General track repair
Richard Boisvert
Hi Udio. Often times I find myself with a completed song that I am in love with, but it has distortions at various points throughout the song. Things like strange background atmospherics, drums that sound compressed, chimes being too loud, etc. These issues are often much too large-scale for inpainting fixes, and remixing comes with the risk of changing parts of a song. My question is: what tools or techniques can I do to perform a "general repair" of my finished track? Or is there an upcoming tool that can help with this?
G
Gimmick7250
Hi!
I don’t quite understand the part where it says, "These issues are often much too large-scale for inpainting fixes." If you inpaint one of the issues and like the output, then you can continue inpainting with that output.
Let me explain my workflow for how I usually enhance tracks.
I’ll use a simple pop song as an example, one that has proper vocals and lyrics:
First, I try to inpaint the parts that don’t sound good (this usually works). This is a kind of preparation for the remix feature. I then remix the inpainted track at 0.1, 0.2, 0.3, or sometimes 0.25 variance, depending on what works. If I like one of the results, I can inpaint and remix it again, and so on.. If I’m happy with the output but still want to improve it, and continuous inpainting or remixing doesn’t help, I usually do one of two things:
Extending: I extend the track (using the same lyrics that are already in the song) with additional promts to the base promt, like "studio quality," "enhance quality," "high quality," or "please enhance the vocal," etc. It might not always be because of these prompts, but it has worked many times. Of course, if you get an output where the audio is improved, you’ll need to extend the entire track again. However, if you adjust the Context Length to around 120-130, it usually reproduces the track successfully.
Using DAW and Plugins: If you don’t want to publish the track on UDIO, I download the stems, load them into FL Studio (or any DAW), and use these iZotope plugins:
RX 11 (Intelligent Noise Reduction and Audio Repair Toolkit)
Neutron 5 (Mix Assistant)
Nectar 4 (Improved Vocal Assistant)
Ozone 11 (Master Assistant)
These plugins have AI assistants, and you only need to fine-tune the results. Once this is done, I export the file so that it’s no larger than 25 MB (that’s the limit UDIO allows for uploads). From there, it’s pretty much the same process as the first one, depending on what you want to do.
Oh yes, I forgot to mention: if you want to separate the stems even more effectively, I recommend the UVR-5 open-source software, which you can run directly on your computer. It includes a lot of models, and in my opinion, it offers the highest-quality stem separation available (it has options for noise, echo, crowd separation, etc.).
These methods require some experimentation because they’re not straightforward. I hope this helps a bit!