Algorithm info: Algorithm name: UVR-MDX-NET Voc_FT - A5 Settings (L+R channel processed with UVR)
See: https://media.discordapp.net/attachments/708579735583588366/1127363323378290729/2023-07-08_23_19_04-Ultimate_Vocal_Remover.png
Advanced: https://media.discordapp.net/attachments/708579735583588366/1127362547763400805/2023-07-08_23_15_53-Ultimate_Vocal_Remover.png
1. split unprocessed stereo track to 2x mono (when I pan them in audacity left and right, compare to unprocessed by ear, is indeed identical) = OK:
for f in *; do ffmpeg -i "$f" -map_channel 0.0.0 "$f_left.wav" -map_channel 0.0.1 "$f_right.wav"; done
2. process those 2x mono with UVR and put them into /join (bulk rename in between, since UVR adds some stuff to the filenames)
3. re-convert them back to mono, which will be saved in /join_mono (since UVR only outputs in stereo):
for f in *; do ffmpeg -i "$f" -ac 1 "../join_mono/$f"; done
4. join those 2x mono to stereo (so copy files from /join_mono into /join), which will be saved in /output:
for f in original/*; do ffmpeg -i "join/$(basename -- $f)-left.wav" -i "join/$(basename -- $f)-right.wav" -filter_complex "[0:a][1:a]join=inputs=2:channel_layout=stereo[a]" -map "[a]" "output/$(basename -- $f)"; done
Date added: 2023-07-09 |
