Add Word-Level Alignment

src/senselab/audio/tasks/forced_alignment/data_structures.py

@@ -51,13 +51,6 @@ class TranscriptionResult(TypedDict):
     language: str
 
 
-class AlignedTranscriptionResult(TypedDict):
-    """A list of segments and word segments of a speech."""
-
-    segments: List[SingleAlignedSegment]
-    word_segments: List[SingleWordSegment]
-
-
 @dataclass
 class Point:
     """Represents a point in the alignment path.

src/senselab/audio/tasks/forced_alignment/evaluation.py

@@ -0,0 +1,26 @@
+"""Provides alignment evaluation functions."""
+
+from senselab.utils.data_structures.script_line import ScriptLine
+
+
+def compare_alignments(alignment_one: ScriptLine, alignment_two: ScriptLine, difference_tolerance: float = 0.1) -> None:
+    """Check if two alignments are within the specified difference tolerance.
+
+    Args:
+        alignment_one (ScriptLine): The first alignment segment.
+        alignment_two (ScriptLine): The second alignment segment.
+        difference_tolerance (float): Allowed difference in start and end times (seconds).
+
+    Raises:
+        AssertionError: If the start or end times differ by more than the tolerance.
+    """
+    print(f"Texts:    {alignment_one.text} | {alignment_two.text}")
+    if alignment_one.text and alignment_two.text:
+        assert alignment_one.text.lower() == alignment_two.text.lower(), f"{alignment_one.text} {alignment_two.text}"
+    if alignment_one.start is not None and alignment_two.start is not None:
+        assert abs(alignment_one.start - alignment_two.start) < difference_tolerance
+    if alignment_one.end is not None and alignment_two.end is not None:
+        assert abs(alignment_one.end - alignment_two.end) < difference_tolerance
+    if alignment_one.chunks and alignment_two.chunks and len(alignment_one.chunks) == len(alignment_two.chunks):
+        for a1, a2 in zip(alignment_one.chunks, alignment_two.chunks):
+            compare_alignments(a1, a2)