Remove option for different initializations for greedy ORC WER

fgnt · Aug 26, 2024 · ee533dd · ee533dd
1 parent 6998d7d
commit ee533dd
Show file tree

Hide file tree

Showing 4 changed files with 1 addition and 25 deletions.
diff --git a/meeteval/wer/__main__.py b/meeteval/wer/__main__.py
@@ -189,7 +189,6 @@ def greedy_orcwer(
         regex=None,
         reference_sort='segment_if_available',
         hypothesis_sort='segment_if_available',
-        assignment_initialization='cp',
         uem=None,
         partial=False,
         normalizer=None,
@@ -203,7 +202,6 @@ def greedy_orcwer(
         uem=uem,
         partial=partial,
         normalizer=normalizer,
-        assignment_initialization=assignment_initialization,
     )
     _save_results(results, hypothesis, per_reco_out, average_out)
 
@@ -551,17 +549,6 @@ def add_argument(self, command_parser, name, p):
                      '- lower,rm(.?!,): Lowercase the transcript and remove punctuations (.,?!).',
                 choices=[None, 'lower,rm(.?!,)'],
             )
-        elif name == 'assignment_initialization':
-            command_parser.add_argument(
-                '--assignment-initialization',
-                choices=['cp', 'random', 'constant'],
-                help='Specifies how the greedy search should be initialized.\n'
-                     'Choices:\n'
-                     '- cp: Use the assignment from the cpWER. Ensures fast convergence when the speaker labels '
-                     'are already somewhat correct. Requires speaker/stream labels in both, reference and hypothesis.\n'
-                     '- random: Initializes the assignment randomly. This is not deterministic.\n'
-                     '- constant: Initializes the label for all segments to 0.'
-            )
         elif name == 'partial':
             command_parser.add_argument(
                 '--partial',

diff --git a/meeteval/wer/api.py b/meeteval/wer/api.py
@@ -127,7 +127,6 @@ def greedy_orcwer(
         regex=None,
         reference_sort='segment_if_available',
         hypothesis_sort='segment_if_available',
-        assignment_initialization='cp',
         uem=None,
         partial=False,
         normalizer=None,
@@ -142,7 +141,6 @@ def greedy_orcwer(
         reference, hypothesis, partial=partial,
         reference_sort=reference_sort,
         hypothesis_sort=hypothesis_sort,
-        assignment_initialization=assignment_initialization,
     )
     return results
 

diff --git a/meeteval/wer/wer/orc.py b/meeteval/wer/wer/orc.py
@@ -316,8 +316,6 @@ def greedy_orc_word_error_rate(
         hypothesis: 'SegLST',
         reference_sort='segment_if_available',
         hypothesis_sort='segment_if_available',
-        *,
-        assignment_initialization='cp',
 ):
     """
     # All correct on a single channel
@@ -355,7 +353,7 @@ def matching(reference, hypothesis):
         distance, assignment = greedy_combination_matching(
             reference.T['words'],
             [[w for words in stream.T['words'] for w in words] for stream in hypothesis.values()],
-            initial_assignment=initialize_assignment(reference, hypothesis, initialization=assignment_initialization),
+            initial_assignment=initialize_assignment(reference, hypothesis, initialization='cp'),
         )
         return distance, assignment
 
@@ -394,7 +392,6 @@ def greedy_orc_word_error_rate_multifile(
         partial=False,
         reference_sort='segment_if_available',
         hypothesis_sort='segment_if_available',
-        assignment_initialization='cp',
 ) -> 'dict[str, OrcErrorRate]':
     """
     Computes the ORC WER for each example in the reference and hypothesis files.
@@ -408,7 +405,6 @@ def greedy_orc_word_error_rate_multifile(
             greedy_orc_word_error_rate,
             reference_sort=reference_sort,
             hypothesis_sort=hypothesis_sort,
-            assignment_initialization=assignment_initialization,
         ), reference, hypothesis,
         partial=partial
     )
diff --git a/tests/test_cli.py b/tests/test_cli.py
@@ -89,11 +89,6 @@ def test_burn_greedy_orc():
     # Test sort option. Only test the ones that are available for ORC
     run('python -m meeteval.wer greedy_orcwer -h hyp.stm -r ref.stm --reference-sort "segment" --hypothesis-sort "false"')
 
-    # Test assignment initialization
-    run(f'meeteval-wer greedy_orcwer -h hyp.stm -r ref.stm --assignment-initialization random')
-    run(f'meeteval-wer greedy_orcwer -h hyp.stm -r ref.stm --assignment-initialization constant')
-
-
 
 def test_burn_mimo():
     run(f'python -m meeteval.wer mimower -h hyp.stm -r ref.stm')