Jean-Marc Liotier wrote:richie wrote:I encourage you to check out Ultimate Vocal Remover man
How does it compare to the other demucs front-ends ? Is there any difference ?
If you're asking how it compares in relation to user interface, it's simple - load wav file, select the model config settings you want and output. If you're asking about quality of audio output, it varies based on the models you select.
Some models have strengths and weaknesses, so a lot of my experimentation as of recent is in running multiple models, you can select from several different demucs, mdx net, VR models and how you want them used accordingly. Ensemble mode is the most interesting as you can choose multiple models to process at the same time, so say you like mdx net for drums, a specific demucs for bass, VR for vocals, etc, you can do that. You just need to wait longer as it processes (unless you happen to have a decent GPU)
There is also a recent option where you have the option to test out Multi Stem, which provides more than just the typical 4 stems, Vocal, Drums, Bass, Other, there's the htdemucs_6d model which makes a decent attempt at adding piano and guitar too. They don't sound great but it's still something that'll only improve with time.
For those with computers that have recent Nvidia GPU, you can have UVR process the stemming on the GPU which is WAY faster than what even higher end higher core cpus can compute.