Remove tensorflow1.x dependency

.gitignore

@@ -0,0 +1,4 @@
+
+*.pyc
+*.pt
+waveglow

config/default.yaml

@@ -0,0 +1,74 @@
+epochs: 500
+iters_per_checkpoint: 1000
+seed: 1234
+dynamic_loss_scaling: True
+fp16_run: False
+distributed_run: False
+dist_backend: "nccl"
+dist_url: "tcp://localhost:54321"
+cudnn_enabled: True
+cudnn_benchmark: False
+ignore_layers: ['embedding.weight']
+
+################################
+# Data Parameters             #
+################################
+load_mel_from_disk: False
+training_files: 'filelists/ljs_audio_text_train_filelist.txt'
+validation_files: 'filelists/ljs_audio_text_val_filelist.txt'
+text_cleaners: ['english_cleaners']
+
+################################
+# Audio Parameters             #
+################################
+max_wav_value: 32768.0
+sampling_rate: 22050
+filter_length: 1024
+hop_length: 256
+win_length: 1024
+n_mel_channels: 80
+mel_fmin: 0.0
+mel_fmax: 8000.0
+
+################################
+# Model Parameters             #
+################################
+n_symbols: 148
+symbols_embedding_dim: 512
+
+# Encoder parameters
+encoder_kernel_size: 5
+encoder_n_convolutions: 3
+encoder_embedding_dim: 512
+
+# Decoder parameters
+n_frames_per_step: 1  # currently only 1 is supported
+decoder_rnn_dim: 1024
+prenet_dim: 256
+max_decoder_steps: 1000
+gate_threshold: 0.5
+p_attention_dropout: 0.1
+p_decoder_dropout: 0.1
+
+# Attention parameters
+attention_rnn_dim: 1024
+attention_dim: 128
+
+# Location Layer parameters
+attention_location_n_filters: 32
+attention_location_kernel_size: 31
+
+# Mel-post processing network parameters
+postnet_embedding_dim: 512
+postnet_kernel_size: 5
+postnet_n_convolutions: 5
+
+################################
+# Optimization Hyperparameters #
+################################
+use_saved_learning_rate: False
+learning_rate: 1e-3
+weight_decay: 1e-6
+grad_clip_thresh: 1.0
+batch_size: 64
+mask_padding: True  # set model's padded outputs to padded values

config/hparams.py

@@ -0,0 +1,65 @@
+import os
+import yaml
+
+
+def load_hparam_str(hp_str):
+    path = "temp-restore.yaml"
+    with open(path, "w") as f:
+        f.write(hp_str)
+    ret = HParam(path)
+    os.remove(path)
+    return ret
+
+
+def load_hparam(filename):
+    stream = open(filename, "r")
+    docs = yaml.load_all(stream, Loader=yaml.Loader)
+    hparam_dict = dict()
+    for doc in docs:
+        for k, v in doc.items():
+            hparam_dict[k] = v
+    return hparam_dict
+
+
+def merge_dict(user, default):
+    if isinstance(user, dict) and isinstance(default, dict):
+        for k, v in default.items():
+            if k not in user:
+                user[k] = v
+            else:
+                user[k] = merge_dict(user[k], v)
+    return user
+
+
+class Dotdict(dict):
+    """
+    a dictionary that supports dot notation
+    as well as dictionary access notation
+    usage: d = DotDict() or d = DotDict({'val1':'first'})
+    set attributes: d.val2 = 'second' or d['val2'] = 'second'
+    get attributes: d.val2 or d['val2']
+    """
+
+    __getattr__ = dict.__getitem__
+    __setattr__ = dict.__setitem__
+    __delattr__ = dict.__delitem__
+
+    def __init__(self, dct=None):
+        dct = dict() if not dct else dct
+        for key, value in dct.items():
+            if hasattr(value, "keys"):
+                value = Dotdict(value)
+            self[key] = value
+
+
+class HParam(Dotdict):
+    def __init__(self, file):
+        super(Dotdict, self).__init__()
+        hp_dict = load_hparam(file)
+        hp_dotdict = Dotdict(hp_dict)
+        for k, v in hp_dotdict.items():
+            setattr(self, k, v)
+
+    __getattr__ = Dotdict.__getitem__
+    __setattr__ = Dotdict.__setitem__
+    __delattr__ = Dotdict.__delitem__

requirements.txt

@@ -1,5 +1,4 @@
 matplotlib==2.1.0
-tensorflow==1.15.2
 numpy==1.13.3
 inflect==0.2.5
 librosa==0.6.0

train.py

@@ -14,7 +14,7 @@
 from data_utils import TextMelLoader, TextMelCollate
 from loss_function import Tacotron2Loss
 from logger import Tacotron2Logger
-from hparams import create_hparams
+from config.hparams import HParam
 
 
 def reduce_tensor(tensor, n_gpus):
@@ -272,10 +272,12 @@ def train(output_directory, log_directory, checkpoint_path, warm_start, n_gpus,
     parser.add_argument('--group_name', type=str, default='group_name',
                         required=False, help='Distributed group name')
     parser.add_argument('--hparams', type=str,
-                        required=False, help='comma separated name=value pairs')
+                        required=True, help='path to the yaml config file ')
 
     args = parser.parse_args()
-    hparams = create_hparams(args.hparams)
+    hparams = HParam(args.hparams)
+    with open(args.hparams, "r") as f:
+        hp_str = "".join(f.readlines())
 
     torch.backends.cudnn.enabled = hparams.cudnn_enabled
     torch.backends.cudnn.benchmark = hparams.cudnn_benchmark