, and no additional cost ... why am I saying this - I wonder whether I better should just sit and wait a couple of months instead of fiddling around with my limited knowledge :-)

Regarding the docker info - I know myself: while sound synth is my favorite playgrund today, it might be something else tomorrow, so I won't be able to follow along and update them and a PR would make no sense. If you can use the notes (thats why I put them in markdown notation) feel free to place them wherever you see fit and do that without credit.
Regards Uli