Demucs.StemSeparation v1.0.0 Experimental
by Bogdan Teleaga
Stem separation model based on Demucs that can split an audio mix into bass, drums, vocals, other using the four knobs. Since we only output one stereo channel at a time, four different instances of the plugin need to be loaded to extract every stem at the same time.
The internal model is a retrained, smaller version of the hybrid demucs version hdemucs_mmi. The model is trained with a short segment size to exhibit better behavior with the short input buffers coming in from the DAW at inference time. Overlap add is used to merge these input buffers.