Revert "Merge pull request #3424 from mozilla/io-fixes"

This reverts commit ab1288ffde, reversing
changes made to 08d18d7328.

.dockerignore

@@ -1,2 +0,0 @@
-tensorflow/
-data/

bin/compare_samples.py

@@ -15,8 +15,8 @@ def fail(message):
 
 
 def compare_samples():
-    sample1 = load_sample(CLI_ARGS.sample1).unpack()
-    sample2 = load_sample(CLI_ARGS.sample2).unpack()
+    sample1 = load_sample(CLI_ARGS.sample1)
+    sample2 = load_sample(CLI_ARGS.sample2)
     if sample1.audio_format != sample2.audio_format:
         fail('Samples differ on: audio-format ({} and {})'.format(sample1.audio_format, sample2.audio_format))
     if sample1.duration != sample2.duration:

training/deepspeech_training/train.py

@@ -811,7 +811,7 @@ def export():
         load_graph_for_evaluation(session)
 
         output_filename = FLAGS.export_file_name + '.pb'
-        if FLAGS.remove_export:
+        if FLAGS.remove_remote_export:
             if isdir_remote(FLAGS.export_dir):
                 log_info('Removing old export')
                 remove_remote(FLAGS.export_dir)

training/deepspeech_training/util/audio.py

@@ -118,19 +118,15 @@ class Sample:
         self.audio_type = new_audio_type
 
 
-def _unpack_and_change_audio_type(sample_and_audio_type):
-    packed_sample, audio_type, bitrate = sample_and_audio_type
-    if hasattr(sample, 'unpack'):
-        sample = packed_sample.unpack()
-    else:
-        sample = packed_sample
+def _change_audio_type(sample_and_audio_type):
+    sample, audio_type, bitrate = sample_and_audio_type
     sample.change_audio_type(audio_type, bitrate=bitrate)
     return sample
 
 
-def change_audio_types(packed_samples, audio_type=AUDIO_TYPE_PCM, bitrate=None, processes=None, process_ahead=None):
+def change_audio_types(samples, audio_type=AUDIO_TYPE_PCM, bitrate=None, processes=None, process_ahead=None):
     with LimitingPool(processes=processes, process_ahead=process_ahead) as pool:
-        yield from pool.imap(_unpack_and_change_audio_type, map(lambda s: (s, audio_type, bitrate), packed_samples))
+        yield from pool.imap(_change_audio_type, map(lambda s: (s, audio_type, bitrate), samples))
 
 
 def get_audio_type_from_extension(ext):

training/deepspeech_training/util/augmentations.py

@@ -152,10 +152,7 @@ def _init_augmentation_worker(preparation_context):
 
 def _load_and_augment_sample(timed_sample, context=None):
     sample, clock = timed_sample
-    if hasattr(sample, 'unpack'):
-        realized_sample = sample.unpack()
-    else:
-        realized_sample = sample
+    realized_sample = sample.unpack()
     return _augment_sample((realized_sample, clock), context)
 
 

training/deepspeech_training/util/taskcluster.py

@@ -14,6 +14,8 @@ import sys
 
 from pkg_resources import parse_version
 
+from .io import isdir_remote, open_remote, is_remote_path
+
 DEFAULT_SCHEMES = {
     'deepspeech': 'https://community-tc.services.mozilla.com/api/index/v1/task/project.deepspeech.deepspeech.native_client.%(branch_name)s.%(arch_string)s/artifacts/public/%(artifact_name)s',
     'tensorflow': 'https://community-tc.services.mozilla.com/api/index/v1/task/project.deepspeech.tensorflow.pip.%(branch_name)s.%(arch_string)s/artifacts/public/%(artifact_name)s'
@@ -41,7 +43,7 @@ def maybe_download_tc(target_dir, tc_url, progress=True):
 
     assert target_dir is not None
 
-    if not os.path.isdir(target_dir):
+    if not is_remote_path(target_dir):
         try:
             os.makedirs(target_dir)
         except OSError as e:
@@ -60,7 +62,7 @@ def maybe_download_tc(target_dir, tc_url, progress=True):
         print('File already exists: %s' % target_file)
 
     if is_gzip:
-        with open(target_file, "r+b") as frw:
+        with open_remote(target_file, "r+b") as frw:
             decompressed = gzip.decompress(frw.read())
             frw.seek(0)
             frw.write(decompressed)