search

SUC-DriverOld / MSST-WebUI

Music Source Separation Training Inference Webui, besides, we packed UVR together!
GNU Affero General Public License v3.0
45 stars 9 forks source link

How about add Kim mel roformer? #19

Closed hundrix closed 3 days ago

hundrix commented 1 week ago

Since this is one of the best vocal models, how about add it to the model list?

https://github.com/KimberleyJensen/Mel-Band-Roformer-Vocal-Model

SUC-DriverOld commented 1 week ago

It has been added in version 1.5, however the name of the model in the program is called "model_vocals_mel_band_roformer_kj_sdr_10.98.ckpt"(Kim_MelBandRoformer). Therefore, some people may not be able to find this model. I will consider changing the model name in future version. Besides, if you use version 1.6 preview, you can manually add models in “WebUI - Download modes - Install unofficial MSST model”.

hundrix commented 4 days ago

Thanks for the great project. I've replaced UVR with MSST in my workflow as the first step for cleaning live stream recordings, specifically for RVC training. The following preset gets the job done effectively:

model_bs_roformer_ep_317_sdr_12.9755.ckpt model_mel_band_roformer_karaoke_aufr33_viperx_sdr_10.1956.ckpt deverb_bs_roformer_8_256dim_8depth.ckpt denoise_mel_band_roformer_aufr33_sdr_27.9959.ckpt

I'll tryout the "Install unofficial MSST model" method.