fix: guard empty English timestamp segments

Avoid indexing empty sentence buffers when punctuation and timestamp entries outnumber split words, and add regression coverage for whitespace-only input.

Co-authored-by: Amir Fathi <[email protected]>
Signed-off-by: LauraGPT <[email protected]>

funasr/utils/timestamp_tools.py

@@ -281,16 +281,16 @@ def timestamp_sentence_en(
 
         punc_id = int(punc_id) if punc_id is not None else 1
         sentence_end = timestamp[1] if timestamp is not None else sentence_end
-        sentence_text = sentence_text[1:] if sentence_text[0] == ' ' else sentence_text
+        if sentence_text.startswith(" "):
+            sentence_text = sentence_text[1:]
         if is_sentence_start:
             sentence_start = timestamp[0] if timestamp is not None else sentence_start
             is_sentence_start = False
         if punc_id > 1:
             is_sentence_start = True
             sentence_text += punc_list[punc_id - 2]
-            sentence_text_seg = (
-                sentence_text_seg[:-1] if sentence_text_seg[-1] == " " else sentence_text_seg
-            )
+            if sentence_text_seg.endswith(" "):
+                sentence_text_seg = sentence_text_seg[:-1]
             if return_raw_text:
                 res.append(
                     {

tests/test_timestamp_tools.py

@@ -0,0 +1,39 @@
+from funasr.utils.timestamp_tools import timestamp_sentence_en
+
+
+def test_timestamp_sentence_en_handles_whitespace_only_segment():
+    result = timestamp_sentence_en(
+        punc_id_list=[3],
+        timestamp_postprocessed=[[0, 100]],
+        text_postprocessed=" ",
+        return_raw_text=True,
+    )
+
+    assert result == [
+        {
+            "text": ".",
+            "start": 0,
+            "end": 100,
+            "timestamp": [[0, 100]],
+            "raw_text": "",
+        }
+    ]
+
+
+def test_timestamp_sentence_en_preserves_normal_sentence_output():
+    result = timestamp_sentence_en(
+        punc_id_list=[1, 3],
+        timestamp_postprocessed=[[0, 100], [100, 220]],
+        text_postprocessed="hello world",
+        return_raw_text=True,
+    )
+
+    assert result == [
+        {
+            "text": "hello world.",
+            "start": 0,
+            "end": 220,
+            "timestamp": [[0, 100], [100, 220]],
+            "raw_text": "hello world",
+        }
+    ]