Merge branch 'gtzan-downloader'

audiomate/corpus/io/__init__.py

@@ -6,7 +6,7 @@
 from .base import CorpusDownloader, CorpusReader, CorpusWriter
 from .broadcast import BroadcastReader  # noqa: F401
 from .default import DefaultReader, DefaultWriter  # noqa: F401
-from .gtzan import GtzanReader  # noqa: F401
+from .gtzan import GtzanDownloader, GtzanReader  # noqa: F401
 from .kaldi import KaldiReader, KaldiWriter  # noqa: F401
 from .musan import MusanDownloader, MusanReader  # noqa: F401
 from .speech_commands import SpeechCommandsReader  # noqa: F401

audiomate/corpus/io/gtzan.py

@@ -2,11 +2,46 @@
 
 import audiomate
 from audiomate.corpus import assets
+from audiomate.utils import download
+from audiomate.utils import files
 from . import base
 
+DOWNLOAD_URL = 'http://opihi.cs.uvic.ca/sound/music_speech.tar.gz'
 DIRECTORIES = {'music_wav': 'music', 'speech_wav': 'speech'}
 
 
+class GtzanDownloader(base.CorpusDownloader):
+    """
+    Downloader for the GTZAN Corpus.
+
+    Args:
+        url (str): The url to download the dataset from. If not given the default URL is used.
+                   It is expected to be a tar.gz file.
+    """
+
+    def __init__(self, url=None):
+        if url is None:
+            self.url = DOWNLOAD_URL
+        else:
+            self.url = url
+
+    @classmethod
+    def type(cls):
+        return 'gtzan'
+
+    def _download(self, target_path):
+        os.makedirs(target_path, exist_ok=True)
+        tmp_file = os.path.join(target_path, 'tmp_ark.tar.gz')
+
+        download.download_file(self.url, tmp_file)
+        download.extract_tar(tmp_file, target_path)
+
+        # We use copy since subfolders in the archive are read-only, hence throws permission error when trying to move.
+        files.move_all_files_from_subfolders_to_top(target_path, delete_subfolders=True, copy=True)
+
+        os.remove(tmp_file)
+
+
 class GtzanReader(base.CorpusReader):
     """
     Reader for the GTZAN music/speech corpus. The corpus consits of 64 music and 64 speech tracks that are each 30

audiomate/utils/files.py

@@ -2,13 +2,14 @@
 import shutil
 
 
-def move_all_files_from_subfolders_to_top(folder_path, delete_subfolders=False):
+def move_all_files_from_subfolders_to_top(folder_path, delete_subfolders=False, copy=False):
     """
     Move all files/folder from all subfolders of `folder_path` on top into `folder_path`.
 
     Args:
         folder_path (str): Path of the folder.
         delete_subfolders (bool): If True the subfolders are deleted after all items are moved out of it.
+        copy (bool): If True copies the files instead of moving. (default False)
     """
     for item in os.listdir(folder_path):
         sub_path = os.path.join(folder_path, item)
@@ -18,7 +19,14 @@ def move_all_files_from_subfolders_to_top(folder_path, delete_subfolders=False):
             for sub_item in os.listdir(sub_path):
                 src = os.path.join(sub_path, sub_item)
                 target = os.path.join(folder_path, sub_item)
-                shutil.move(src, target)
+
+                if copy:
+                    if os.path.isfile(src):
+                        shutil.copy(src, target)
+                    else:
+                        shutil.copytree(src, target)
+                else:
+                    shutil.move(src, target)
 
             if delete_subfolders:
                 shutil.rmtree(sub_path)

docs/notes/changelog.rst

@@ -16,7 +16,10 @@ Next Version
   But it only works for the prepared corpus.
 
 * Added function (:func:`audiomate.corpus.utils.label_cleaning.merge_consecutive_labels_with_same_values`)
-  for merging consecutive labels with the same value.
+  for merging consecutive labels with the same value
+
+* Added downloader (:class:`audiomate.corpus.io.GtzanDownloader`) for the
+  `GTZAN Music/Speech <https://marsyasweb.appspot.com/download/data_sets/>`_.
 
 **Fixes**
 

docs/reference/io.rst

@@ -44,7 +44,7 @@ Implementations
   Free-Spoken-Digit-Dataset       x         x
   Folder                                    x
   Google Speech Commands                    x
-  GTZAN                                     x
+  GTZAN                           x         x
   Kaldi                                     x      x
   Mozilla DeepSpeech                               x
   MUSAN                           x         x
@@ -109,6 +109,9 @@ Google Speech Commands
 
 GTZAN
 ^^^^^
+.. autoclass:: GtzanDownloader
+   :members:
+
 .. autoclass:: GtzanReader
    :members:
 

tests/corpus/io/test_gtzan.py

@@ -1,9 +1,11 @@
 import os
 
 import pytest
+import requests_mock
 
 from audiomate import corpus
 from audiomate.corpus import io
+from audiomate.corpus.io import gtzan
 from tests import resources
 
 
@@ -17,6 +19,29 @@ def data_path():
     return resources.sample_corpus_path('gtzan')
 
 
+@pytest.fixture()
+def tar_data():
+    with open(resources.get_resource_path(['sample_files', 'cv_corpus_v1.tar.gz']), 'rb') as f:
+        return f.read()
+
+
+class TestGtzanDownloader:
+
+    def test_download(self, tar_data, tmpdir):
+        target_folder = tmpdir.strpath
+        downloader = io.GtzanDownloader()
+
+        with requests_mock.Mocker() as mock:
+            mock.get(gtzan.DOWNLOAD_URL, content=tar_data)
+
+            downloader.download(target_folder)
+
+        assert os.path.isfile(os.path.join(target_folder, 'cv-valid-dev.csv'))
+        assert os.path.isdir(os.path.join(target_folder, 'cv-valid-dev'))
+        assert os.path.isfile(os.path.join(target_folder, 'cv-valid-train.csv'))
+        assert os.path.isdir(os.path.join(target_folder, 'cv-valid-train'))
+
+
 class TestGtzanReader:
 
     def test_load_files(self, reader, data_path):