[BC breaking] fix issue with VCTK dataset (#484)

* fix issue with VCTK dataset

* update docstring

* filter out folder p315

* add hidden except_folder has hidden variable

* maintain structure

* lint

* remove space
Co-authored-by: Vincent QB <vincentqb@users.noreply.github.com>

torchaudio/datasets/utils.py

@@ -244,9 +244,9 @@ def walk_files(root, suffix, prefix=False, remove_suffix=False):
         suffix (str or tuple): Suffix of the files to match, e.g. '.png' or ('.jpg', '.png').
             It uses the Python "str.endswith" method and is passed directly
         prefix (bool, optional): If true, prepends the full path to each result, otherwise
-            only returns the name of the files found
+            only returns the name of the files found (Default: ``False``)
         remove_suffix (bool, optional): If true, removes the suffix to each result defined in suffix,
-            otherwise will return the result as found.
+            otherwise will return the result as found (Default: ``False``).
     """
 
     root = os.path.expanduser(root)

torchaudio/datasets/vctk.py

@@ -39,12 +39,16 @@ class VCTK(Dataset):
     """
     Create a Dataset for VCTK. Each item is a tuple of the form:
     (waveform, sample_rate, utterance, speaker_id, utterance_id)
+
+    Folder `p315` will be ignored due to the non-existent corresponding text files.
+    For more information about the dataset visit: https://datashare.is.ed.ac.uk/handle/10283/3443
     """
 
     _folder_txt = "txt"
     _folder_audio = "wav48"
     _ext_txt = ".txt"
     _ext_audio = ".wav"
+    _except_folder = "p315"
 
     def __init__(
         self,
@@ -93,6 +97,7 @@ class VCTK(Dataset):
         walker = walk_files(
             self._path, suffix=self._ext_audio, prefix=False, remove_suffix=True
         )
+        walker = filter(lambda w: self._except_folder not in w, walker)
         self._walker = list(walker)
 
     def __getitem__(self, n):