heimoshuiyu/ultimatevocalremovergui
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Add files via upload

Browse Source
This commit is contained in:
Anjok07
2022-04-14 23:41:57 -05:00
committed by GitHub
parent 18fbabc673
commit 4d9122581e
3 changed files with 1017 additions and 315 deletions
Download Patch File Download Diff File Expand all files Collapse all files

228
VocalRemover.py
View File

@@ -22,14 +22,15 @@ from collections import defaultdict
import queue import queue
import threading # Run the algorithm inside a thread import threading # Run the algorithm inside a thread
from pathlib import Path from pathlib import Path
import inference_v5 import inference_v5
import win32gui, win32con import inference_v5_ensemble
# import win32gui, win32con
the_program_to_hide = win32gui.GetForegroundWindow() # the_program_to_hide = win32gui.GetForegroundWindow()
win32gui.ShowWindow(the_program_to_hide , win32con.SW_HIDE) # win32gui.ShowWindow(the_program_to_hide , win32con.SW_HIDE)
# Change the current working directory to the directory # Change the current working directory to the directory
# this file sits in # this file sits in
@@ -44,19 +45,21 @@ os.chdir(base_path) # Change the current working directory to the base path
instrumentalModels_dir = os.path.join(base_path, 'models') instrumentalModels_dir = os.path.join(base_path, 'models')
banner_path = os.path.join(base_path, 'img', 'UVR-banner.png') banner_path = os.path.join(base_path, 'img', 'UVR-banner.png')
refresh_path = os.path.join(base_path, 'img', 'refresh.png') efile_path = os.path.join(base_path, 'img', 'file.png')
DEFAULT_DATA = { DEFAULT_DATA = {
'exportPath': '', 'exportPath': '',
'inputPaths': [], 'inputPaths': [],
'gpu': False, 'gpu': False,
'postprocess': False, 'postprocess': False,
'tta': False, 'tta': False,
'save': True,
'output_image': False, 'output_image': False,
'window_size': '512', 'window_size': '512',
'agg': 10, 'agg': 10,
'modelFolder': False, 'modelFolder': False,
'modelInstrumentalLabel': '', 'modelInstrumentalLabel': '',
#'aiModel': 'v5', 'aiModel': 'Single Model',
'ensChoose': 'HP1 Models',
'useModel': 'instrumental', 'useModel': 'instrumental',
'lastDir': None, 'lastDir': None,
} }
@@ -196,7 +199,7 @@ class MainWindow(TkinterDnD.Tk):
PADDING = 10 PADDING = 10
COL1_ROWS = 6 COL1_ROWS = 6
COL2_ROWS = 5 COL2_ROWS = 6
COL3_ROWS = 6 COL3_ROWS = 6
def __init__(self): def __init__(self):
@@ -223,7 +226,7 @@ class MainWindow(TkinterDnD.Tk):
# --Variables-- # --Variables--
self.logo_img = open_image(path=banner_path, self.logo_img = open_image(path=banner_path,
size=(self.winfo_width(), 9999)) size=(self.winfo_width(), 9999))
self.refresh_img = open_image(path=refresh_path, self.efile_img = open_image(path=efile_path,
size=(20, 20)) size=(20, 20))
self.instrumentalLabel_to_path = defaultdict(lambda: '') self.instrumentalLabel_to_path = defaultdict(lambda: '')
self.lastInstrumentalModels = [] self.lastInstrumentalModels = []
@@ -236,6 +239,7 @@ class MainWindow(TkinterDnD.Tk):
self.gpuConversion_var = tk.BooleanVar(value=data['gpu']) self.gpuConversion_var = tk.BooleanVar(value=data['gpu'])
self.postprocessing_var = tk.BooleanVar(value=data['postprocess']) self.postprocessing_var = tk.BooleanVar(value=data['postprocess'])
self.tta_var = tk.BooleanVar(value=data['tta']) self.tta_var = tk.BooleanVar(value=data['tta'])
self.save_var = tk.BooleanVar(value=data['save'])
self.outputImage_var = tk.BooleanVar(value=data['output_image']) self.outputImage_var = tk.BooleanVar(value=data['output_image'])
# Models # Models
self.instrumentalModel_var = tk.StringVar(value=data['modelInstrumentalLabel']) self.instrumentalModel_var = tk.StringVar(value=data['modelInstrumentalLabel'])
@@ -245,8 +249,10 @@ class MainWindow(TkinterDnD.Tk):
self.winSize_var = tk.StringVar(value=data['window_size']) self.winSize_var = tk.StringVar(value=data['window_size'])
self.agg_var = tk.StringVar(value=data['agg']) self.agg_var = tk.StringVar(value=data['agg'])
# AI model # AI model
#self.aiModel_var = tk.StringVar(value=data['aiModel']) self.aiModel_var = tk.StringVar(value=data['aiModel'])
#self.last_aiModel = self.aiModel_var.get() self.last_aiModel = self.aiModel_var.get()
self.ensChoose_var = tk.StringVar(value=data['ensChoose'])
self.last_ensChoose = self.ensChoose_var.get()
# Other # Other
self.inputPathsEntry_var = tk.StringVar(value='') self.inputPathsEntry_var = tk.StringVar(value='')
self.lastDir = data['lastDir'] # nopep8 self.lastDir = data['lastDir'] # nopep8
@@ -277,9 +283,9 @@ class MainWindow(TkinterDnD.Tk):
self.conversion_Button = ttk.Button(master=self, self.conversion_Button = ttk.Button(master=self,
text='Start Conversion', text='Start Conversion',
command=self.start_conversion) command=self.start_conversion)
self.refresh_Button = ttk.Button(master=self, self.efile_Button = ttk.Button(master=self,
image=self.refresh_img, image=self.efile_img,
command=self.restart) command=self.open_newModel_filedialog)
self.progressbar = ttk.Progressbar(master=self, self.progressbar = ttk.Progressbar(master=self,
variable=self.progress_var) variable=self.progress_var)
@@ -288,7 +294,7 @@ class MainWindow(TkinterDnD.Tk):
background='#a0a0a0', background='#a0a0a0',
borderwidth=0,) borderwidth=0,)
self.command_Text.write(f'COMMAND LINE [{datetime.now().strftime("%Y-%m-%d %H:%M:%S")}]') # nopep8 self.command_Text.write(f'COMMAND LINE [{datetime.now().strftime("%Y-%m-%d %H:%M:%S")}]') # nopep8
def configure_widgets(self): def configure_widgets(self):
"""Change widget styling and appearance""" """Change widget styling and appearance"""
@@ -322,7 +328,7 @@ class MainWindow(TkinterDnD.Tk):
relx=0, rely=0, relwidth=1, relheight=0) relx=0, rely=0, relwidth=1, relheight=0)
self.conversion_Button.place(x=10, y=self.IMAGE_HEIGHT + self.FILEPATHS_HEIGHT + self.OPTIONS_HEIGHT + self.PADDING*2, width=-20 - 40, height=self.CONVERSIONBUTTON_HEIGHT, self.conversion_Button.place(x=10, y=self.IMAGE_HEIGHT + self.FILEPATHS_HEIGHT + self.OPTIONS_HEIGHT + self.PADDING*2, width=-20 - 40, height=self.CONVERSIONBUTTON_HEIGHT,
relx=0, rely=0, relwidth=1, relheight=0) relx=0, rely=0, relwidth=1, relheight=0)
self.refresh_Button.place(x=-10 - 35, y=self.IMAGE_HEIGHT + self.FILEPATHS_HEIGHT + self.OPTIONS_HEIGHT + self.PADDING*2, width=35, height=self.CONVERSIONBUTTON_HEIGHT, self.efile_Button.place(x=-10 - 35, y=self.IMAGE_HEIGHT + self.FILEPATHS_HEIGHT + self.OPTIONS_HEIGHT + self.PADDING*2, width=35, height=self.CONVERSIONBUTTON_HEIGHT,
relx=1, rely=0, relwidth=0, relheight=0) relx=1, rely=0, relwidth=0, relheight=0)
self.command_Text.place(x=15, y=self.IMAGE_HEIGHT + self.FILEPATHS_HEIGHT + self.OPTIONS_HEIGHT + self.CONVERSIONBUTTON_HEIGHT + self.PADDING*3, width=-30, height=self.COMMAND_HEIGHT, self.command_Text.place(x=15, y=self.IMAGE_HEIGHT + self.FILEPATHS_HEIGHT + self.OPTIONS_HEIGHT + self.CONVERSIONBUTTON_HEIGHT + self.PADDING*3, width=-30, height=self.COMMAND_HEIGHT,
relx=0, rely=0, relwidth=1, relheight=0) relx=0, rely=0, relwidth=1, relheight=0)
@@ -380,11 +386,18 @@ class MainWindow(TkinterDnD.Tk):
text='TTA', text='TTA',
variable=self.tta_var, variable=self.tta_var,
) )
# Save Ensemble Outputs
self.options_save_Checkbutton = ttk.Checkbutton(master=self.options_Frame,
text='Save All Outputs',
variable=self.save_var,
)
# Save Image # Save Image
self.options_image_Checkbutton = ttk.Checkbutton(master=self.options_Frame, self.options_image_Checkbutton = ttk.Checkbutton(master=self.options_Frame,
text='Output Image', text='Output Image',
variable=self.outputImage_var, variable=self.outputImage_var,
) )
# Model Test Mode
self.options_modelFolder_Checkbutton = ttk.Checkbutton(master=self.options_Frame, self.options_modelFolder_Checkbutton = ttk.Checkbutton(master=self.options_Frame,
text='Model Test Mode', text='Model Test Mode',
variable=self.modelFolder_var, variable=self.modelFolder_var,
@@ -407,12 +420,20 @@ class MainWindow(TkinterDnD.Tk):
background='#404040', font=self.font, foreground='white', relief="groove") background='#404040', font=self.font, foreground='white', relief="groove")
# AI model # AI model
# self.options_aiModel_Label = tk.Label(master=self.options_Frame, self.options_aiModel_Label = tk.Label(master=self.options_Frame,
# text='Choose AI Engine', anchor=tk.CENTER, text='Choose Conversion Method', anchor=tk.CENTER,
# background='#63605f', font=self.font, foreground='white', relief="sunken") background='#404040', font=self.font, foreground='white', relief="groove")
# self.options_aiModel_Optionmenu = ttk.OptionMenu(self.options_Frame, self.options_aiModel_Optionmenu = ttk.OptionMenu(self.options_Frame,
# self.aiModel_var, self.aiModel_var,
# None, 'v5') None, 'Single Model', 'Ensemble Mode')
# Ensemble Mode
self.options_ensChoose_Label = tk.Label(master=self.options_Frame,
text='Choose Ensemble', anchor=tk.CENTER,
background='#404040', font=self.font, foreground='white', relief="groove")
self.options_ensChoose_Optionmenu = ttk.OptionMenu(self.options_Frame,
self.ensChoose_var,
None, 'HP1 Models', 'HP2 Models', 'All HP Models', 'Vocal Models')
# "Save to", "Select Your Audio File(s)"", and "Start Conversion" Button Style # "Save to", "Select Your Audio File(s)"", and "Start Conversion" Button Style
@@ -428,10 +449,10 @@ class MainWindow(TkinterDnD.Tk):
self.instrumentalModel_var) self.instrumentalModel_var)
# Add Open Export Directory Button # Add Open Export Directory Button
self.options_export_Button = ttk.Button(master=self.options_Frame, # self.options_export_Button = ttk.Button(master=self.options_Frame,
text='Open Export Directory', # text='Open Export Directory',
style="Bold.TButton", # style="Bold.TButton",
command=self.open_newModel_filedialog) # command=self.open_newModel_filedialog)
# -Place Widgets- # -Place Widgets-
# -Column 1- # -Column 1-
self.options_gpu_Checkbutton.place(x=0, y=0, width=0, height=0, self.options_gpu_Checkbutton.place(x=0, y=0, width=0, height=0,
@@ -440,6 +461,8 @@ class MainWindow(TkinterDnD.Tk):
relx=0, rely=1/self.COL1_ROWS, relwidth=1/3, relheight=1/self.COL1_ROWS) relx=0, rely=1/self.COL1_ROWS, relwidth=1/3, relheight=1/self.COL1_ROWS)
self.options_tta_Checkbutton.place(x=0, y=0, width=0, height=0, self.options_tta_Checkbutton.place(x=0, y=0, width=0, height=0,
relx=0, rely=2/self.COL1_ROWS, relwidth=1/3, relheight=1/self.COL1_ROWS) relx=0, rely=2/self.COL1_ROWS, relwidth=1/3, relheight=1/self.COL1_ROWS)
self.options_save_Checkbutton.place(x=0, y=0, width=0, height=0,
relx=0, rely=4/self.COL1_ROWS, relwidth=1/3, relheight=1/self.COL1_ROWS)
self.options_image_Checkbutton.place(x=0, y=0, width=0, height=0, self.options_image_Checkbutton.place(x=0, y=0, width=0, height=0,
relx=0, rely=3/self.COL1_ROWS, relwidth=1/3, relheight=1/self.COL1_ROWS) relx=0, rely=3/self.COL1_ROWS, relwidth=1/3, relheight=1/self.COL1_ROWS)
self.options_modelFolder_Checkbutton.place(x=0, y=0, width=0, height=0, self.options_modelFolder_Checkbutton.place(x=0, y=0, width=0, height=0,
@@ -447,12 +470,22 @@ class MainWindow(TkinterDnD.Tk):
# -Column 2- # -Column 2-
self.options_instrumentalModel_Label.place(x=-15, y=6, width=0, height=-10, self.options_instrumentalModel_Label.place(x=-15, y=6, width=0, height=-10,
relx=1/3, rely=0/self.COL2_ROWS, relwidth=1/3, relheight=1/self.COL2_ROWS) relx=1/3, rely=2/self.COL2_ROWS, relwidth=1/3, relheight=1/self.COL2_ROWS)
self.options_instrumentalModel_Optionmenu.place(x=-15, y=6, width=0, height=-10, self.options_instrumentalModel_Optionmenu.place(x=-15, y=6, width=0, height=-10,
relx=1/3, rely=1/self.COL2_ROWS, relwidth=1/3, relheight=1/self.COL2_ROWS)
self.options_export_Button.place(x=0, y=0, width=-30, height=-8,
relx=1/3, rely=3/self.COL2_ROWS, relwidth=1/3, relheight=1/self.COL2_ROWS) relx=1/3, rely=3/self.COL2_ROWS, relwidth=1/3, relheight=1/self.COL2_ROWS)
self.options_ensChoose_Label.place(x=-15, y=6, width=0, height=-10,
relx=1/3, rely=0/self.COL2_ROWS, relwidth=1/3, relheight=1/self.COL2_ROWS)
self.options_ensChoose_Optionmenu.place(x=-15, y=6, width=0, height=-10,
relx=1/3, rely=1/self.COL2_ROWS, relwidth=1/3, relheight=1/self.COL2_ROWS)
# self.options_export_Button.place(x=0, y=0, width=-30, height=-8,
# relx=2/3, rely=4/self.COL3_ROWS, relwidth=1/3, relheight=1/self.COL3_ROWS)
# -Column 3- # -Column 3-
@@ -469,15 +502,17 @@ class MainWindow(TkinterDnD.Tk):
relx=2/3, rely=3/self.COL3_ROWS, relwidth=1/3, relheight=1/self.COL3_ROWS) relx=2/3, rely=3/self.COL3_ROWS, relwidth=1/3, relheight=1/self.COL3_ROWS)
# AI model # Conversion Method
# self.options_aiModel_Label.place(x=5, y=-5, width=-30, height=-8, self.options_aiModel_Label.place(x=-15, y=6, width=0, height=-10,
# relx=1/3, rely=5/self.COL2_ROWS, relwidth=1/3, relheight=1/self.COL2_ROWS) relx=1/3, rely=0/self.COL2_ROWS, relwidth=1/3, relheight=1/self.COL2_ROWS)
# self.options_aiModel_Optionmenu.place(x=5, y=-5, width=-30, height=-8, self.options_aiModel_Optionmenu.place(x=-15, y=4, width=0, height=-10,
# relx=1/3, rely=6/self.COL2_ROWS, relwidth=1/3, relheight=1/self.COL2_ROWS) relx=1/3, rely=1/self.COL2_ROWS, relwidth=1/3, relheight=1/self.COL2_ROWS)
# Model deselect # Model deselect
# self.aiModel_var.trace_add('write', self.aiModel_var.trace_add('write',
# lambda *args: self.deselect_models()) lambda *args: self.deselect_models())
# Opening filedialogs # Opening filedialogs
def open_file_filedialog(self): def open_file_filedialog(self):
@@ -532,6 +567,7 @@ class MainWindow(TkinterDnD.Tk):
else: else:
window_size = int(self.winSize_var.get()) window_size = int(self.winSize_var.get())
agg = int(self.agg_var.get()) agg = int(self.agg_var.get())
ensChoose = str(self.ensChoose_var.get())
except ValueError: # Non integer was put in entry box except ValueError: # Non integer was put in entry box
tk.messagebox.showwarning(master=self, tk.messagebox.showwarning(master=self,
title='Invalid Input', title='Invalid Input',
@@ -551,25 +587,25 @@ class MainWindow(TkinterDnD.Tk):
message='You have selected an invalid music file! Please make sure that the file still exists!', message='You have selected an invalid music file! Please make sure that the file still exists!',
detail=f'File path: {path}') detail=f'File path: {path}')
return return
if not os.path.isfile(instrumentalModel_path): if self.aiModel_var.get() == 'Single Model':
tk.messagebox.showwarning(master=self, if not os.path.isfile(instrumentalModel_path):
title='Invalid Main Model File', tk.messagebox.showwarning(master=self,
message='You have selected an invalid main model file!\nPlease make sure that your model file still exists!') title='Invalid Main Model File',
return message='You have selected an invalid main model file!\nPlease make sure that your model file still exists!')
return
if not os.path.isdir(export_path): if not os.path.isdir(export_path):
tk.messagebox.showwarning(master=self, tk.messagebox.showwarning(master=self,
title='Invalid Export Directory', title='Invalid Export Directory',
message='You have selected an invalid export directory!\nPlease make sure that your directory still exists!') message='You have selected an invalid export directory!\nPlease make sure that your directory still exists!')
return return
# if self.aiModel_var.get() == 'v4': if self.aiModel_var.get() == 'Single Model':
# inference = inference_v4 inference = inference_v5
# elif self.aiModel_var.get() == 'v5': elif self.aiModel_var.get() == 'Ensemble Mode':
# inference = inference_v5 inference = inference_v5_ensemble
# else: else:
# raise TypeError('This error should not occur.') raise TypeError('This error should not occur.')
inference = inference_v5
# -Run the algorithm- # -Run the algorithm-
threading.Thread(target=inference.main, threading.Thread(target=inference.main,
@@ -580,7 +616,8 @@ class MainWindow(TkinterDnD.Tk):
# Processing Options # Processing Options
'gpu': 0 if self.gpuConversion_var.get() else -1, 'gpu': 0 if self.gpuConversion_var.get() else -1,
'postprocess': self.postprocessing_var.get(), 'postprocess': self.postprocessing_var.get(),
'tta': self.tta_var.get(), # not needed for v2 'tta': self.tta_var.get(),
'save': self.save_var.get(),
'output_image': self.outputImage_var.get(), 'output_image': self.outputImage_var.get(),
# Models # Models
'instrumentalModel': instrumentalModel_path, 'instrumentalModel': instrumentalModel_path,
@@ -591,10 +628,12 @@ class MainWindow(TkinterDnD.Tk):
# Constants # Constants
'window_size': window_size, 'window_size': window_size,
'agg': agg, 'agg': agg,
'ensChoose': ensChoose,
# Other Variables (Tkinter) # Other Variables (Tkinter)
'window': self, 'window': self,
'text_widget': self.command_Text, 'text_widget': self.command_Text,
'button_widget': self.conversion_Button, 'button_widget': self.conversion_Button,
'inst_menu': self.options_instrumentalModel_Optionmenu,
'progress_var': self.progress_var, 'progress_var': self.progress_var,
}, },
daemon=True daemon=True
@@ -647,23 +686,55 @@ class MainWindow(TkinterDnD.Tk):
on certain selections on certain selections
""" """
# Models if self.aiModel_var.get() == 'Single Model':
# self.options_instrumentalModel_Label.configure(foreground='#000') self.options_ensChoose_Label.place_forget()
# self.options_instrumentalModel_Optionmenu.configure(state=tk.NORMAL) # nopep8 self.options_ensChoose_Optionmenu.place_forget()
self.options_save_Checkbutton.configure(state=tk.DISABLED)
# if self.aiModel_var.get() == 'v5': self.options_save_Checkbutton.place_forget()
# self.options_tta_Checkbutton.configure(state=tk.NORMAL) self.options_modelFolder_Checkbutton.configure(state=tk.NORMAL)
# self.options_agg_Label.place(x=5, y=-5, width=-30, height=-8, self.options_modelFolder_Checkbutton.place(x=0, y=0, width=0, height=0,
# relx=1/3, rely=3/self.COL2_ROWS, relwidth=1/3, relheight=1/self.COL2_ROWS) relx=0, rely=4/self.COL1_ROWS, relwidth=1/3, relheight=1/self.COL1_ROWS)
# self.options_agg_Entry.place(x=5, y=-4, width=-30, height=-8, self.options_instrumentalModel_Label.place(x=-15, y=6, width=0, height=-10,
# relx=1/3, rely=4/self.COL2_ROWS, relwidth=1/3, relheight=1/self.COL2_ROWS) relx=1/3, rely=3/self.COL2_ROWS, relwidth=1/3, relheight=1/self.COL2_ROWS)
self.options_instrumentalModel_Optionmenu.place(x=-15, y=6, width=0, height=-10,
relx=1/3, rely=4/self.COL2_ROWS, relwidth=1/3, relheight=1/self.COL2_ROWS)
else:
self.options_instrumentalModel_Label.place_forget()
self.options_instrumentalModel_Optionmenu.place_forget()
self.options_modelFolder_Checkbutton.place_forget()
self.options_modelFolder_Checkbutton.configure(state=tk.DISABLED)
self.options_save_Checkbutton.configure(state=tk.NORMAL)
self.options_save_Checkbutton.place(x=0, y=0, width=0, height=0,
relx=0, rely=4/self.COL1_ROWS, relwidth=1/3, relheight=1/self.COL1_ROWS)
self.options_ensChoose_Label.place(x=-15, y=6, width=0, height=-10,
relx=1/3, rely=3/self.COL2_ROWS, relwidth=1/3, relheight=1/self.COL2_ROWS)
self.options_ensChoose_Optionmenu.place(x=-15, y=6, width=0, height=-10,
relx=1/3, rely=4/self.COL2_ROWS, relwidth=1/3, relheight=1/self.COL2_ROWS)
# else: if self.aiModel_var.get() == 'Ensemble Mode':
# self.options_tta_Checkbutton.configure(state=tk.NORMAL) self.options_instrumentalModel_Label.place_forget()
# self.options_agg_Label.place(x=5, y=-5, width=-30, height=-8, self.options_instrumentalModel_Optionmenu.place_forget()
# relx=1/3, rely=3/self.COL2_ROWS, relwidth=1/3, relheight=1/self.COL2_ROWS) self.options_modelFolder_Checkbutton.place_forget()
# self.options_agg_Entry.place(x=5, y=-4, width=-30, height=-8, self.options_modelFolder_Checkbutton.configure(state=tk.DISABLED)
# relx=1/3, rely=4/self.COL2_ROWS, relwidth=1/3, relheight=1/self.COL2_ROWS) self.options_save_Checkbutton.configure(state=tk.NORMAL)
self.options_save_Checkbutton.place(x=0, y=0, width=0, height=0,
relx=0, rely=4/self.COL1_ROWS, relwidth=1/3, relheight=1/self.COL1_ROWS)
self.options_ensChoose_Label.place(x=-15, y=6, width=0, height=-10,
relx=1/3, rely=3/self.COL2_ROWS, relwidth=1/3, relheight=1/self.COL2_ROWS)
self.options_ensChoose_Optionmenu.place(x=-15, y=6, width=0, height=-10,
relx=1/3, rely=4/self.COL2_ROWS, relwidth=1/3, relheight=1/self.COL2_ROWS)
else:
self.options_ensChoose_Label.place_forget()
self.options_ensChoose_Optionmenu.place_forget()
self.options_save_Checkbutton.configure(state=tk.DISABLED)
self.options_save_Checkbutton.place_forget()
self.options_modelFolder_Checkbutton.configure(state=tk.NORMAL)
self.options_modelFolder_Checkbutton.place(x=0, y=0, width=0, height=0,
relx=0, rely=4/self.COL1_ROWS, relwidth=1/3, relheight=1/self.COL1_ROWS)
self.options_instrumentalModel_Label.place(x=-15, y=6, width=0, height=-10,
relx=1/3, rely=3/self.COL2_ROWS, relwidth=1/3, relheight=1/self.COL2_ROWS)
self.options_instrumentalModel_Optionmenu.place(x=-15, y=6, width=0, height=-10,
relx=1/3, rely=4/self.COL2_ROWS, relwidth=1/3, relheight=1/self.COL2_ROWS)
self.update_inputPaths() self.update_inputPaths()
@@ -678,23 +749,26 @@ class MainWindow(TkinterDnD.Tk):
self.last_aiModel = self.aiModel_var.get() self.last_aiModel = self.aiModel_var.get()
self.instrumentalModel_var.set('') self.instrumentalModel_var.set('')
self.ensChoose_var.set('HP1 Models')
self.winSize_var.set(DEFAULT_DATA['window_size']) self.winSize_var.set(DEFAULT_DATA['window_size'])
self.agg_var.set(DEFAULT_DATA['agg']) self.agg_var.set(DEFAULT_DATA['agg'])
self.modelFolder_var.set(DEFAULT_DATA['modelFolder'])
self.update_available_models() self.update_available_models()
self.update_states() self.update_states()
def restart(self): # def restart(self):
""" # """
Restart the application after asking for confirmation # Restart the application after asking for confirmation
""" # """
save = tk.messagebox.askyesno(title='Confirmation', # save = tk.messagebox.askyesno(title='Confirmation',
message='The application will restart. Do you want to save the data?') # message='The application will restart. Do you want to save the data?')
if save: # if save:
self.save_values() # self.save_values()
subprocess.Popen(f'python "{__file__}"', shell=True) # subprocess.Popen(f'..App\Python\python.exe "{__file__}"')
exit() # exit()
def save_values(self): def save_values(self):
""" """
@@ -716,6 +790,7 @@ class MainWindow(TkinterDnD.Tk):
'gpu': self.gpuConversion_var.get(), 'gpu': self.gpuConversion_var.get(),
'postprocess': self.postprocessing_var.get(), 'postprocess': self.postprocessing_var.get(),
'tta': self.tta_var.get(), 'tta': self.tta_var.get(),
'save': self.save_var.get(),
'output_image': self.outputImage_var.get(), 'output_image': self.outputImage_var.get(),
'window_size': window_size, 'window_size': window_size,
'agg': agg, 'agg': agg,
@@ -723,7 +798,8 @@ class MainWindow(TkinterDnD.Tk):
'lastDir': self.lastDir, 'lastDir': self.lastDir,
'modelFolder': self.modelFolder_var.get(), 'modelFolder': self.modelFolder_var.get(),
'modelInstrumentalLabel': self.instrumentalModel_var.get(), 'modelInstrumentalLabel': self.instrumentalModel_var.get(),
#'aiModel': self.aiModel_var.get(), 'aiModel': self.aiModel_var.get(),
'ensChoose': self.ensChoose_var.get(),
}) })
self.destroy() self.destroy()

481
inference_v5.py
View File

@@ -30,70 +30,8 @@ class VocalRemover(object):
self.text_widget = text_widget self.text_widget = text_widget
self.models = defaultdict(lambda: None) self.models = defaultdict(lambda: None)
self.devices = defaultdict(lambda: None) self.devices = defaultdict(lambda: None)
self._load_models()
# self.offset = model.offset # self.offset = model.offset
def _load_models(self):
self.text_widget.write('Loading models...\n') # nopep8 Write Command Text
nn_arch_sizes = [
31191, # default
33966, 123821, 123812, 537238 # custom
]
global args
global model_params_d
p = argparse.ArgumentParser()
p.add_argument('--paramone', type=str, default='lib_v5/modelparams/4band_44100.json')
p.add_argument('--paramtwo', type=str, default='lib_v5/modelparams/4band_v2.json')
p.add_argument('--paramthree', type=str, default='lib_v5/modelparams/3band_44100_msb2.json')
p.add_argument('--paramfour', type=str, default='lib_v5/modelparams/4band_v2_sn.json')
p.add_argument('--aggressiveness',type=float, default=data['agg']/100)
p.add_argument('--nn_architecture', type=str, choices= ['auto'] + list('{}KB'.format(s) for s in nn_arch_sizes), default='auto')
p.add_argument('--high_end_process', type=str, default='mirroring')
args = p.parse_args()
if 'auto' == args.nn_architecture:
model_size = math.ceil(os.stat(data['instrumentalModel']).st_size / 1024)
args.nn_architecture = '{}KB'.format(min(nn_arch_sizes, key=lambda x:abs(x-model_size)))
nets = importlib.import_module('lib_v5.nets' + f'_{args.nn_architecture}'.replace('_{}KB'.format(nn_arch_sizes[0]), ''), package=None)
ModelName=(data['instrumentalModel'])
ModelParam1="4BAND_44100"
ModelParam2="4BAND_44100_B"
ModelParam3="MSB2"
ModelParam4="4BAND_44100_SN"
if ModelParam1 in ModelName:
model_params_d=args.paramone
if ModelParam2 in ModelName:
model_params_d=args.paramtwo
if ModelParam3 in ModelName:
model_params_d=args.paramthree
if ModelParam4 in ModelName:
model_params_d=args.paramfour
print(model_params_d)
mp = ModelParameters(model_params_d)
# -Instrumental-
if os.path.isfile(data['instrumentalModel']):
device = torch.device('cpu')
model = nets.CascadedASPPNet(mp.param['bins'] * 2)
model.load_state_dict(torch.load(self.data['instrumentalModel'],
map_location=device))
if torch.cuda.is_available() and self.data['gpu'] >= 0:
device = torch.device('cuda:{}'.format(self.data['gpu']))
model.to(device)
self.models['instrumental'] = model
self.devices['instrumental'] = device
self.text_widget.write('Done!\n')
data = { data = {
# Paths # Paths
@@ -152,6 +90,26 @@ def determineModelFolderName():
def main(window: tk.Wm, text_widget: tk.Text, button_widget: tk.Button, progress_var: tk.Variable, def main(window: tk.Wm, text_widget: tk.Text, button_widget: tk.Button, progress_var: tk.Variable,
**kwargs: dict): **kwargs: dict):
global args
global model_params_d
global nn_arch_sizes
nn_arch_sizes = [
31191, # default
33966, 123821, 123812, 537238 # custom
]
p = argparse.ArgumentParser()
p.add_argument('--paramone', type=str, default='lib_v5/modelparams/4band_44100.json')
p.add_argument('--paramtwo', type=str, default='lib_v5/modelparams/4band_v2.json')
p.add_argument('--paramthree', type=str, default='lib_v5/modelparams/3band_44100_msb2.json')
p.add_argument('--paramfour', type=str, default='lib_v5/modelparams/4band_v2_sn.json')
p.add_argument('--aggressiveness',type=float, default=data['agg']/100)
p.add_argument('--nn_architecture', type=str, choices= ['auto'] + list('{}KB'.format(s) for s in nn_arch_sizes), default='auto')
p.add_argument('--high_end_process', type=str, default='mirroring')
args = p.parse_args()
def save_files(wav_instrument, wav_vocals): def save_files(wav_instrument, wav_vocals):
"""Save output music files""" """Save output music files"""
@@ -215,210 +173,255 @@ def main(window: tk.Wm, text_widget: tk.Text, button_widget: tk.Button, progress
os.mkdir(folder_path) os.mkdir(folder_path)
# Separation Preperation # Separation Preperation
try: try: #Load File(s)
for file_num, music_file in enumerate(data['input_paths'], start=1): for file_num, music_file in enumerate(data['input_paths'], start=1):
# Determine File Name # Determine File Name
base_name = f'{data["export_path"]}{modelFolderName}/{file_num}_{os.path.splitext(os.path.basename(music_file))[0]}' base_name = f'{data["export_path"]}{modelFolderName}/{file_num}_{os.path.splitext(os.path.basename(music_file))[0]}'
# Start Separation
model_name = os.path.basename(data[f'{data["useModel"]}Model'])
model = vocal_remover.models[data['useModel']]
device = vocal_remover.devices[data['useModel']]
# -Get text and update progress-
base_text = get_baseText(total_files=len(data['input_paths']),
file_num=file_num)
progress_kwargs = {'progress_var': progress_var,
'total_files': len(data['input_paths']),
'file_num': file_num}
update_progress(**progress_kwargs,
step=0)
mp = ModelParameters(model_params_d)
# -Go through the different steps of seperation- model_name = os.path.basename(data[f'{data["useModel"]}Model'])
# Wave source model = vocal_remover.models[data['useModel']]
text_widget.write(base_text + 'Loading wave source...\n') device = vocal_remover.devices[data['useModel']]
# -Get text and update progress-
X_wave, y_wave, X_spec_s, y_spec_s = {}, {}, {}, {} base_text = get_baseText(total_files=len(data['input_paths']),
file_num=file_num)
bands_n = len(mp.param['band']) progress_kwargs = {'progress_var': progress_var,
'total_files': len(data['input_paths']),
for d in range(bands_n, 0, -1): 'file_num': file_num}
bp = mp.param['band'][d] update_progress(**progress_kwargs,
step=0)
if d == bands_n: # high-end band
X_wave[d], _ = librosa.load(
music_file, bp['sr'], False, dtype=np.float32, res_type=bp['res_type']) #Load Model(s)
text_widget.write(base_text + 'Loading models...')
if X_wave[d].ndim == 1:
X_wave[d] = np.asarray([X_wave[d], X_wave[d]])
else: # lower bands if 'auto' == args.nn_architecture:
X_wave[d] = librosa.resample(X_wave[d+1], mp.param['band'][d+1]['sr'], bp['sr'], res_type=bp['res_type']) model_size = math.ceil(os.stat(data['instrumentalModel']).st_size / 1024)
args.nn_architecture = '{}KB'.format(min(nn_arch_sizes, key=lambda x:abs(x-model_size)))
nets = importlib.import_module('lib_v5.nets' + f'_{args.nn_architecture}'.replace('_{}KB'.format(nn_arch_sizes[0]), ''), package=None)
ModelName=(data['instrumentalModel'])
ModelParam1="4BAND_44100"
ModelParam2="4BAND_44100_B"
ModelParam3="MSB2"
ModelParam4="4BAND_44100_SN"
if ModelParam1 in ModelName:
model_params_d=args.paramone
if ModelParam2 in ModelName:
model_params_d=args.paramtwo
if ModelParam3 in ModelName:
model_params_d=args.paramthree
if ModelParam4 in ModelName:
model_params_d=args.paramfour
# Stft of wave source print('Model Parameters:', model_params_d)
X_spec_s[d] = spec_utils.wave_to_spectrogram_mt(X_wave[d], bp['hl'], bp['n_fft'], mp.param['mid_side'], mp = ModelParameters(model_params_d)
mp.param['mid_side_b2'], mp.param['reverse'])
if d == bands_n and args.high_end_process != 'none': # -Instrumental-
input_high_end_h = (bp['n_fft']//2 - bp['crop_stop']) + (mp.param['pre_filter_stop'] - mp.param['pre_filter_start']) if os.path.isfile(data['instrumentalModel']):
input_high_end = X_spec_s[d][:, bp['n_fft']//2-input_high_end_h:bp['n_fft']//2, :] device = torch.device('cpu')
model = nets.CascadedASPPNet(mp.param['bins'] * 2)
text_widget.write(base_text + 'Done!\n') model.load_state_dict(torch.load(data['instrumentalModel'],
map_location=device))
update_progress(**progress_kwargs, if torch.cuda.is_available() and data['gpu'] >= 0:
step=0.1) device = torch.device('cuda:{}'.format(data['gpu']))
model.to(device)
text_widget.write(base_text + 'Stft of wave source...\n')
text_widget.write(base_text + 'Done!\n')
text_widget.write(base_text + "Please Wait..\n")
X_spec_m = spec_utils.combine_spectrograms(X_spec_s, mp)
del X_wave, X_spec_s
def inference(X_spec, device, model, aggressiveness):
def _execute(X_mag_pad, roi_size, n_window, device, model, aggressiveness):
model.eval()
with torch.no_grad(): vocal_remover.models['instrumental'] = model
preds = [] vocal_remover.devices['instrumental'] = device
iterations = [n_window]
total_iterations = sum(iterations) text_widget.write(' Done!\n')
model_name = os.path.basename(data[f'{data["useModel"]}Model'])
mp = ModelParameters(model_params_d)
text_widget.write(base_text + "Length: "f"{total_iterations} Slices\n") # -Go through the different steps of seperation-
# Wave source
for i in tqdm(range(n_window)): text_widget.write(base_text + 'Loading wave source...')
update_progress(**progress_kwargs,
step=(0.1 + (0.8/n_window * i))) X_wave, y_wave, X_spec_s, y_spec_s = {}, {}, {}, {}
start = i * roi_size
X_mag_window = X_mag_pad[None, :, :, start:start + data['window_size']] bands_n = len(mp.param['band'])
X_mag_window = torch.from_numpy(X_mag_window).to(device)
for d in range(bands_n, 0, -1):
pred = model.predict(X_mag_window, aggressiveness) bp = mp.param['band'][d]
pred = pred.detach().cpu().numpy() if d == bands_n: # high-end band
preds.append(pred[0]) X_wave[d], _ = librosa.load(
music_file, bp['sr'], False, dtype=np.float32, res_type=bp['res_type'])
pred = np.concatenate(preds, axis=2) if X_wave[d].ndim == 1:
X_wave[d] = np.asarray([X_wave[d], X_wave[d]])
else: # lower bands
X_wave[d] = librosa.resample(X_wave[d+1], mp.param['band'][d+1]['sr'], bp['sr'], res_type=bp['res_type'])
# Stft of wave source
return pred X_spec_s[d] = spec_utils.wave_to_spectrogram_mt(X_wave[d], bp['hl'], bp['n_fft'], mp.param['mid_side'],
mp.param['mid_side_b2'], mp.param['reverse'])
if d == bands_n and args.high_end_process != 'none':
input_high_end_h = (bp['n_fft']//2 - bp['crop_stop']) + (mp.param['pre_filter_stop'] - mp.param['pre_filter_start'])
input_high_end = X_spec_s[d][:, bp['n_fft']//2-input_high_end_h:bp['n_fft']//2, :]
def preprocess(X_spec): text_widget.write('Done!\n')
X_mag = np.abs(X_spec)
X_phase = np.angle(X_spec)
return X_mag, X_phase update_progress(**progress_kwargs,
step=0.1)
text_widget.write(base_text + 'Stft of wave source...')
X_mag, X_phase = preprocess(X_spec) text_widget.write(' Done!\n')
coef = X_mag.max()
X_mag_pre = X_mag / coef
n_frame = X_mag_pre.shape[2]
pad_l, pad_r, roi_size = dataset.make_padding(n_frame,
data['window_size'], model.offset)
n_window = int(np.ceil(n_frame / roi_size))
X_mag_pad = np.pad(
X_mag_pre, ((0, 0), (0, 0), (pad_l, pad_r)), mode='constant')
pred = _execute(X_mag_pad, roi_size, n_window, text_widget.write(base_text + "Please Wait...\n")
device, model, aggressiveness)
pred = pred[:, :, :n_frame] X_spec_m = spec_utils.combine_spectrograms(X_spec_s, mp)
if data['tta']: del X_wave, X_spec_s
pad_l += roi_size // 2
pad_r += roi_size // 2 def inference(X_spec, device, model, aggressiveness):
n_window += 1
def _execute(X_mag_pad, roi_size, n_window, device, model, aggressiveness):
model.eval()
with torch.no_grad():
preds = []
iterations = [n_window]
total_iterations = sum(iterations)
text_widget.write(base_text + "Processing "f"{total_iterations} Slices... ")
for i in tqdm(range(n_window)):
update_progress(**progress_kwargs,
step=(0.1 + (0.8/n_window * i)))
start = i * roi_size
X_mag_window = X_mag_pad[None, :, :, start:start + data['window_size']]
X_mag_window = torch.from_numpy(X_mag_window).to(device)
pred = model.predict(X_mag_window, aggressiveness)
pred = pred.detach().cpu().numpy()
preds.append(pred[0])
pred = np.concatenate(preds, axis=2)
text_widget.write('Done!\n')
return pred
def preprocess(X_spec):
X_mag = np.abs(X_spec)
X_phase = np.angle(X_spec)
return X_mag, X_phase
X_mag, X_phase = preprocess(X_spec)
coef = X_mag.max()
X_mag_pre = X_mag / coef
n_frame = X_mag_pre.shape[2]
pad_l, pad_r, roi_size = dataset.make_padding(n_frame,
data['window_size'], model.offset)
n_window = int(np.ceil(n_frame / roi_size))
X_mag_pad = np.pad( X_mag_pad = np.pad(
X_mag_pre, ((0, 0), (0, 0), (pad_l, pad_r)), mode='constant') X_mag_pre, ((0, 0), (0, 0), (pad_l, pad_r)), mode='constant')
pred = _execute(X_mag_pad, roi_size, n_window,
device, model, aggressiveness)
pred = pred[:, :, :n_frame]
if data['tta']:
pad_l += roi_size // 2
pad_r += roi_size // 2
n_window += 1
pred_tta = _execute(X_mag_pad, roi_size, n_window, X_mag_pad = np.pad(
device, model, aggressiveness) X_mag_pre, ((0, 0), (0, 0), (pad_l, pad_r)), mode='constant')
pred_tta = pred_tta[:, :, roi_size // 2:]
pred_tta = pred_tta[:, :, :n_frame]
return (pred + pred_tta) * 0.5 * coef, X_mag, np.exp(1.j * X_phase) pred_tta = _execute(X_mag_pad, roi_size, n_window,
device, model, aggressiveness)
pred_tta = pred_tta[:, :, roi_size // 2:]
pred_tta = pred_tta[:, :, :n_frame]
return (pred + pred_tta) * 0.5 * coef, X_mag, np.exp(1.j * X_phase)
else:
return pred * coef, X_mag, np.exp(1.j * X_phase)
aggressiveness = {'value': args.aggressiveness, 'split_bin': mp.param['band'][1]['crop_stop']}
if data['tta']:
text_widget.write(base_text + "Running Inferences (TTA)...\n")
else: else:
return pred * coef, X_mag, np.exp(1.j * X_phase) text_widget.write(base_text + "Running Inference...\n")
pred, X_mag, X_phase = inference(X_spec_m,
aggressiveness = {'value': args.aggressiveness, 'split_bin': mp.param['band'][1]['crop_stop']} device,
model, aggressiveness)
if data['tta']:
text_widget.write(base_text + "Running Inferences (TTA)...\n")
else:
text_widget.write(base_text + "Running Inference...\n")
pred, X_mag, X_phase = inference(X_spec_m,
device,
model, aggressiveness)
text_widget.write(base_text + 'Done!\n')
update_progress(**progress_kwargs,
step=0.9)
# Postprocess
if data['postprocess']:
text_widget.write(base_text + 'Post processing...\n')
pred_inv = np.clip(X_mag - pred, 0, np.inf)
pred = spec_utils.mask_silence(pred, pred_inv)
text_widget.write(base_text + 'Done!\n')
update_progress(**progress_kwargs, update_progress(**progress_kwargs,
step=0.95) step=0.9)
# Postprocess
if data['postprocess']:
text_widget.write(base_text + 'Post processing...')
pred_inv = np.clip(X_mag - pred, 0, np.inf)
pred = spec_utils.mask_silence(pred, pred_inv)
text_widget.write(' Done!\n')
# Inverse stft update_progress(**progress_kwargs,
text_widget.write(base_text + 'Inverse stft of instruments and vocals...\n') # nopep8 step=0.95)
y_spec_m = pred * X_phase
v_spec_m = X_spec_m - y_spec_m
if args.high_end_process.startswith('mirroring'):
input_high_end_ = spec_utils.mirroring(args.high_end_process, y_spec_m, input_high_end, mp)
wav_instrument = spec_utils.cmb_spectrogram_to_wave(y_spec_m, mp, input_high_end_h, input_high_end_)
else:
wav_instrument = spec_utils.cmb_spectrogram_to_wave(y_spec_m, mp)
if args.high_end_process.startswith('mirroring'): # Inverse stft
input_high_end_ = spec_utils.mirroring(args.high_end_process, v_spec_m, input_high_end, mp) text_widget.write(base_text + 'Inverse stft of instruments and vocals...') # nopep8
y_spec_m = pred * X_phase
v_spec_m = X_spec_m - y_spec_m
if args.high_end_process.startswith('mirroring'):
input_high_end_ = spec_utils.mirroring(args.high_end_process, y_spec_m, input_high_end, mp)
wav_vocals = spec_utils.cmb_spectrogram_to_wave(v_spec_m, mp, input_high_end_h, input_high_end_) wav_instrument = spec_utils.cmb_spectrogram_to_wave(y_spec_m, mp, input_high_end_h, input_high_end_)
else: else:
wav_vocals = spec_utils.cmb_spectrogram_to_wave(v_spec_m, mp) wav_instrument = spec_utils.cmb_spectrogram_to_wave(y_spec_m, mp)
text_widget.write(base_text + 'Done!\n')
update_progress(**progress_kwargs, if args.high_end_process.startswith('mirroring'):
step=1) input_high_end_ = spec_utils.mirroring(args.high_end_process, v_spec_m, input_high_end, mp)
# Save output music files
text_widget.write(base_text + 'Saving Files...\n')
save_files(wav_instrument, wav_vocals)
text_widget.write(base_text + 'Done!\n')
update_progress(**progress_kwargs, wav_vocals = spec_utils.cmb_spectrogram_to_wave(v_spec_m, mp, input_high_end_h, input_high_end_)
step=1) else:
wav_vocals = spec_utils.cmb_spectrogram_to_wave(v_spec_m, mp)
text_widget.write('Done!\n')
# Save output image update_progress(**progress_kwargs,
if data['output_image']: step=1)
with open('{}_Instruments.jpg'.format(base_name), mode='wb') as f: # Save output music files
image = spec_utils.spectrogram_to_image(y_spec_m) text_widget.write(base_text + 'Saving Files...')
_, bin_image = cv2.imencode('.jpg', image) save_files(wav_instrument, wav_vocals)
bin_image.tofile(f) text_widget.write(' Done!\n')
with open('{}_Vocals.jpg'.format(base_name), mode='wb') as f:
image = spec_utils.spectrogram_to_image(v_spec_m)
_, bin_image = cv2.imencode('.jpg', image)
bin_image.tofile(f)
text_widget.write(base_text + 'Completed Seperation!\n\n') update_progress(**progress_kwargs,
step=1)
# Save output image
if data['output_image']:
with open('{}_Instruments.jpg'.format(base_name), mode='wb') as f:
image = spec_utils.spectrogram_to_image(y_spec_m)
_, bin_image = cv2.imencode('.jpg', image)
bin_image.tofile(f)
with open('{}_Vocals.jpg'.format(base_name), mode='wb') as f:
image = spec_utils.spectrogram_to_image(v_spec_m)
_, bin_image = cv2.imencode('.jpg', image)
bin_image.tofile(f)
text_widget.write(base_text + 'Completed Seperation!\n\n')
except Exception as e: except Exception as e:
traceback_text = ''.join(traceback.format_tb(e.__traceback__)) traceback_text = ''.join(traceback.format_tb(e.__traceback__))
message = f'Traceback Error: "{traceback_text}"\n{type(e).__name__}: "{e}"\nFile: {music_file}\nPlease contact the creator and attach a screenshot of this error with the file and settings that caused it!' message = f'Traceback Error: "{traceback_text}"\n{type(e).__name__}: "{e}"\nFile: {music_file}\nPlease contact the creator and attach a screenshot of this error with the file and settings that caused it!'
@@ -433,9 +436,9 @@ def main(window: tk.Wm, text_widget: tk.Text, button_widget: tk.Button, progress
return return
os.remove('temp.wav') os.remove('temp.wav')
progress_var.set(0) progress_var.set(0)
text_widget.write(f'Conversion(s) Completed and Saving all Files!\n') text_widget.write(f'\nConversion(s) Completed!\n')
text_widget.write(f'Time Elapsed: {time.strftime("%H:%M:%S", time.gmtime(int(time.perf_counter() - stime)))}') # nopep8 text_widget.write(f'Time Elapsed: {time.strftime("%H:%M:%S", time.gmtime(int(time.perf_counter() - stime)))}') # nopep8
torch.cuda.empty_cache() torch.cuda.empty_cache()
button_widget.configure(state=tk.NORMAL) # Enable Button button_widget.configure(state=tk.NORMAL) # Enable Button

623
inference_v5_ensemble.py Normal file
View File

@@ -0,0 +1,623 @@
from functools import total_ordering
import pprint
import argparse
import os
from statistics import mode
import cv2
import librosa
import numpy as np
import soundfile as sf
import shutil
from tqdm import tqdm
from lib_v5 import dataset
from lib_v5 import spec_utils
from lib_v5.model_param_init import ModelParameters
import torch
# Command line text parsing and widget manipulation
from collections import defaultdict
import tkinter as tk
import traceback # Error Message Recent Calls
import time # Timer
class VocalRemover(object):
def __init__(self, data, text_widget: tk.Text):
self.data = data
self.text_widget = text_widget
# self.offset = model.offset
data = {
# Paths
'input_paths': None,
'export_path': None,
# Processing Options
'gpu': -1,
'postprocess': True,
'tta': True,
'save': True,
'output_image': True,
# Models
'instrumentalModel': None,
'useModel': None,
# Constants
'window_size': 512,
'agg': 10,
'ensChoose': 'HP1 Models'
}
default_window_size = data['window_size']
default_agg = data['agg']
def update_progress(progress_var, total_files, file_num, step: float = 1):
"""Calculate the progress for the progress widget in the GUI"""
base = (100 / total_files)
progress = base * (file_num - 1)
progress += base * step
progress_var.set(progress)
def get_baseText(total_files, file_num):
"""Create the base text for the command widget"""
text = 'File {file_num}/{total_files} '.format(file_num=file_num,
total_files=total_files)
return text
def main(window: tk.Wm, text_widget: tk.Text, button_widget: tk.Button, progress_var: tk.Variable,
**kwargs: dict):
global args
global nn_arch_sizes
nn_arch_sizes = [
31191, # default
33966, 123821, 123812, 537238 # custom
]
p = argparse.ArgumentParser()
p.add_argument('--aggressiveness',type=float, default=data['agg']/100)
p.add_argument('--high_end_process', type=str, default='mirroring')
args = p.parse_args()
def save_files(wav_instrument, wav_vocals):
"""Save output music files"""
vocal_name = '(Vocals)'
instrumental_name = '(Instrumental)'
save_path = os.path.dirname(base_name)
# Swap names if vocal model
VModel="Vocal"
if VModel in model_name:
# Reverse names
vocal_name, instrumental_name = instrumental_name, vocal_name
# Save Temp File
# For instrumental the instrumental is the temp file
# and for vocal the instrumental is the temp file due
# to reversement
sf.write(f'temp.wav',
wav_instrument, mp.param['sr'])
# -Save files-
# Instrumental
if instrumental_name is not None:
instrumental_path = '{save_path}/{file_name}.wav'.format(
save_path=save_path,
file_name = f'{os.path.basename(base_name)}_{ModelName_1}_{instrumental_name}',
)
sf.write(instrumental_path,
wav_instrument, mp.param['sr'])
# Vocal
if vocal_name is not None:
vocal_path = '{save_path}/{file_name}.wav'.format(
save_path=save_path,
file_name=f'{os.path.basename(base_name)}_{ModelName_1}_{vocal_name}',
)
sf.write(vocal_path,
wav_vocals, mp.param['sr'])
data.update(kwargs)
# Update default settings
global default_window_size
global default_agg
default_window_size = data['window_size']
default_agg = data['agg']
stime = time.perf_counter()
progress_var.set(0)
text_widget.clear()
button_widget.configure(state=tk.DISABLED) # Disable Button
# Separation Preperation
try: #Load File(s)
HP1_Models = [
{
'model_name':'HP_4BAND_44100_A',
'model_params':'lib_v5/modelparams/4band_44100.json',
'model_location':'models/Main Models/HP_4BAND_44100_A.pth',
'using_archtecture': '123821KB',
'loop_name': 'Ensemble Mode - Model 1/2'
},
{
'model_name':'HP_4BAND_44100_B',
'model_params':'lib_v5/modelparams/4band_v2.json',
'model_location':'models/Main Models/HP_4BAND_44100_B.pth',
'using_archtecture': '123821KB',
'loop_name': 'Ensemble Mode - Model 2/2'
}
]
HP2_Models = [
{
'model_name':'HP2_4BAND_44100_1',
'model_params':'lib_v5/modelparams/4band_44100.json',
'model_location':'models/Main Models/HP2_4BAND_44100_1.pth',
'using_archtecture': '537238KB',
'loop_name': 'Ensemble Mode - Model 1/3'
},
{
'model_name':'HP2_4BAND_44100_2',
'model_params':'lib_v5/modelparams/4band_44100.json',
'model_location':'models/Main Models/HP2_4BAND_44100_2.pth',
'using_archtecture': '537238KB',
'loop_name': 'Ensemble Mode - Model 2/3'
},
{
'model_name':'HP2_3BAND_44100_MSB2',
'model_params':'lib_v5/modelparams/3band_44100_msb2.json',
'model_location':'models/Main Models/HP2_3BAND_44100_MSB2.pth',
'using_archtecture': '537227KB',
'loop_name': 'Ensemble Mode - Model 3/3'
}
]
All_HP_Models = [
{
'model_name':'HP_4BAND_44100_A',
'model_params':'lib_v5/modelparams/4band_44100.json',
'model_location':'models/Main Models/HP_4BAND_44100_A.pth',
'using_archtecture': '123821KB',
'loop_name': 'Ensemble Mode - Model 1/5'
},
{
'model_name':'HP_4BAND_44100_B',
'model_params':'lib_v5/modelparams/4band_v2.json',
'model_location':'models/Main Models/HP_4BAND_44100_B.pth',
'using_archtecture': '123821KB',
'loop_name': 'Ensemble Mode - Model 2/5'
},
{
'model_name':'HP2_4BAND_44100_1',
'model_params':'lib_v5/modelparams/4band_44100.json',
'model_location':'models/Main Models/HP2_4BAND_44100_1.pth',
'using_archtecture': '537238KB',
'loop_name': 'Ensemble Mode - Model 3/5'
},
{
'model_name':'HP2_4BAND_44100_2',
'model_params':'lib_v5/modelparams/4band_44100.json',
'model_location':'models/Main Models/HP2_4BAND_44100_2.pth',
'using_archtecture': '537238KB',
'loop_name': 'Ensemble Mode - Model 4/5'
},
{
'model_name':'HP2_3BAND_44100_MSB2',
'model_params':'lib_v5/modelparams/3band_44100_msb2.json',
'model_location':'models/Main Models/HP2_3BAND_44100_MSB2.pth',
'using_archtecture': '537227KB',
'loop_name': 'Ensemble Mode - Model 5/5'
}
]
Vocal_Models = [
{
'model_name':'HP_Vocal_4BAND_44100',
'model_params':'lib_v5/modelparams/4band_44100.json',
'model_location':'models/Main Models/HP_Vocal_4BAND_44100.pth',
'using_archtecture': '123821KB',
'loop_name': 'Ensemble Mode - Model 1/2'
},
{
'model_name':'HP_Vocal_AGG_4BAND_44100',
'model_params':'lib_v5/modelparams/4band_44100.json',
'model_location':'models/Main Models/HP_Vocal_AGG_4BAND_44100.pth',
'using_archtecture': '123821KB',
'loop_name': 'Ensemble Mode - Model 2/2'
}
]
if data['ensChoose'] == 'HP1 Models':
loops = HP1_Models
ensefolder = 'HP_Models_Saved_Outputs'
ensemode = 'HP_Models'
if data['ensChoose'] == 'HP2 Models':
loops = HP2_Models
ensefolder = 'HP2_Models_Saved_Outputs'
ensemode = 'HP2_Models'
if data['ensChoose'] == 'All HP Models':
loops = All_HP_Models
ensefolder = 'All_HP_Models_Saved_Outputs'
ensemode = 'All_HP_Models'
if data['ensChoose'] == 'Vocal Models':
loops = Vocal_Models
ensefolder = 'Vocal_Models_Saved_Outputs'
ensemode = 'Vocal_Models'
for file_num, music_file in enumerate(data['input_paths'], start=1):
# -Get text and update progress-
base_text = get_baseText(total_files=len(data['input_paths']),
file_num=file_num)
progress_kwargs = {'progress_var': progress_var,
'total_files': len(data['input_paths']),
'file_num': file_num}
update_progress(**progress_kwargs,
step=0)
#Load Model(s)
#text_widget.write(base_text + 'Loading models...')
for i, c in tqdm(enumerate(loops), disable=True, desc='Iterations..'):
text_widget.write(c['loop_name'] + '\n\n')
text_widget.write(base_text + 'Loading ' + c['model_name'] + '... ')
arch_now = c['using_archtecture']
if arch_now == '123821KB':
from lib_v5 import nets_123821KB as nets
elif arch_now == '537238KB':
from lib_v5 import nets_537238KB as nets
elif arch_now == '537227KB':
from lib_v5 import nets_537227KB as nets
def determineenseFolderName():
"""
Determine the name that is used for the folder and appended
to the back of the music files
"""
enseFolderName = ''
# -Instrumental-
if str(ensefolder):
enseFolderName += os.path.splitext(os.path.basename(ensefolder))[0]
if enseFolderName:
enseFolderName = '/' + enseFolderName
return enseFolderName
enseFolderName = determineenseFolderName()
if enseFolderName:
folder_path = f'{data["export_path"]}{enseFolderName}'
if not os.path.isdir(folder_path):
os.mkdir(folder_path)
# Determine File Name
base_name = f'{data["export_path"]}{enseFolderName}/{file_num}_{os.path.splitext(os.path.basename(music_file))[0]}'
enseExport = f'{data["export_path"]}{enseFolderName}/'
trackname = f'{file_num}_{os.path.splitext(os.path.basename(music_file))[0]}'
ModelName_1=(c['model_name'])
print('Model Parameters:', c['model_params'])
mp = ModelParameters(c['model_params'])
# -Instrumental-
if os.path.isfile(c['model_location']):
device = torch.device('cpu')
model = nets.CascadedASPPNet(mp.param['bins'] * 2)
model.load_state_dict(torch.load(c['model_location'],
map_location=device))
if torch.cuda.is_available() and data['gpu'] >= 0:
device = torch.device('cuda:{}'.format(data['gpu']))
model.to(device)
text_widget.write('Done!\n')
model_name = os.path.basename(c["model_name"])
# -Go through the different steps of seperation-
# Wave source
text_widget.write(base_text + 'Loading wave source... ')
X_wave, y_wave, X_spec_s, y_spec_s = {}, {}, {}, {}
bands_n = len(mp.param['band'])
for d in range(bands_n, 0, -1):
bp = mp.param['band'][d]
if d == bands_n: # high-end band
X_wave[d], _ = librosa.load(
music_file, bp['sr'], False, dtype=np.float32, res_type=bp['res_type'])
if X_wave[d].ndim == 1:
X_wave[d] = np.asarray([X_wave[d], X_wave[d]])
else: # lower bands
X_wave[d] = librosa.resample(X_wave[d+1], mp.param['band'][d+1]['sr'], bp['sr'], res_type=bp['res_type'])
# Stft of wave source
X_spec_s[d] = spec_utils.wave_to_spectrogram_mt(X_wave[d], bp['hl'], bp['n_fft'], mp.param['mid_side'],
mp.param['mid_side_b2'], mp.param['reverse'])
if d == bands_n and args.high_end_process != 'none':
input_high_end_h = (bp['n_fft']//2 - bp['crop_stop']) + (mp.param['pre_filter_stop'] - mp.param['pre_filter_start'])
input_high_end = X_spec_s[d][:, bp['n_fft']//2-input_high_end_h:bp['n_fft']//2, :]
text_widget.write('Done!\n')
update_progress(**progress_kwargs,
step=0.1)
text_widget.write(base_text + 'Stft of wave source... ')
text_widget.write('Done!\n')
text_widget.write(base_text + "Please Wait...\n")
X_spec_m = spec_utils.combine_spectrograms(X_spec_s, mp)
del X_wave, X_spec_s
def inference(X_spec, device, model, aggressiveness):
def _execute(X_mag_pad, roi_size, n_window, device, model, aggressiveness):
model.eval()
with torch.no_grad():
preds = []
iterations = [n_window]
total_iterations = sum(iterations)
text_widget.write(base_text + "Processing "f"{total_iterations} Slices... ")
for i in tqdm(range(n_window)):
update_progress(**progress_kwargs,
step=(0.1 + (0.8/n_window * i)))
start = i * roi_size
X_mag_window = X_mag_pad[None, :, :, start:start + data['window_size']]
X_mag_window = torch.from_numpy(X_mag_window).to(device)
pred = model.predict(X_mag_window, aggressiveness)
pred = pred.detach().cpu().numpy()
preds.append(pred[0])
pred = np.concatenate(preds, axis=2)
text_widget.write('Done!\n')
return pred
def preprocess(X_spec):
X_mag = np.abs(X_spec)
X_phase = np.angle(X_spec)
return X_mag, X_phase
X_mag, X_phase = preprocess(X_spec)
coef = X_mag.max()
X_mag_pre = X_mag / coef
n_frame = X_mag_pre.shape[2]
pad_l, pad_r, roi_size = dataset.make_padding(n_frame,
data['window_size'], model.offset)
n_window = int(np.ceil(n_frame / roi_size))
X_mag_pad = np.pad(
X_mag_pre, ((0, 0), (0, 0), (pad_l, pad_r)), mode='constant')
pred = _execute(X_mag_pad, roi_size, n_window,
device, model, aggressiveness)
pred = pred[:, :, :n_frame]
if data['tta']:
pad_l += roi_size // 2
pad_r += roi_size // 2
n_window += 1
X_mag_pad = np.pad(
X_mag_pre, ((0, 0), (0, 0), (pad_l, pad_r)), mode='constant')
pred_tta = _execute(X_mag_pad, roi_size, n_window,
device, model, aggressiveness)
pred_tta = pred_tta[:, :, roi_size // 2:]
pred_tta = pred_tta[:, :, :n_frame]
return (pred + pred_tta) * 0.5 * coef, X_mag, np.exp(1.j * X_phase)
else:
return pred * coef, X_mag, np.exp(1.j * X_phase)
aggressiveness = {'value': args.aggressiveness, 'split_bin': mp.param['band'][1]['crop_stop']}
if data['tta']:
text_widget.write(base_text + "Running Inferences (TTA)... \n")
else:
text_widget.write(base_text + "Running Inference... \n")
pred, X_mag, X_phase = inference(X_spec_m,
device,
model, aggressiveness)
update_progress(**progress_kwargs,
step=0.85)
# Postprocess
if data['postprocess']:
text_widget.write(base_text + 'Post processing... ')
pred_inv = np.clip(X_mag - pred, 0, np.inf)
pred = spec_utils.mask_silence(pred, pred_inv)
text_widget.write('Done!\n')
update_progress(**progress_kwargs,
step=0.85)
# Inverse stft
text_widget.write(base_text + 'Inverse stft of instruments and vocals... ') # nopep8
y_spec_m = pred * X_phase
v_spec_m = X_spec_m - y_spec_m
if args.high_end_process.startswith('mirroring'):
input_high_end_ = spec_utils.mirroring(args.high_end_process, y_spec_m, input_high_end, mp)
wav_instrument = spec_utils.cmb_spectrogram_to_wave(y_spec_m, mp, input_high_end_h, input_high_end_)
else:
wav_instrument = spec_utils.cmb_spectrogram_to_wave(y_spec_m, mp)
if args.high_end_process.startswith('mirroring'):
input_high_end_ = spec_utils.mirroring(args.high_end_process, v_spec_m, input_high_end, mp)
wav_vocals = spec_utils.cmb_spectrogram_to_wave(v_spec_m, mp, input_high_end_h, input_high_end_)
else:
wav_vocals = spec_utils.cmb_spectrogram_to_wave(v_spec_m, mp)
text_widget.write('Done!\n')
update_progress(**progress_kwargs,
step=0.9)
# Save output music files
text_widget.write(base_text + 'Saving Files... ')
save_files(wav_instrument, wav_vocals)
text_widget.write('Done!\n')
# Save output image
if data['output_image']:
with open('{}_Instruments.jpg'.format(base_name), mode='wb') as f:
image = spec_utils.spectrogram_to_image(y_spec_m)
_, bin_image = cv2.imencode('.jpg', image)
bin_image.tofile(f)
with open('{}_Vocals.jpg'.format(base_name), mode='wb') as f:
image = spec_utils.spectrogram_to_image(v_spec_m)
_, bin_image = cv2.imencode('.jpg', image)
bin_image.tofile(f)
text_widget.write(base_text + 'Completed Seperation!\n\n')
# Emsembling Outputs
def get_files(folder="", prefix="", suffix=""):
return [f"{folder}{i}" for i in os.listdir(folder) if i.startswith(prefix) if i.endswith(suffix)]
ensambles = [
{
'algorithm':'min_mag',
'model_params':'lib_v5/modelparams/1band_sr44100_hl512.json',
'files':get_files(folder=enseExport, prefix=trackname, suffix="_(Instrumental).wav"),
'output':'{}_Ensembled_{}_Instrumentals'.format(trackname, ensemode),
'type': 'Instrumentals'
},
{
'algorithm':'max_mag',
'model_params':'lib_v5/modelparams/1band_sr44100_hl512.json',
'files':get_files(folder=enseExport, prefix=trackname, suffix="_(Vocals).wav"),
'output': '{}_Ensembled_{}_Vocals'.format(trackname, ensemode),
'type': 'Vocals'
}
]
for i, e in tqdm(enumerate(ensambles), desc="Ensembling..."):
text_widget.write(base_text + "Ensembling " + e['type'] + "... ")
wave, specs = {}, {}
mp = ModelParameters(e['model_params'])
for i in range(len(e['files'])):
spec = {}
for d in range(len(mp.param['band']), 0, -1):
bp = mp.param['band'][d]
if d == len(mp.param['band']): # high-end band
wave[d], _ = librosa.load(
e['files'][i], bp['sr'], False, dtype=np.float32, res_type=bp['res_type'])
if len(wave[d].shape) == 1: # mono to stereo
wave[d] = np.array([wave[d], wave[d]])
else: # lower bands
wave[d] = librosa.resample(wave[d+1], mp.param['band'][d+1]['sr'], bp['sr'], res_type=bp['res_type'])
spec[d] = spec_utils.wave_to_spectrogram(wave[d], bp['hl'], bp['n_fft'], mp.param['mid_side'], mp.param['mid_side_b2'], mp.param['reverse'])
specs[i] = spec_utils.combine_spectrograms(spec, mp)
del wave
sf.write(os.path.join('{}'.format(data['export_path']),'{}.wav'.format(e['output'])),
spec_utils.cmb_spectrogram_to_wave(spec_utils.ensembling(e['algorithm'],
specs), mp), mp.param['sr'])
if not data['save']: # Deletes all outputs if Save All Outputs: is checked
files = e['files']
for file in files:
os.remove(file)
text_widget.write("Done!\n")
update_progress(**progress_kwargs,
step=0.95)
text_widget.write("\n")
except Exception as e:
traceback_text = ''.join(traceback.format_tb(e.__traceback__))
message = f'Traceback Error: "{traceback_text}"\n{type(e).__name__}: "{e}"\nFile: {music_file}\nPlease contact the creator and attach a screenshot of this error with the file and settings that caused it!'
tk.messagebox.showerror(master=window,
title='Untracked Error',
message=message)
print(traceback_text)
print(type(e).__name__, e)
print(message)
progress_var.set(0)
button_widget.configure(state=tk.NORMAL) # Enable Button
return
if len(os.listdir(enseExport)) == 0: # Check if the folder is empty
shutil.rmtree(folder_path)
update_progress(**progress_kwargs,
step=1)
print('Done!')
os.remove('temp.wav')
progress_var.set(0)
text_widget.write(f'Conversions Completed!\n')
text_widget.write(f'Time Elapsed: {time.strftime("%H:%M:%S", time.gmtime(int(time.perf_counter() - stime)))}') # nopep8
torch.cuda.empty_cache()
button_widget.configure(state=tk.NORMAL) # Enable Button