I just discovered this app a month ago, so forgive me if you already know about it! It’s pretty cool because you can use it to isolate just the guitar. I’ve been using it to dial in some presets and play over the track.
You can technically use YouTube! What I do is find a YouTube to mp3 converter then load it onto the app.Where do you get the tracks from if you can't use any streaming services or youtube?
Good lord, I’m really late to using this app. Haha!Very nice!. Yes, if you search for moises you'll find plenty of discussion of it on the forum.
haha. Yeah, the AI isn’t perfect. Still gets you a glimpse into the tones though!It doesn’t always work. Almost always includes organ and other guitars on the same stem.
Yes, it does. I just haven't been able to use it for tone-matching.haha. Yeah, the AI isn’t perfect. Still gets you a glimpse into the tones though!
yeah it leverages a model called "spleeter" I believeIt doesn’t always work. Almost always includes organ and other guitars on the same stem.
yeah it leverages a model called "spleeter" I believe
https://github.com/deezer/spleeter
that was trained primarily to split drums, bass, vocals and everything else. Guitar is "everything else". Someone should train a guitar oriented model to get minus-guitar tracks and it would work better. In order to do that you need the final mix and the stems to train the model to recognized the individual parts. Of all places where that dataset can be sourced this forum is probably the best bet.
I don't know how they do it but spleeter most definitely doesn't support extracting guitar alone. That hasn't changed. from https://github.com/deezer/spleeterFYI, "Guitar" is separate from "Other" in Moises. I've noticed Moises has improved significantly since the date of that blog post, but I don't know any details about how it has changed.
and a discussion on training a guitar specific modelSpleeter is Deezer source separation library with pretrained models written in Python and uses Tensorflow. It makes it easy to train source separation model (assuming you have a dataset of isolated sources), and provides already trained state of the art model for performing various flavour of separation :
I can imagine thats where it struggles for sure. I've only been importing some basic songs but would be curious to throw on some periphery or dream theater to testDifficulties lie with more-than-one-guitar bands.