Cool_Tools/ultimatevocalremovergui
1
0
Fork 0
mirror of https://github.com/Anjok07/ultimatevocalremovergui.git synced 2025-03-03 08:36:33 +01:00
Code Issues Packages Projects Releases Wiki Activity

Add files via upload

Browse Source
This commit is contained in:
Anjok07 2023-10-10 21:48:02 -05:00 committed by GitHub
parent 3630c32124
commit 274ca43b3a
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 39 additions and 23 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
UVR.py
View File

@ -53,6 +53,9 @@ import sys
import yaml import yaml
from ml_collections import ConfigDict from ml_collections import ConfigDict
from collections import Counter from collections import Counter
import torch_directml
is_gpu_available = torch.backends.mps.is_available() if (OPERATING_SYSTEM == 'Darwin') else (torch_directml.is_available() or torch.cuda.is_available())
# Change the current working directory to the directory # Change the current working directory to the directory
# this file sits in # this file sits in
@ -109,6 +112,8 @@ elif OPERATING_SYSTEM=="Windows":
right_click_button = '<Button-3>' right_click_button = '<Button-3>'
application_extension = ".exe" application_extension = ".exe"
clear_gpu_cache = torch.mps.empty_cache if is_macos else torch.cuda.empty_cache
def right_click_release_linux(window, top_win=None): def right_click_release_linux(window, top_win=None):
if OPERATING_SYSTEM=="Linux": if OPERATING_SYSTEM=="Linux":
root.bind('<Button-1>', lambda e:window.destroy()) root.bind('<Button-1>', lambda e:window.destroy())
@ -1437,7 +1442,7 @@ class MainWindow(TkinterDnD.Tk if is_dnd_compatible else tk.Tk):
self.mdx_hash_MAPPER = load_model_hash_data(MDX_HASH_JSON) self.mdx_hash_MAPPER = load_model_hash_data(MDX_HASH_JSON)
self.mdx_name_select_MAPPER = load_model_hash_data(MDX_MODEL_NAME_SELECT) self.mdx_name_select_MAPPER = load_model_hash_data(MDX_MODEL_NAME_SELECT)
self.demucs_name_select_MAPPER = load_model_hash_data(DEMUCS_MODEL_NAME_SELECT) self.demucs_name_select_MAPPER = load_model_hash_data(DEMUCS_MODEL_NAME_SELECT)
self.is_gpu_available = torch.cuda.is_available() if not OPERATING_SYSTEM == 'Darwin' else torch.backends.mps.is_available() self.is_gpu_available = is_gpu_available
self.is_process_stopped = False self.is_process_stopped = False
self.inputs_from_dir = [] self.inputs_from_dir = []
self.iteration = 0 self.iteration = 0
@ -5479,7 +5484,7 @@ class MainWindow(TkinterDnD.Tk if is_dnd_compatible else tk.Tk):
if self.clear_cache_torch: if self.clear_cache_torch:
#self.set_app_font(is_chosen_font=True) #self.set_app_font(is_chosen_font=True)
torch.cuda.empty_cache() clear_gpu_cache()
self.clear_cache_torch = False self.clear_cache_torch = False
if self.is_process_stopped: if self.is_process_stopped:
@ -5492,7 +5497,7 @@ class MainWindow(TkinterDnD.Tk if is_dnd_compatible else tk.Tk):
self.conversion_Button_Text_var.set(START_PROCESSING) self.conversion_Button_Text_var.set(START_PROCESSING)
self.conversion_Button.configure(state=tk.NORMAL) self.conversion_Button.configure(state=tk.NORMAL)
self.progress_bar_main_var.set(0) self.progress_bar_main_var.set(0)
torch.cuda.empty_cache() clear_gpu_cache()
self.is_process_stopped = False self.is_process_stopped = False
if self.is_confirm_error_var.get(): if self.is_confirm_error_var.get():
@ -6589,7 +6594,7 @@ class MainWindow(TkinterDnD.Tk if is_dnd_compatible else tk.Tk):
if os.path.isfile(audio_file): if os.path.isfile(audio_file):
os.remove(audio_file) os.remove(audio_file)
torch.cuda.empty_cache() clear_gpu_cache()
shutil.rmtree(export_path) if is_ensemble and len(os.listdir(export_path)) == 0 else None shutil.rmtree(export_path) if is_ensemble and len(os.listdir(export_path)) == 0 else None

39
separate.py
View File

@ -31,6 +31,10 @@ import math
#import random #import random
from onnx import load from onnx import load
from onnx2pytorch import ConvertModel from onnx2pytorch import ConvertModel
import torch_directml
import gc
cpu = 'cpu'
if TYPE_CHECKING: if TYPE_CHECKING:
from UVR import ModelData from UVR import ModelData
@ -38,6 +42,8 @@ if TYPE_CHECKING:
warnings.filterwarnings("ignore") warnings.filterwarnings("ignore")
cpu = torch.device('cpu') cpu = torch.device('cpu')
clear_gpu_cache = torch.mps.empty_cache if is_macos else torch.cuda.empty_cache
class SeperateAttributes: class SeperateAttributes:
def __init__(self, model_data: ModelData, def __init__(self, model_data: ModelData,
process_data: dict, process_data: dict,
@ -137,8 +143,9 @@ class SeperateAttributes:
self.stem_path_init = os.path.join(self.export_path, f'{self.audio_file_base}_({self.secondary_stem}).wav') self.stem_path_init = os.path.join(self.export_path, f'{self.audio_file_base}_({self.secondary_stem}).wav')
self.deverb_vocal_opt = model_data.deverb_vocal_opt self.deverb_vocal_opt = model_data.deverb_vocal_opt
self.is_save_vocal_only = model_data.is_save_vocal_only self.is_save_vocal_only = model_data.is_save_vocal_only
self.device = 'cpu' self.device = cpu
self.run_type = ['CPUExecutionProvider'] self.run_type = ['CPUExecutionProvider']
self.is_opencl = False
if self.is_inst_only_voc_splitter or self.is_sec_bv_rebalance: if self.is_inst_only_voc_splitter or self.is_sec_bv_rebalance:
self.is_primary_stem_only = False self.is_primary_stem_only = False
@ -149,11 +156,11 @@ class SeperateAttributes:
if self.is_gpu_conversion >= 0: if self.is_gpu_conversion >= 0:
if OPERATING_SYSTEM == 'Darwin' and torch.backends.mps.is_available(): if OPERATING_SYSTEM == 'Darwin' and torch.backends.mps.is_available():
self.device = 'mps' self.device, self.is_mps = 'mps', True
self.is_mps = True elif torch_directml.is_available():
self.device, self.is_mps, self.is_opencl = torch_directml.device(), True, True
elif torch.cuda.is_available(): elif torch.cuda.is_available():
self.device = 'cuda:0' self.device, self.run_type = 'cuda:0', ['CUDAExecutionProvider']
self.run_type = ['CUDAExecutionProvider']
if model_data.process_method == MDX_ARCH_TYPE: if model_data.process_method == MDX_ARCH_TYPE:
self.is_mdx_ckpt = model_data.is_mdx_ckpt self.is_mdx_ckpt = model_data.is_mdx_ckpt
@ -166,6 +173,7 @@ class SeperateAttributes:
self.mdx_segment_size = model_data.mdx_segment_size self.mdx_segment_size = model_data.mdx_segment_size
if self.is_mdx_c: if self.is_mdx_c:
self.device = cpu if self.is_opencl else self.device
if not self.is_4_stem_ensemble: if not self.is_4_stem_ensemble:
self.primary_stem = model_data.ensemble_primary_stem if process_data['is_ensemble_master'] else model_data.primary_stem self.primary_stem = model_data.ensemble_primary_stem if process_data['is_ensemble_master'] else model_data.primary_stem
self.secondary_stem = model_data.ensemble_secondary_stem if process_data['is_ensemble_master'] else model_data.secondary_stem self.secondary_stem = model_data.ensemble_secondary_stem if process_data['is_ensemble_master'] else model_data.secondary_stem
@ -192,7 +200,7 @@ class SeperateAttributes:
self.is_demucs_combine_stems = model_data.is_demucs_combine_stems self.is_demucs_combine_stems = model_data.is_demucs_combine_stems
self.demucs_stem_count = model_data.demucs_stem_count self.demucs_stem_count = model_data.demucs_stem_count
self.pre_proc_model = model_data.pre_proc_model self.pre_proc_model = model_data.pre_proc_model
self.device = 'cpu' if self.is_mps and not self.demucs_version == DEMUCS_V4 else self.device self.device = cpu if self.is_mps and not self.demucs_version == DEMUCS_V4 else self.device
self.primary_stem = model_data.ensemble_primary_stem if process_data['is_ensemble_master'] else model_data.primary_stem self.primary_stem = model_data.ensemble_primary_stem if process_data['is_ensemble_master'] else model_data.primary_stem
self.secondary_stem = model_data.ensemble_secondary_stem if process_data['is_ensemble_master'] else model_data.secondary_stem self.secondary_stem = model_data.ensemble_secondary_stem if process_data['is_ensemble_master'] else model_data.secondary_stem
@ -458,6 +466,7 @@ class SeperateMDX(SeperateAttributes):
self.running_inference_console_write() self.running_inference_console_write()
mix = prepare_mix(self.audio_file) mix = prepare_mix(self.audio_file)
source = self.demix(mix) source = self.demix(mix)
if not self.is_vocal_split_model: if not self.is_vocal_split_model:
@ -485,7 +494,7 @@ class SeperateMDX(SeperateAttributes):
self.primary_source_map = self.final_process(primary_stem_path, self.primary_source, self.secondary_source_primary, self.primary_stem, samplerate) self.primary_source_map = self.final_process(primary_stem_path, self.primary_source, self.secondary_source_primary, self.primary_stem, samplerate)
torch.cuda.empty_cache() clear_gpu_cache()
secondary_sources = {**self.primary_source_map, **self.secondary_source_map} secondary_sources = {**self.primary_source_map, **self.secondary_source_map}
@ -499,7 +508,7 @@ class SeperateMDX(SeperateAttributes):
self.trim = self.n_fft//2 self.trim = self.n_fft//2
self.chunk_size = self.hop * (self.mdx_segment_size-1) self.chunk_size = self.hop * (self.mdx_segment_size-1)
self.gen_size = self.chunk_size-2*self.trim self.gen_size = self.chunk_size-2*self.trim
self.stft = STFT(self.n_fft, self.hop, self.dim_f) self.stft = STFT(self.n_fft, self.hop, self.dim_f, self.device)
def demix(self, mix, is_match_mix=False): def demix(self, mix, is_match_mix=False):
self.initialize_model_settings() self.initialize_model_settings()
@ -685,7 +694,7 @@ class SeperateMDXC(SeperateAttributes):
self.primary_source_map = self.final_process(primary_stem_path, self.primary_source, self.secondary_source_primary, self.primary_stem, samplerate) self.primary_source_map = self.final_process(primary_stem_path, self.primary_source, self.secondary_source_primary, self.primary_stem, samplerate)
torch.cuda.empty_cache() clear_gpu_cache()
secondary_sources = {**self.primary_source_map, **self.secondary_source_map} secondary_sources = {**self.primary_source_map, **self.secondary_source_map}
self.process_vocal_split_chain(secondary_sources) self.process_vocal_split_chain(secondary_sources)
@ -699,8 +708,8 @@ class SeperateMDXC(SeperateAttributes):
if self.is_pitch_change: if self.is_pitch_change:
mix, sr_pitched = spec_utils.change_pitch_semitones(mix, 44100, semitone_shift=-self.semitone_shift) mix, sr_pitched = spec_utils.change_pitch_semitones(mix, 44100, semitone_shift=-self.semitone_shift)
model = TFC_TDF_net(self.mdx_c_configs).eval().to(self.device) model = TFC_TDF_net(self.mdx_c_configs, device=self.device).eval().to(self.device)
model.load_state_dict(torch.load(self.model_path, map_location=self.device)) model.load_state_dict(torch.load(self.model_path, map_location=cpu))
mix = torch.tensor(mix, dtype=torch.float32) mix = torch.tensor(mix, dtype=torch.float32)
try: try:
@ -830,7 +839,7 @@ class SeperateDemucs(SeperateAttributes):
self.write_to_console(DONE, base_text='') self.write_to_console(DONE, base_text='')
del self.demucs del self.demucs
torch.cuda.empty_cache() clear_gpu_cache()
if isinstance(inst_source, np.ndarray): if isinstance(inst_source, np.ndarray):
source_reshape = spec_utils.reshape_sources(inst_source[self.demucs_source_map[VOCAL_STEM]], source[self.demucs_source_map[VOCAL_STEM]]) source_reshape = spec_utils.reshape_sources(inst_source[self.demucs_source_map[VOCAL_STEM]], source[self.demucs_source_map[VOCAL_STEM]])
@ -1050,7 +1059,7 @@ class SeperateVR(SeperateAttributes):
self.secondary_source_map = self.final_process(secondary_stem_path, self.secondary_source, self.secondary_source_secondary, self.secondary_stem, 44100) self.secondary_source_map = self.final_process(secondary_stem_path, self.secondary_source, self.secondary_source_secondary, self.secondary_stem, 44100)
torch.cuda.empty_cache() clear_gpu_cache()
secondary_sources = {**self.primary_source_map, **self.secondary_source_map} secondary_sources = {**self.primary_source_map, **self.secondary_source_map}
self.process_vocal_split_chain(secondary_sources) self.process_vocal_split_chain(secondary_sources)
@ -1330,8 +1339,10 @@ def vr_denoiser(X, device, hop_length=1024, n_fft=2048, cropsize=256, is_deverbe
hop_length=1024 hop_length=1024
nout, nout_lstm = 16, 128 nout, nout_lstm = 16, 128
device = cpu if not device in ['cuda', 'cpu'] else device
model = nets_new.CascadedNet(n_fft, nout=nout, nout_lstm=nout_lstm) model = nets_new.CascadedNet(n_fft, nout=nout, nout_lstm=nout_lstm)
model.load_state_dict(torch.load(model_path, map_location=device)) model.load_state_dict(torch.load(model_path, map_location=cpu))
model.to(device) model.to(device)
if mp is None: if mp is None: