Skip to content

Open-source speech recognition toolkit for training, inference, streaming ASR, VAD, punctuation, speaker diarization pipelines, and OpenAI-compatible/MCP serving.

fix: count speaker overlap once per diarization segment

LauraGPTe7006a93159d
diff --git a/.github/workflows/test-moss-adapter.yml b/.github/workflows/test-moss-adapter.yml
index c13a3a37d..48acbe57c 100644
--- a/.github/workflows/test-moss-adapter.yml
+++ b/.github/workflows/test-moss-adapter.yml
@@ -4,6 +4,8 @@ on:
pull_request:
paths:
- "funasr/models/moss_transcribe_diarize/**"
+ - "funasr/models/campplus/utils.py"
+ - "tests/test_campplus_utils.py"
- "funasr/auto/auto_model.py"
- "funasr/bin/_server_app.py"
- "funasr/bin/server.py"
@@ -21,6 +23,8 @@ on:
- main
paths:
- "funasr/models/moss_transcribe_diarize/**"
+ - "funasr/models/campplus/utils.py"
+ - "tests/test_campplus_utils.py"
- "funasr/auto/auto_model.py"
- "funasr/bin/_server_app.py"
- "funasr/bin/server.py"
@@ -58,6 +62,7 @@ jobs:
- name: Check formatting and syntax
run: |
python -m black --check \
+ tests/test_campplus_utils.py \
funasr/models/moss_transcribe_diarize/model.py \
tests/test_moss_transcribe_diarize_model.py \
tests/test_moss_transcribe_diarize_docs.py \
@@ -71,6 +76,7 @@ jobs:
- name: Run adapter and documentation contracts
run: |
python -m pytest -q \
+ tests/test_campplus_utils.py \
tests/test_moss_transcribe_diarize_model.py \
tests/test_moss_transcribe_diarize_docs.py \
tests/test_server_app_openai_segments.py \
diff --git a/funasr/models/campplus/utils.py b/funasr/models/campplus/utils.py
index dc2ff0300..9ea12a7ed 100644
--- a/funasr/models/campplus/utils.py
+++ b/funasr/models/campplus/utils.py
@@ -254,26 +254,21 @@ def smooth(res, mindur=0.7):
def distribute_spk(sentence_list, sd_time_list):
- """Distribute spk.
-
- Args:
- sentence_list: TODO.
- sd_time_list: TODO.
- """
+ """Assign each sentence to the speaker with the greatest total overlap.
+
+ Sentence times are milliseconds; diarization times are seconds. Equal
+ totals retain the first overlapping speaker, and no overlap defaults to 0.
+ """
sd_time_list = [(spk_st * 1000, spk_ed * 1000, spk) for spk_st, spk_ed, spk in sd_time_list]
for d in sentence_list:
sentence_start = d['start']
sentence_end = d['end']
- sentence_spk = 0
- max_overlap = 0
+ speaker_overlaps = {}
for spk_st, spk_ed, spk in sd_time_list:
overlap = max(min(sentence_end, spk_ed) - max(sentence_start, spk_st), 0)
- if overlap > max_overlap:
- max_overlap = overlap
- sentence_spk = spk
- if overlap > 0 and sentence_spk == spk:
- max_overlap += overlap
- d['spk'] = int(sentence_spk)
+ if overlap > 0:
+ speaker_overlaps[spk] = speaker_overlaps.get(spk, 0) + overlap
+ d['spk'] = int(max(speaker_overlaps, key=speaker_overlaps.get, default=0))
return sentence_list
diff --git a/tests/test_campplus_utils.py b/tests/test_campplus_utils.py
new file mode 100644
index 000000000..94c60b47c
--- /dev/null
+++ b/tests/test_campplus_utils.py
@@ -0,0 +1,71 @@
+"""Speaker assignment uses the actual overlap with each sentence."""
+
+from itertools import permutations
+
+import numpy as np
+import pytest
+
+from funasr.models.campplus.utils import distribute_spk
+
+
+def test_distribute_spk_does_not_double_count_the_first_overlap():
+ sentences = [{"start": 0, "end": 10000, "text": "first and second"}]
+
+ result = distribute_spk(sentences, [(0, 4, 0), (4, 10, 1)])
+
+ assert result == [{"start": 0, "end": 10000, "text": "first and second", "spk": 1}]
+
+
+ "timeline",
+ list(permutations([(0, 2, 7), (2, 5, 9), (5, 7, 7), (7, 10, 9)])),
+)
+def test_distribute_spk_totals_nonconsecutive_turns_independent_of_order(timeline):
+ sentences = [{"start": 0, "end": 10000}]
+
+ assert distribute_spk(sentences, timeline) == [{"start": 0, "end": 10000, "spk": 9}]
+
+
+ ("timeline", "expected"),
+ [
+ ([(0, 2, 7), (2, 7, 9), (7, 10, 7)], 7),
+ ([(2, 7, 9), (0, 2, 7), (7, 10, 7)], 9),
+ ([(-2, 0, 9), (0, 5, 7), (5, 10, 9)], 7),
+ ],
+)
+def test_distribute_spk_breaks_total_ties_by_first_positive_overlap(timeline, expected):
+ assert distribute_spk([{"start": 0, "end": 10000}], timeline) == [
+ {"start": 0, "end": 10000, "spk": expected}
+ ]
+
+
+def test_distribute_spk_clips_to_each_sentence_and_preserves_objects():
+ first = {"start": 4000, "end": 6000, "text": "clipped", "spk": 99}
+ second = {"start": 6000, "end": 7000, "timestamp": [[6000, 7000]]}
+ sentences = [first, second]
+ timeline = [(0, 4, 8), (4, 5.5, np.int64(3)), (5.5, 100, np.int64(9))]
+
+ result = distribute_spk(sentences, timeline)
+
+ assert result is sentences
+ assert result[0] is first
+ assert result[1] is second
+ assert result == [
+ {"start": 4000, "end": 6000, "text": "clipped", "spk": 3},
+ {"start": 6000, "end": 7000, "timestamp": [[6000, 7000]], "spk": 9},
+ ]
+ assert type(first["spk"]) is int
+ assert type(second["spk"]) is int
+
+
[email protected]("timeline", [[], [(0, 1, 8), (2, 3, 9)]])
+def test_distribute_spk_defaults_to_zero_without_positive_overlap(timeline):
+ assert distribute_spk([{"start": 1000, "end": 2000}], timeline) == [
+ {"start": 1000, "end": 2000, "spk": 0}
+ ]
+
+
+def test_distribute_spk_accepts_empty_sentences():
+ sentences = []
+ assert distribute_spk(sentences, [(0, 1, 7)]) is sentences