Approximate audio length from file size

TTS/tts/configs/tacotron_config.py

@@ -106,7 +106,7 @@ class TacotronConfig(BaseTTSConfig):
             Weight decay coefficient. Defaults to `1e-6`.
         grad_clip (float):
             Gradient clipping threshold. Defaults to `5`.
-        seq_len_notm (bool):
+        seq_len_norm (bool):
             enable / disable the sequnce length normalization in the loss functions. If set True, loss of a sample
             is divided by the sequence length. Defaults to False.
         loss_masking (bool):

TTS/tts/datasets/dataset.py

@@ -330,7 +330,7 @@ class TTSDataset(Dataset):
         if by_audio_len:
             lengths = []
             for item in self.items:
-                lengths.append(os.path.getsize(item[1]))
+                lengths.append(os.path.getsize(item[1]) / 16 * 8)  # assuming 16bit audio
             lengths = np.array(lengths)
         else:
             lengths = np.array([len(ins[0]) for ins in self.items])

TTS/tts/models/base_tacotron.py

@@ -242,4 +242,4 @@ class BaseTacotron(BaseTTS):
             self.decoder.set_r(r)
             if trainer.config.bidirectional_decoder:
                 trainer.model.decoder_backward.set_r(r)
-            print(f"\n > Number of output frames: {self.decoder.r}")
+            print(f"\n > Number of output frames: {self.decoder.r}")

TTS/tts/models/base_tts.py

@@ -20,9 +20,9 @@ from TTS.utils.audio import AudioProcessor
 
 
 class BaseTTS(BaseModel):
-    """Abstract `tts` class. Every new `tts` model must inherit this.
+    """Base `tts` class. Every new `tts` model must inherit this.
 
-    It defines `tts` specific functions on top of `Model`.
+    It defines common `tts` specific functions on top of `Model` implementation.
 
     Notes on input/output tensor shapes:
         Any input or output tensor of the model must be shaped as