Gracefully handle web and cloud data sources when scanning for audio …

…files. PiperOrigin-RevId: 645884486
google-research · Jun 25, 2024 · 5a374b9 · 5a374b9
1 parent 6c77181
commit 5a374b9
Show file tree

Hide file tree

Showing 2 changed files with 12 additions and 2 deletions.
diff --git a/chirp/inference/embed_lib.py b/chirp/inference/embed_lib.py
@@ -84,7 +84,13 @@ def create_source_infos(
   """
   source_files = []
   for pattern in source_file_patterns:
-    for source_file in epath.Path('').glob(pattern):
+    if '://' in pattern:
+      root, pattern = pattern.split('://')
+      root = root + '://'
+    else:
+      root = ''
+
+    for source_file in epath.Path(root).glob(pattern):
       source_files.append(source_file)
 
   source_file_splits = []

diff --git a/chirp/inference/tests/embed_test.py b/chirp/inference/tests/embed_test.py
@@ -435,7 +435,11 @@ def test_frame_audio(self):
 
   def test_create_source_infos(self):
     # Just one file, but it's all good.
-    globs = [path_utils.get_absolute_path('inference/tests/testdata/clap.wav')]
+    globs = [
+        path_utils.get_absolute_path(
+            'inference/tests/testdata/clap.wav'
+        ).as_posix()
+    ]
     # Disable sharding by setting shard_len_s <= 0.
     got_infos = embed_lib.create_source_infos(
         globs, shard_len_s=-1, num_shards_per_file=100