From 643f443ba259ca5545cc902ee821e7f60dc1ed8f Mon Sep 17 00:00:00 2001
From: LauraGPT <18321252+LauraGPT@users.noreply.github.com>
Date: Thu, 1 Oct 2026 04:54:48 +0000
Subject: [PATCH] fix(runtime): skip empty deduplicated event segments

Signed-off-by: LauraGPT <18321252+LauraGPT@users.noreply.github.com>
---
 .../funasr_torch/utils/postprocess_utils.py   |  2 +
 .../funasr_onnx/utils/postprocess_utils.py    |  2 +
 .../test_rich_transcription_event_segments.py | 52 +++++++++++++++++++
 3 files changed, 56 insertions(+)
 create mode 100644 tests/test_rich_transcription_event_segments.py

diff --git a/runtime/python/libtorch/funasr_torch/utils/postprocess_utils.py b/runtime/python/libtorch/funasr_torch/utils/postprocess_utils.py
index eb4e2232e..93a6d0eb3 100644
--- a/runtime/python/libtorch/funasr_torch/utils/postprocess_utils.py
+++ b/runtime/python/libtorch/funasr_torch/utils/postprocess_utils.py
@@ -355,6 +355,8 @@ def rich_transcription_postprocess(s):
             continue
         if get_event(s_list[i]) == cur_ent_event and get_event(s_list[i]) != None:
             s_list[i] = s_list[i][1:]
+        if len(s_list[i]) == 0:
+            continue
         # else:
         cur_ent_event = get_event(s_list[i])
         if get_emo(s_list[i]) != None and get_emo(s_list[i]) == get_emo(new_s):
diff --git a/runtime/python/onnxruntime/funasr_onnx/utils/postprocess_utils.py b/runtime/python/onnxruntime/funasr_onnx/utils/postprocess_utils.py
index d144d31ab..38eefc49e 100644
--- a/runtime/python/onnxruntime/funasr_onnx/utils/postprocess_utils.py
+++ b/runtime/python/onnxruntime/funasr_onnx/utils/postprocess_utils.py
@@ -409,6 +409,8 @@ def rich_transcription_postprocess(s):
             continue
         if get_event(s_list[i]) == cur_ent_event and get_event(s_list[i]) != None:
             s_list[i] = s_list[i][1:]
+        if len(s_list[i]) == 0:
+            continue
         # else:
         cur_ent_event = get_event(s_list[i])
         if get_emo(s_list[i]) != None and get_emo(s_list[i]) == get_emo(new_s):
diff --git a/tests/test_rich_transcription_event_segments.py b/tests/test_rich_transcription_event_segments.py
new file mode 100644
index 000000000..ed756abfd
--- /dev/null
+++ b/tests/test_rich_transcription_event_segments.py
@@ -0,0 +1,52 @@
+"""Keep event-only language segments consistent across shipped runtimes."""
+
+import importlib.util
+from pathlib import Path
+
+import pytest
+
+
+ROOT = Path(__file__).resolve().parents[1]
+MODULES = [
+    ROOT / "funasr/utils/postprocess_utils.py",
+    ROOT / "runtime/python/onnxruntime/funasr_onnx/utils/postprocess_utils.py",
+    ROOT / "runtime/python/libtorch/funasr_torch/utils/postprocess_utils.py",
+]
+
+
+@pytest.fixture(params=MODULES, ids=["funasr", "onnx", "libtorch"])
+def postprocess(request):
+    spec = importlib.util.spec_from_file_location("_event_segments", request.param)
+    module = importlib.util.module_from_spec(spec)
+    spec.loader.exec_module(module)
+    return module
+
+
+@pytest.mark.parametrize("event", ["<|Cough|>", "<|Sneeze|>", "<|Applause|>"])
+@pytest.mark.parametrize("following_text", ["", "world"])
+def test_repeated_event_only_segments_keep_following_text(postprocess, event, following_text):
+    text = (
+        f"<|en|><|NEUTRAL|>{event}<|woitn|>hello"
+        f"<|zh|><|NEUTRAL|>{event}<|woitn|>"
+        f"<|ja|><|NEUTRAL|>{event}<|woitn|>"
+        f"<|en|><|NEUTRAL|><|Speech|><|woitn|>{following_text}"
+    )
+
+    result = postprocess.rich_transcription_postprocess(text)
+
+    assert result == postprocess.event_dict[event] + "hello" + following_text
+
+
+@pytest.mark.parametrize("event", ["<|Cough|>", "<|Sneeze|>", "<|Applause|>"])
+def test_repeated_event_only_segments_without_speech(postprocess, event):
+    text = f"<|en|><|NEUTRAL|>{event}<|zh|><|NEUTRAL|>{event}"
+
+    assert postprocess.rich_transcription_postprocess(text) == postprocess.event_dict[event]
+
+
+def test_distinct_event_only_segments_remain_visible(postprocess):
+    text = "<|en|><|NEUTRAL|><|Applause|><|zh|><|NEUTRAL|><|Laughter|>"
+
+    assert postprocess.rich_transcription_postprocess(text) == (
+        postprocess.event_dict["<|Applause|>"] + postprocess.event_dict["<|Laughter|>"]
+    )