Merge branch 'main' of https://github.com/DanRuta/xva-trainer into main

python/audio_norm/model.py

@@ -1,6 +1,7 @@
 import os
 import json
 import traceback
+import platform
 
 # Not a model, but it was easier to just integrate the code this way
 
@@ -62,6 +63,8 @@ def __init__(self, logger, PROD, device, models_manager):
         self.model = None
         self.isReady = True
         self.ffmpeg_path = f'{"./resources/app" if self.PROD else "."}/python/ffmpeg.exe'
+        if platform.system() != 'Windows':
+            self.ffmpeg_path = 'ffmpeg'
 
 
     def load_state_dict (self, ckpt_path, sd):

python/pyinstaller_imports.py

@@ -15,13 +15,19 @@
     import scipy.optimize
     # import scipy.optimize.line_search
     import sklearn.utils._cython_blas
-    import sklearn.neighbors.typedefs
-    import sklearn.neighbors.quad_tree
+    try:
+        import sklearn.neighbors.typedefs
+        import sklearn.neighbors.quad_tree
+    except:
+        import sklearn.neighbors
     import sklearn.tree
     import sklearn.tree._utils
     import sklearn.utils._cython_blas
-    import sklearn.neighbors.typedefs
-    import sklearn.neighbors.quad_tree
+    try:
+        import sklearn.neighbors.typedefs
+        import sklearn.neighbors.quad_tree
+    except:
+        import sklearn.neighbors
     import sklearn.tree._utils
 except:
     print("==== scipy")

python/xvapitch/dataset.py

@@ -174,7 +174,7 @@ def __init__(
         # print("Loading text processors...")
         self.tp = {}
         for lang_code in lang_names.keys():
-            base_dir = "/".join(os.path.abspath(__file__).split("\\")[:-1])+f'/text'
+            base_dir = os.path.join(os.path.dirname(os.path.abspath(__file__)), 'text')
             self.tp[lang_code] = get_text_preprocessor(lang_code, base_dir, override_useAnyG2P=False)
 
 
@@ -983,7 +983,7 @@ def pre_cache_g2p (dataset_roots, lang=None):
 
     # tp = get_text_preprocessor(lang_code, base_dir)
     tp = {}
-    base_dir = "/".join(os.path.abspath(__file__).split("\\")[:-1])+f'/text'
+    base_dir = os.path.join(os.path.dirname(os.path.abspath(__file__)), 'text')
     for lang_code in languages:
         tp[lang_code] = get_text_preprocessor(lang_code, base_dir, override_useAnyG2P=True)
 

python/xvapitch/xva_train.py

@@ -1144,7 +1144,9 @@ def init_model(self, args, device, gpus):
         model = xVAPitch(args)
         model = model.to(device)
         try:
-            model = torch.compile(model)
+            print("No torch.compile()") # Waiting for someone to verify this works in linux first
+            # model = torch.compile(model)
+            pass
         except:
             print("No torch.compile()") # Waiting for windows support
             pass

server.py

@@ -6,6 +6,11 @@
 import wave
 import contextlib
 import numpy as np
+import platform
+
+if platform.system() == 'Linux':
+    import pyximport
+    pyximport.install(setup_args={"script_args" : ["--verbose"]})
 
 if __name__ == '__main__':
     multiprocessing.freeze_support()