In an attempt to improve the quality of the output stems, I upsampled the input audio (48K to 96K) in the naive belief that this might make the separation better, but the final output audio quality was not very high.
The problem is definitely not with these parameter settings such as Shifts or Overlap, as I tweaked these parameters to EXTREMES (Shifts:19 / Overlap:0.983). I'm currently speculating that the reason for the less-than-ideal separation results is that the Demucs V4 model is trained on either 44.1kHz or 48kHz, and it only works best with audio at these two common sample rates?
❓ Questions
(PS. I'm using UVR 5.6.0)
Hi Decums Fam.
In an attempt to improve the quality of the output stems, I upsampled the input audio (48K to 96K) in the naive belief that this might make the separation better, but the final output audio quality was not very high.
The problem is definitely not with these parameter settings such as Shifts or Overlap, as I tweaked these parameters to EXTREMES (Shifts:19 / Overlap:0.983). I'm currently speculating that the reason for the less-than-ideal separation results is that the Demucs V4 model is trained on either 44.1kHz or 48kHz, and it only works best with audio at these two common sample rates?