fix: Update client to properly infer that flac (along with other browser renderable audio types) are loaded as DatapointType.AUDIO (#783)

This was previously system dependent. Some systems would load this from /etc/mime.types, but it was not defined in the Python's built-in mimetypes.

Author: munkyshi

kolena/dataset/dataset.py

@@ -114,6 +114,9 @@ def _infer_datatype_value_from_url(url: str) -> str:
             return datatype
     elif url.endswith(".pcd"):
         return DatapointType.POINT_CLOUD.value
+    # Explicit handling for formats not consistently in Python's built-in mimetypes
+    elif url.lower().endswith((".flac", ".ogg", ".opus", ".aac", ".m4a")):
+        return DatapointType.AUDIO.value
 
     return DatapointType.TABULAR.value
 

tests/unit/dataset/test_dataset.py

@@ -52,6 +52,12 @@
         ("test.pcd", DatapointType.POINT_CLOUD),
         ("gcp://summary.pdf", DatapointType.DOCUMENT),
         ("//my.mp3", DatapointType.AUDIO),
+        ("s3://bucket/audio.flac", DatapointType.AUDIO),
+        ("https://example.com/audio.FLAC", DatapointType.AUDIO),
+        ("audio.ogg", DatapointType.AUDIO),
+        ("/path/to/audio.opus", DatapointType.AUDIO),
+        ("audio.aac", DatapointType.AUDIO),
+        ("audio.m4a", DatapointType.AUDIO),
         (None, DatapointType.TABULAR),
         (123, DatapointType.TABULAR),
     ],
@@ -72,6 +78,13 @@ def test__infer_datatype_value(uri: Any, expected: str) -> None:
         (".csv", DatapointType.DOCUMENT),
         (".pdf", DatapointType.DOCUMENT),
         (".mp3", DatapointType.AUDIO),
+        (".flac", DatapointType.AUDIO),
+        ("flac", DatapointType.AUDIO),
+        ("FLAC", DatapointType.AUDIO),
+        (".ogg", DatapointType.AUDIO),
+        (".opus", DatapointType.AUDIO),
+        (".aac", DatapointType.AUDIO),
+        (".m4a", DatapointType.AUDIO),
         ("tabular", DatapointType.TABULAR),
         (None, DatapointType.TABULAR),
         (123, DatapointType.TABULAR),