mirror of
https://github.com/Anjok07/ultimatevocalremovergui.git
synced 2024-11-24 15:30:11 +01:00
1139 lines
32 KiB
Batchfile
1139 lines
32 KiB
Batchfile
@echo off
|
|
|
|
cls
|
|
:start
|
|
set choice=
|
|
set /p choice=Run with TTA? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto tta
|
|
if '%choice%'=='y' goto tta
|
|
if '%choice%'=='N' goto notta
|
|
if '%choice%'=='n' goto notta
|
|
if '%choice%'=='' goto notta
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:tta
|
|
ECHO =======================================
|
|
ECHO Choose Your Model
|
|
ECHO =======================================
|
|
ECHO.
|
|
ECHO a. MGM-v5-2Band-32000-BETA1
|
|
ECHO b. MGM-v5-2Band-32000-BETA2
|
|
ECHO c. MGM-v5-3Band-44100-BETA
|
|
ECHO d. MGM-v5-4Band-44100-BETA1
|
|
ECHO e. MGM-v5-4Band-44100-BETA2
|
|
ECHO f. MGM-v5-KAROKEE-32000-BETA1
|
|
ECHO g. MGM-v5-KAROKEE-32000-BETA2-AGR
|
|
ECHO h. MGM-v5-MIDSIDE-44100-BETA1
|
|
ECHO i. MGM-v5-MIDSIDE-44100-BETA2
|
|
ECHO j. MGM-v5-Vocal_2Band-32000-BETA1
|
|
ECHO k. MGM-v5-Vocal_2Band-32000-BETA2
|
|
ECHO l. LOFI_2band_iter5_2
|
|
ECHO m. LOFI_2band_iter5_2
|
|
ECHO n. HighPrecison_4band_1
|
|
ECHO o. HighPrecison_4band_2
|
|
ECHO p. NewLayer_4band_1.pth
|
|
ECHO q. NewLayer_4band_2.pth
|
|
ECHO r. NewLayer_4band_3.pth
|
|
ECHO s. Ensemble 4Band Models (7 Models) - Final Outputs Save to Ensembled Folder!
|
|
ECHO t. Ensemble All Models (12 Models) - Final Outputs Save to Ensembled Folder!
|
|
set choice=
|
|
set /p choice=Type the letter associated with the option you choose and hit 'Enter':
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='a' goto model1
|
|
if '%choice%'=='b' goto model2
|
|
if '%choice%'=='c' goto model3
|
|
if '%choice%'=='d' goto model4
|
|
if '%choice%'=='e' goto model5
|
|
if '%choice%'=='f' goto model6
|
|
if '%choice%'=='g' goto model7
|
|
if '%choice%'=='h' goto model8
|
|
if '%choice%'=='i' goto model9
|
|
if '%choice%'=='j' goto model10
|
|
if '%choice%'=='k' goto model11
|
|
if '%choice%'=='l' goto model12
|
|
if '%choice%'=='m' goto model13
|
|
if '%choice%'=='n' goto model14
|
|
if '%choice%'=='o' goto model15
|
|
if '%choice%'=='p' goto model16
|
|
if '%choice%'=='q' goto model17
|
|
if '%choice%'=='r' goto model18
|
|
if '%choice%'=='s' goto model19
|
|
if '%choice%'=='t' goto model20
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:model1
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
ECHO Running MGM-v5-2Band-32000-BETA1
|
|
set model=MGM-v5-2Band-32000-BETA1
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\2band_32000.json -D -w 352 -P models\%model%.pth -t -i %1
|
|
goto end
|
|
:default
|
|
ECHO Running MGM-v5-2Band-32000-BETA1
|
|
set model=MGM-v5-2Band-32000-BETA1
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\2band_32000.json -w 352 -P models\%model%.pth -t -i %1
|
|
goto end
|
|
:model2
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
|
|
ECHO Running Model MGM-v5-2Band-32000-BETA2
|
|
set model=Running Model MGM-v5-2Band-32000-BETA2
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\2band_32000.json -w 352 -D -P models\%model%.pth -t -i %1
|
|
goto end
|
|
:default
|
|
|
|
ECHO Running Model MGM-v5-2Band-32000-BETA2
|
|
set model=Running Model MGM-v5-2Band-32000-BETA2
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\2band_32000.json -w 352 -P models\%model%.pth -t -i %1
|
|
goto end
|
|
:model3
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
|
|
ECHO Running Model MGM-v5-3Band-44100-BETA
|
|
set model=MGM-v5-3Band-44100-BETA
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\3band_44100.json -w 352 -D -P models\%model%.pth -t -i %1
|
|
goto end
|
|
:default
|
|
|
|
ECHO Running Model MGM-v5-3Band-44100-BETA
|
|
set model=MGM-v5-3Band-44100-BETA
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\3band_44100.json -w 352 -P models\%model%.pth -t -i %1
|
|
goto end
|
|
:model4
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
|
|
ECHO Running Model MGM-v5-4Band-44100-BETA1
|
|
set model=MGM-v5-4Band-44100-BETA1
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\4band_44100.json -w 352 -D -P models\%model%.pth -t -i %1
|
|
goto end
|
|
:default
|
|
|
|
ECHO Running Model MGM-v5-4Band-44100-BETA1
|
|
set model=MGM-v5-4Band-44100-BETA1
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\4band_44100.json -w 352 -P models\%model%.pth -t -i %1
|
|
goto end
|
|
:model5
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
|
|
ECHO Running Model MGM-v5-4Band-44100-BETA2
|
|
set model=MGM-v5-4Band-44100-BETA2
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\4band_44100.json -D -w 352 -P models\%model%.pth -t -i %1
|
|
goto end
|
|
:default
|
|
|
|
ECHO Running Model MGM-v5-4Band-44100-BETA2
|
|
set model=MGM-v5-4Band-44100-BETA2
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\4band_44100.json -w 352 -P models\%model%.pth -t -i %1
|
|
goto end
|
|
:model6
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
|
|
ECHO Running Model MGM-v5-KAROKEE-32000-BETA1
|
|
set model=MGM-v5-KAROKEE-32000-BETA1
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\2band_32000.json -D -w 352 -P models\%model%.pth -t -i %1
|
|
goto end
|
|
:default
|
|
|
|
ECHO Running Model MGM-v5-KAROKEE-32000-BETA1
|
|
set model=MGM-v5-KAROKEE-32000-BETA1
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\2band_32000.json -w 352 -P models\%model%.pth -t -i %1
|
|
goto end
|
|
:model7
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
|
|
ECHO Running Model MGM-v5-KAROKEE-32000-BETA2-AGR
|
|
set model=MGM-v5-KAROKEE-32000-BETA2-AGR
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\2band_32000.json -D -w 352 -P models\%model%.pth -t -i %1
|
|
goto end
|
|
:default
|
|
|
|
ECHO Running Model MGM-v5-KAROKEE-32000-BETA2-AGR
|
|
set model=MGM-v5-KAROKEE-32000-BETA2-AGR
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\2band_32000.json -w 352 -P models\%model%.pth -t -i %1
|
|
goto end
|
|
:model8
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
|
|
ECHO Running Model MGM-v5-MIDSIDE-44100-BETA1
|
|
set model=MGM-v5-MIDSIDE-44100-BETA1
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\3band_44100_mid.json -D -w 352 -P models\%model%.pth -t -i %1
|
|
goto end
|
|
:default
|
|
|
|
ECHO Running Model MGM-v5-MIDSIDE-44100-BETA1
|
|
set model=MGM-v5-MIDSIDE-44100-BETA1
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\3band_44100_mid.json -w 352 -P models\%model%.pth -t -i %1
|
|
goto end
|
|
:model9
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
|
|
ECHO Running Model MGM-v5-MIDSIDE-44100-BETA2
|
|
set model=MGM-v5-MIDSIDE-44100-BETA2
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\3band_44100_mid.json -D -w 352 -P models\%model%.pth -t -i %1
|
|
goto end
|
|
:default
|
|
|
|
ECHO Running Model MGM-v5-MIDSIDE-44100-BETA2
|
|
set model=MGM-v5-MIDSIDE-44100-BETA2
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\3band_44100_mid.json -w 352 -P models\%model%.pth -t -i %1
|
|
goto end
|
|
:model10
|
|
ECHO Running Model MGM-v5-Vocal_2Band-32000-BETA1
|
|
set model=MGM-v5-Vocal_2Band-32000-BETA1
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\2band_32000.json -w 352 -P models\%model%.pth -t -i %1
|
|
goto end
|
|
:model11
|
|
ECHO Running Model MGM-v5-Vocal_2Band-32000-BETA2
|
|
set model=MGM-v5-Vocal_2Band-32000-BETA2
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\2band_32000.json -w 352 -P models\%model%.pth -t -i %1
|
|
goto end
|
|
:model12
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
|
|
ECHO Running Model LOFI_2band-1_33966KB
|
|
set model=LOFI_2band-32000-1_33966KB
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\2band_44100_lofi.json -D -w 352 -n 33966KB -P models\%model%.pth -t -i %1
|
|
goto end
|
|
:default
|
|
|
|
ECHO Running Model LOFI_2band-1_33966KB
|
|
set model=LOFI_2band-32000-1_33966KB
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\2band_44100_lofi.json -w 352 -n 33966KB -P models\%model%.pth -t -i %1
|
|
goto end
|
|
:model13
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
|
|
ECHO Running Model LOFI_2band-2_33966KB
|
|
set model=LOFI_2band-2_33966KB
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\2band_44100_lofi.json -D -w 352 -n 33966KB -P models\%model%.pth -t -i %1
|
|
goto end
|
|
:default
|
|
|
|
ECHO Running Model LOFI_2band-2_33966KB
|
|
set model=LOFI_2band-2_33966KB
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\2band_44100_lofi.json -w 352 -n 33966KB -P models\%model%.pth -t -i %1
|
|
goto end
|
|
:model14
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
|
|
ECHO Running Model HighPrecison_4band_1
|
|
set model=HighPrecison_4band_1
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\4band_44100.json -D -w 352 -n 123821KB -P models\%model%.pth -t -i %1
|
|
goto end
|
|
:default
|
|
|
|
ECHO Running Model HighPrecison_4band_1
|
|
set model=HighPrecison_4band_1
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\4band_44100.json -w 352 -n 123821KB -P models\%model%.pth -t -i %1
|
|
goto end
|
|
:model15
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
|
|
ECHO Running Model HighPrecison_4band_2
|
|
set model=HighPrecison_4band_2
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\4band_44100.json -D -w 352 -n 123821KB -P models\%model%.pth -t -i %1
|
|
goto end
|
|
:default
|
|
|
|
ECHO Running Model HighPrecison_4band_2
|
|
set model=HighPrecison_4band_2
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\4band_44100.json -w 352 -n 123821KB -P models\%model%.pth -t -i %1
|
|
goto end
|
|
:model16
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
ECHO Running Model NewLayer_4band_1
|
|
set model=NewLayer_4band_1
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\4band_44100.json -w 352 -D -n 129605KB -P models\%model%.pth -t -i %1
|
|
goto start
|
|
:default
|
|
ECHO Running Model NewLayer_4band_1
|
|
set model=NewLayer_4band_1
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\4band_44100.json -w 352 -n 129605KB -P models\%model%.pth -t -i %1
|
|
goto end
|
|
:model17
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
|
|
ECHO Running Model NewLayer_4band_2
|
|
set model=NewLayer_4band_2
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\4band_44100.json -w 352 -D -n 129605KB -P models\%model%.pth -t -i %1
|
|
goto end
|
|
:default
|
|
|
|
ECHO Running Model NewLayer_4band_2
|
|
set model=NewLayer_4band_2
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\4band_44100.json -w 352 -n 129605KB -P models\%model%.pth -t -i %1
|
|
goto end
|
|
:model18
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
|
|
ECHO Running Model
|
|
set model=NewLayer_4band_3
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\4band_44100.json -w 352 -D -n 129605KB -P models\%model%.pth -t -i %1
|
|
goto end
|
|
:default
|
|
|
|
ECHO Running Model
|
|
set model=NewLayer_4band_3
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\4band_44100.json -w 352 -n 129605KB -P models\%model%.pth -t -i %1
|
|
goto end
|
|
:model19
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
|
|
ECHO Ensemble All 4Band Models
|
|
cd /d %~dp0
|
|
|
|
python 4Band_ens_inference.py -g 0 -D -w 352 -D -s -t -i %1
|
|
goto end
|
|
:default
|
|
|
|
ECHO Ensemble All 4Band Models
|
|
cd /d %~dp0
|
|
|
|
python 4Band_ens_inference.py -g 0 -w 352 -D -s -t -i %1
|
|
goto end
|
|
:model20
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
|
|
ECHO Ensemble All 44100-Models
|
|
cd /d %~dp0
|
|
|
|
python allmodels_ens_inference.py -g 0 -w 352 -D -t -i %1
|
|
goto end
|
|
:default
|
|
|
|
ECHO Ensemble All 44100-Models
|
|
cd /d %~dp0
|
|
|
|
python allmodels_ens_inference.py -g 0 -w 352 -t -i %1
|
|
pause
|
|
:notta
|
|
ECHO =======================================
|
|
ECHO Choose Your Model
|
|
ECHO =======================================
|
|
ECHO.
|
|
ECHO a. MGM-v5-2Band-32000-BETA1
|
|
ECHO b. MGM-v5-2Band-32000-BETA2
|
|
ECHO c. MGM-v5-3Band-44100-BETA
|
|
ECHO d. MGM-v5-4Band-44100-BETA1
|
|
ECHO e. MGM-v5-4Band-44100-BETA2
|
|
ECHO f. MGM-v5-KAROKEE-32000-BETA1
|
|
ECHO g. MGM-v5-KAROKEE-32000-BETA2-AGR
|
|
ECHO h. MGM-v5-MIDSIDE-44100-BETA1
|
|
ECHO i. MGM-v5-MIDSIDE-44100-BETA2
|
|
ECHO j. MGM-v5-Vocal_2Band-32000-BETA1
|
|
ECHO k. MGM-v5-Vocal_2Band-32000-BETA2
|
|
ECHO l. LOFI_2band_iter5_2
|
|
ECHO m. LOFI_2band_iter5_2
|
|
ECHO n. HighPrecison_4band_1
|
|
ECHO o. HighPrecison_4band_2
|
|
ECHO p. NewLayer_4band_1
|
|
ECHO q. NewLayer_4band_2
|
|
ECHO r. NewLayer_4band_3
|
|
ECHO s. Ensemble 4Band Models (7 Models) - Final Outputs Save to Ensembled Folder!
|
|
ECHO t. Ensemble All Models (12 Models) - Final Outputs Save to Ensembled Folder!
|
|
set choice=
|
|
set /p choice=Type the letter associated with the option you choose and hit 'Enter':
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='a' goto model1
|
|
if '%choice%'=='b' goto model2
|
|
if '%choice%'=='c' goto model3
|
|
if '%choice%'=='d' goto model4
|
|
if '%choice%'=='e' goto model5
|
|
if '%choice%'=='f' goto model6
|
|
if '%choice%'=='g' goto model7
|
|
if '%choice%'=='h' goto model8
|
|
if '%choice%'=='i' goto model9
|
|
if '%choice%'=='j' goto model10
|
|
if '%choice%'=='k' goto model11
|
|
if '%choice%'=='l' goto model12
|
|
if '%choice%'=='m' goto model13
|
|
if '%choice%'=='n' goto model14
|
|
if '%choice%'=='o' goto model15
|
|
if '%choice%'=='p' goto model16
|
|
if '%choice%'=='q' goto model17
|
|
if '%choice%'=='r' goto model18
|
|
if '%choice%'=='s' goto model19
|
|
if '%choice%'=='t' goto model20
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:model1
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
ECHO Running MGM-v5-2Band-32000-BETA1
|
|
set model=MGM-v5-2Band-32000-BETA1
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\2band_32000.json -D -w 352 -P models\%model%.pth -i %1
|
|
goto end
|
|
:default
|
|
ECHO Running MGM-v5-2Band-32000-BETA1
|
|
set model=MGM-v5-2Band-32000-BETA1
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\2band_32000.json -w 352 -P models\%model%.pth -i %1
|
|
goto end
|
|
:model2
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
|
|
ECHO Running Model MGM-v5-2Band-32000-BETA2
|
|
set model=Running Model MGM-v5-2Band-32000-BETA2
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\2band_32000.json -w 352 -D -P models\%model%.pth -i %1
|
|
goto end
|
|
:default
|
|
|
|
ECHO Running Model MGM-v5-2Band-32000-BETA2
|
|
set model=Running Model MGM-v5-2Band-32000-BETA2
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\2band_32000.json -w 352 -P models\%model%.pth -i %1
|
|
goto end
|
|
:model3
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
|
|
ECHO Running Model MGM-v5-3Band-44100-BETA
|
|
set model=MGM-v5-3Band-44100-BETA
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\3band_44100.json -w 352 -D -P models\%model%.pth -i %1
|
|
goto end
|
|
:default
|
|
|
|
ECHO Running Model MGM-v5-3Band-44100-BETA
|
|
set model=MGM-v5-3Band-44100-BETA
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\3band_44100.json -w 352 -P models\%model%.pth -i %1
|
|
goto end
|
|
:model4
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
|
|
ECHO Running Model MGM-v5-4Band-44100-BETA1
|
|
set model=MGM-v5-4Band-44100-BETA1
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\4band_44100.json -w 352 -D -P models\%model%.pth -i %1
|
|
goto end
|
|
:default
|
|
|
|
ECHO Running Model MGM-v5-4Band-44100-BETA1
|
|
set model=MGM-v5-4Band-44100-BETA1
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\4band_44100.json -w 352 -P models\%model%.pth -i %1
|
|
goto end
|
|
:model5
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
|
|
ECHO Running Model MGM-v5-4Band-44100-BETA2
|
|
set model=MGM-v5-4Band-44100-BETA2
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\4band_44100.json -D -w 352 -P models\%model%.pth -i %1
|
|
goto end
|
|
:default
|
|
|
|
ECHO Running Model MGM-v5-4Band-44100-BETA2
|
|
set model=MGM-v5-4Band-44100-BETA2
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\4band_44100.json -w 352 -P models\%model%.pth -i %1
|
|
goto end
|
|
:model6
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
|
|
ECHO Running Model MGM-v5-KAROKEE-32000-BETA1
|
|
set model=MGM-v5-KAROKEE-32000-BETA1
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\2band_32000.json -D -w 352 -P models\%model%.pth -i %1
|
|
goto end
|
|
:default
|
|
|
|
ECHO Running Model MGM-v5-KAROKEE-32000-BETA1
|
|
set model=MGM-v5-KAROKEE-32000-BETA1
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\2band_32000.json -w 352 -P models\%model%.pth -i %1
|
|
goto end
|
|
:model7
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
|
|
ECHO Running Model MGM-v5-KAROKEE-32000-BETA2-AGR
|
|
set model=MGM-v5-KAROKEE-32000-BETA2-AGR
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\2band_32000.json -D -w 352 -P models\%model%.pth -i %1
|
|
goto end
|
|
:default
|
|
|
|
ECHO Running Model MGM-v5-KAROKEE-32000-BETA2-AGR
|
|
set model=MGM-v5-KAROKEE-32000-BETA2-AGR
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\2band_32000.json -w 352 -P models\%model%.pth -i %1
|
|
goto end
|
|
:model8
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
|
|
ECHO Running Model MGM-v5-MIDSIDE-44100-BETA1
|
|
set model=MGM-v5-MIDSIDE-44100-BETA1
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\3band_44100_mid.json -D -w 352 -P models\%model%.pth -i %1
|
|
goto end
|
|
:default
|
|
|
|
ECHO Running Model MGM-v5-MIDSIDE-44100-BETA1
|
|
set model=MGM-v5-MIDSIDE-44100-BETA1
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\3band_44100_mid.json -w 352 -P models\%model%.pth -i %1
|
|
goto end
|
|
:model9
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
|
|
ECHO Running Model MGM-v5-MIDSIDE-44100-BETA2
|
|
set model=MGM-v5-MIDSIDE-44100-BETA2
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\3band_44100_mid.json -D -w 352 -P models\%model%.pth -i %1
|
|
goto end
|
|
:default
|
|
|
|
ECHO Running Model MGM-v5-MIDSIDE-44100-BETA2
|
|
set model=MGM-v5-MIDSIDE-44100-BETA2
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\3band_44100_mid.json -w 352 -P models\%model%.pth -i %1
|
|
goto end
|
|
:model10
|
|
ECHO Running Model MGM-v5-Vocal_2Band-32000-BETA1
|
|
set model=MGM-v5-Vocal_2Band-32000-BETA1
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\2band_32000.json -w 352 -P models\%model%.pth -i %1
|
|
goto end
|
|
:model11
|
|
ECHO Running Model MGM-v5-Vocal_2Band-32000-BETA2
|
|
set model=MGM-v5-Vocal_2Band-32000-BETA2
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\2band_32000.json -w 352 -P models\%model%.pth -i %1
|
|
goto end
|
|
:model12
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
|
|
ECHO Running Model LOFI_2band-1_33966KB
|
|
set model=LOFI_2band-32000-1_33966KB
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\2band_44100_lofi.json -D -w 352 -n 33966KB -P models\%model%.pth -i %1
|
|
goto end
|
|
:default
|
|
|
|
ECHO Running Model LOFI_2band-1_33966KB
|
|
set model=LOFI_2band-32000-1_33966KB
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\2band_44100_lofi.json -w 352 -n 33966KB -P models\%model%.pth -i %1
|
|
goto end
|
|
:model13
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
|
|
ECHO Running Model LOFI_2band-2_33966KB
|
|
set model=LOFI_2band-2_33966KB
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\2band_44100_lofi.json -D -w 352 -n 33966KB -P models\%model%.pth -i %1
|
|
goto end
|
|
:default
|
|
|
|
ECHO Running Model LOFI_2band-2_33966KB
|
|
set model=LOFI_2band-2_33966KB
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\2band_44100_lofi.json -w 352 -n 33966KB -P models\%model%.pth -i %1
|
|
goto end
|
|
:model14
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
|
|
ECHO Running Model HighPrecison_4band_1
|
|
set model=HighPrecison_4band_1
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\4band_44100.json -D -w 352 -n 123821KB -P models\%model%.pth -i %1
|
|
goto end
|
|
:default
|
|
|
|
ECHO Running Model HighPrecison_4band_1
|
|
set model=HighPrecison_4band_1
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\4band_44100.json -w 352 -n 123821KB -P models\%model%.pth -i %1
|
|
goto end
|
|
:model15
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
|
|
ECHO Running Model HighPrecison_4band_2
|
|
set model=HighPrecison_4band_2
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\4band_44100.json -D -w 352 -n 123821KB -P models\%model%.pth -i %1
|
|
goto end
|
|
:default
|
|
|
|
ECHO Running Model HighPrecison_4band_2
|
|
set model=HighPrecison_4band_2
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\4band_44100.json -w 352 -n 123821KB -P models\%model%.pth -i %1
|
|
goto end
|
|
:model16
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
ECHO Running Model NewLayer_4band_1
|
|
set model=NewLayer_4band_1
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\4band_44100.json -w 352 -D -n 129605KB -P models\%model%.pth -i %1
|
|
goto start
|
|
:default
|
|
ECHO Running Model NewLayer_4band_1
|
|
set model=NewLayer_4band_1
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\4band_44100.json -w 352 -n 129605KB -P models\%model%.pth -i %1
|
|
goto end
|
|
:model17
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
|
|
ECHO Running Model NewLayer_4band_2
|
|
set model=NewLayer_4band_2
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\4band_44100.json -w 352 -D -n 129605KB -P models\%model%.pth -i %1
|
|
goto end
|
|
:default
|
|
|
|
ECHO Running Model NewLayer_4band_2
|
|
set model=NewLayer_4band_2
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\4band_44100.json -w 352 -n 129605KB -P models\%model%.pth -i %1
|
|
goto end
|
|
:model18
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
|
|
ECHO Running Model
|
|
set model=NewLayer_4band_3
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\4band_44100.json -w 352 -D -n 129605KB -P models\%model%.pth -i %1
|
|
goto end
|
|
:default
|
|
|
|
ECHO Running Model
|
|
set model=NewLayer_4band_3
|
|
cd /d %~dp0
|
|
|
|
python inference.py -g 0 -m modelparams\4band_44100.json -w 352 -n 129605KB -P models\%model%.pth -i %1
|
|
goto end
|
|
:model19
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
|
|
ECHO Ensemble All 4Band Models
|
|
cd /d %~dp0
|
|
|
|
python 4Band_ens_inference.py -g 0 -D -w 352 -D -i %1
|
|
goto end
|
|
:default
|
|
|
|
ECHO Ensemble All 4Band Models
|
|
cd /d %~dp0
|
|
|
|
python 4Band_ens_inference.py -g 0 -w 352 -D -i %1
|
|
goto end
|
|
:model20
|
|
set choice=
|
|
set /p choice=Include Deep Vocal Extraction Instrumental? [Y/N]:
|
|
if not '%choice%'=='' set choice=%choice:~0,1%
|
|
if '%choice%'=='Y' goto deep
|
|
if '%choice%'=='y' goto deep
|
|
if '%choice%'=='N' goto default
|
|
if '%choice%'=='n' goto default
|
|
if '%choice%'=='' goto default
|
|
ECHO "%choice%" is not valid, try again
|
|
ECHO.
|
|
goto start
|
|
:deep
|
|
|
|
ECHO Ensemble All 44100-Models
|
|
cd /d %~dp0
|
|
|
|
python allmodels_ens_inference.py -g 0 -w 352 -D -i %1
|
|
goto end
|
|
:default
|
|
|
|
ECHO Ensemble All 44100-Models
|
|
cd /d %~dp0
|
|
|
|
python allmodels_ens_inference.py -g 0 -w 352 -i %1
|
|
goto end
|
|
:end
|
|
pause |