update processors for loading attention maps

TTS/tts/datasets/preprocess.py

@@ -8,6 +8,9 @@ from tqdm import tqdm
 
 from TTS.tts.utils.generic_utils import split_dataset
 
+####################
+# UTILITIES
+####################
 
 def load_meta_data(datasets, eval_split=True):
     meta_data_train_all = []
@@ -17,9 +20,12 @@ def load_meta_data(datasets, eval_split=True):
         root_path = dataset['path']
         meta_file_train = dataset['meta_file_train']
         meta_file_val = dataset['meta_file_val']
+        # setup the right data processor
         preprocessor = get_preprocessor_by_name(name)
+        # load train set
         meta_data_train = preprocessor(root_path, meta_file_train)
         print(f" | > Found {len(meta_data_train)} files in {Path(root_path).resolve()}")
+        # load evaluation split if set
         if eval_split:
             if meta_file_val is None:
                 meta_data_eval, meta_data_train = split_dataset(meta_data_train)
@@ -27,15 +33,41 @@ def load_meta_data(datasets, eval_split=True):
                 meta_data_eval = preprocessor(root_path, meta_file_val)
             meta_data_eval_all += meta_data_eval
         meta_data_train_all += meta_data_train
+        # load attention masks for duration predictor training
+        if 'meta_file_attn_mask' in dataset:
+            meta_data = dict(load_attention_mask_meta_data(dataset['meta_file_attn_mask']))
+            for idx, ins in enumerate(meta_data_train_all):
+                attn_file = meta_data[ins[1]].strip()
+                meta_data_train_all[idx].append(attn_file)
+            if meta_data_eval_all is not None:
+                for idx, ins in enumerate(meta_data_eval_all):
+                    attn_file = meta_data[ins[1]].strip()
+                    meta_data_eval_all[idx].append(attn_file)
     return meta_data_train_all, meta_data_eval_all
 
 
+def load_attention_mask_meta_data(metafile_path):
+    """Load meta data file created by compute_attention_masks.py"""
+    with open(metafile_path, 'r') as f:
+        lines = f.readlines()
+
+    meta_data = []
+    for line in lines:
+        wav_file, attn_file = line.split('|')
+        meta_data.append([wav_file, attn_file])
+    return meta_data
+
+
 def get_preprocessor_by_name(name):
     """Returns the respective preprocessing function."""
     thismodule = sys.modules[__name__]
     return getattr(thismodule, name.lower())
 
 
+########################
+# DATASETS
+########################
+
 def tweb(root_path, meta_file):
     """Normalize TWEB dataset.
     https://www.kaggle.com/bryanpark/the-world-english-bible-speech-dataset
@@ -52,19 +84,6 @@ def tweb(root_path, meta_file):
     return items
 
 
-# def kusal(root_path, meta_file):
-#     txt_file = os.path.join(root_path, meta_file)
-#     texts = []
-#     wavs = []
-#     with open(txt_file, "r", encoding="utf8") as f:
-#         frames = [
-#             line.split('\t') for line in f
-#             if line.split('\t')[0] in self.wav_files_dict.keys()
-#         ]
-#     # TODO: code the rest
-#     return  {'text': texts, 'wavs': wavs}
-
-
 def mozilla(root_path, meta_file):
     """Normalizes Mozilla meta data files to TTS format"""
     txt_file = os.path.join(root_path, meta_file)