Update synthesis.py

TTS/tts/utils/synthesis.py

@@ -193,14 +193,15 @@ def synthesis(
     # convert outputs to numpy
     # plot results
     wav = None
-    if hasattr(model, "END2END") and model.END2END:
+    model_outputs = model_outputs.squeeze()
-        wav = model_outputs.squeeze(0)
+    if model_outputs.ndim == 2:  # [T, C_spec]
-    else:
         if use_griffin_lim:
             wav = inv_spectrogram(model_outputs, model.ap, CONFIG)
             # trim silence
             if do_trim_silence:
                 wav = trim_silence(wav, model.ap)
+    else:  # [T,]
+        wav = model_outputs
     return_dict = {
         "wav": wav,
         "alignments": alignments,

TTS/utils/training.py

@@ -42,31 +42,3 @@ def gradual_training_scheduler(global_step, config):
         if global_step * num_gpus >= values[0]:
             new_values = values
     return new_values[1], new_values[2]
-
-
-def lr_decay(init_lr, global_step, warmup_steps):
-    r"""from https://github.com/r9y9/tacotron_pytorch/blob/master/train.py
-    It is only being used by the Speaker Encoder trainer."""
-    warmup_steps = float(warmup_steps)
-    step = global_step + 1.0
-    lr = init_lr * warmup_steps**0.5 * np.minimum(step * warmup_steps**-1.5, step**-0.5)
-    return lr
-
-
-# pylint: disable=dangerous-default-value
-def set_weight_decay(model, weight_decay, skip_list={"decoder.attention.v", "rnn", "lstm", "gru", "embedding"}):
-    """
-    Skip biases, BatchNorm parameters, rnns.
-    and attention projection layer v
-    """
-    decay = []
-    no_decay = []
-    for name, param in model.named_parameters():
-        if not param.requires_grad:
-            continue
-
-        if len(param.shape) == 1 or any((skip_name in name for skip_name in skip_list)):
-            no_decay.append(param)
-        else:
-            decay.append(param)
-    return [{"params": no_decay, "weight_decay": 0.0}, {"params": decay, "weight_decay": weight_decay}]