init

control/cli/synthesizer_preprocess_embeds.py

@@ -14,7 +14,7 @@ if __name__ == "__main__":
         "Path to the synthesizer training data that contains the audios and the train.txt file. "
         "If you let everything as default, it should be <datasets_root>/SV2TTS/synthesizer/.")
     parser.add_argument("-e", "--encoder_model_fpath", type=Path, 
-                        default="encoder/saved_models/pretrained.pt", help=\
+                        default="data/ckpt/encoder/pretrained.pt", help=\
         "Path your trained encoder model.")
     parser.add_argument("-n", "--n_processes", type=int, default=4, help= \
         "Number of parallel processes. An encoder is created for each, so you may need to lower "

control/cli/synthesizer_train.py

@@ -3,8 +3,7 @@ from models.synthesizer.train import train
 from utils.argutils import print_args
 import argparse
 
-
-if __name__ == "__main__":
+def new_train():
     parser = argparse.ArgumentParser()
     parser.add_argument("run_id", type=str, help= \
         "Name for this model instance. If a model state from the same run ID was previously "
@@ -13,7 +12,7 @@ if __name__ == "__main__":
     parser.add_argument("syn_dir", type=str, default=argparse.SUPPRESS, help= \
         "Path to the synthesizer directory that contains the ground truth mel spectrograms, "
         "the wavs and the embeds.")
-    parser.add_argument("-m", "--models_dir", type=str, default="synthesizer/saved_models/", help=\
+    parser.add_argument("-m", "--models_dir", type=str, default=f"data/ckpt/synthesizer/", help=\
         "Path to the output directory that will contain the saved model weights and the logs.")
     parser.add_argument("-s", "--save_every", type=int, default=1000, help= \
         "Number of steps between updates of the model on the disk. Set to 0 to never save the "
@@ -28,10 +27,14 @@ if __name__ == "__main__":
     parser.add_argument("--hparams", default="",
                         help="Hyperparameter overrides as a comma-separated list of name=value "
 							 "pairs")
-    args = parser.parse_args()
+    args, _ = parser.parse_known_args()
     print_args(args, parser)
 
     args.hparams = hparams.parse(args.hparams)
 
     # Run the training
     train(**vars(args))
+
+
+if __name__ == "__main__":
+    new_train()

control/cli/train_ppg2mel.py

@@ -0,0 +1,66 @@
+import sys
+import torch
+import argparse
+import numpy as np
+from utils.load_yaml import HpsYaml
+from models.ppg2mel.train.train_linglf02mel_seq2seq_oneshotvc import Solver
+
+# For reproducibility, comment these may speed up training
+torch.backends.cudnn.deterministic = True
+torch.backends.cudnn.benchmark = False
+
+def main():
+    # Arguments
+    parser = argparse.ArgumentParser(description=
+            'Training PPG2Mel VC model.')
+    parser.add_argument('--config', type=str, 
+                        help='Path to experiment config, e.g., config/vc.yaml')
+    parser.add_argument('--name', default=None, type=str, help='Name for logging.')
+    parser.add_argument('--logdir', default='log/', type=str,
+                        help='Logging path.', required=False)
+    parser.add_argument('--ckpdir', default='ppg2mel/saved_models/', type=str,
+                        help='Checkpoint path.', required=False)
+    parser.add_argument('--outdir', default='result/', type=str,
+                        help='Decode output path.', required=False)
+    parser.add_argument('--load', default=None, type=str,
+                        help='Load pre-trained model (for training only)', required=False)
+    parser.add_argument('--warm_start', action='store_true',
+                        help='Load model weights only, ignore specified layers.')
+    parser.add_argument('--seed', default=0, type=int,
+                        help='Random seed for reproducable results.', required=False)
+    parser.add_argument('--njobs', default=8, type=int,
+                        help='Number of threads for dataloader/decoding.', required=False)
+    parser.add_argument('--cpu', action='store_true', help='Disable GPU training.')
+    parser.add_argument('--no-pin', action='store_true',
+                        help='Disable pin-memory for dataloader')
+    parser.add_argument('--test', action='store_true', help='Test the model.')
+    parser.add_argument('--no-msg', action='store_true', help='Hide all messages.')
+    parser.add_argument('--finetune', action='store_true', help='Finetune model')
+    parser.add_argument('--oneshotvc', action='store_true', help='Oneshot VC model')
+    parser.add_argument('--bilstm', action='store_true', help='BiLSTM VC model')
+    parser.add_argument('--lsa', action='store_true', help='Use location-sensitive attention (LSA)')
+
+    ###
+    paras = parser.parse_args()
+    setattr(paras, 'gpu', not paras.cpu)
+    setattr(paras, 'pin_memory', not paras.no_pin)
+    setattr(paras, 'verbose', not paras.no_msg)
+    # Make the config dict dot visitable
+    config = HpsYaml(paras.config)
+
+    np.random.seed(paras.seed)
+    torch.manual_seed(paras.seed)
+    if torch.cuda.is_available():
+        torch.cuda.manual_seed_all(paras.seed)
+
+    print(">>> OneShot VC training ...")
+    mode = "train"
+    solver = Solver(config, paras, mode)
+    solver.load_data()
+    solver.set_model()
+    solver.exec()
+    print(">>> Oneshot VC train finished!")
+    sys.exit(0)
+
+if __name__ == "__main__":
+    main()