Fix CMUARCTIC text transcripts reader. (#4164)

pearu · web-flow · commit cc9f5150c415 · 2026-01-06T10:58:27.000+02:00
diff --git a/src/torchaudio/datasets/cmuarctic.py b/src/torchaudio/datasets/cmuarctic.py
@@ -1,4 +1,3 @@
-import csv
 import os
 from pathlib import Path
 from typing import Tuple, Union
@@ -128,9 +127,8 @@ def __init__(
                 )
         self._text = os.path.join(self._path, self._folder_text, self._file_text)
 
-        with open(self._text, "r") as text:
-            walker = csv.reader(text)
-            self._walker = list(walker)
+        with open(self._text, "r", newline=None) as text:
+            self._walker = [[line.rstrip("\n")] for line in text.readlines()]
 
     def __getitem__(self, n: int) -> Tuple[Tensor, int, str, str]:
         """Load the n-th sample from the dataset.
diff --git a/test/torchaudio_unittest/datasets/cmuarctic_test.py b/test/torchaudio_unittest/datasets/cmuarctic_test.py
@@ -11,7 +11,7 @@ def get_mock_dataset(root_dir):
     """
     mocked_data = []
     sample_rate = 16000
-    transcript = "This is a test transcript."
+    transcript = "This is a test transcript, with comma."
 
     base_dir = os.path.join(root_dir, "ARCTIC", "cmu_us_aew_arctic")
     txt_dir = os.path.join(base_dir, "etc")
