fix: narrow FunASR STT exception handling

Author: LauraGPT

plugins/funasr/pyproject.toml

@@ -28,7 +28,7 @@ raw-options = { root = "..", search_parent_directories = true, fallback_version
 packages = ["vision_agents"]
 
 [tool.hatch.build.targets.sdist]
-include = ["/vision_agents"]
+include = ["vision_agents"]
 
 [tool.uv.sources]
 vision-agents = { workspace = true }

plugins/funasr/tests/test_funasr_stt.py

@@ -1,6 +1,53 @@
+import asyncio
+
+import pytest
+
 from vision_agents.plugins import funasr
 
 
+class _FakeAutoModel:
+    pass
+
+
+class _FakeSamples:
+    def __init__(self, size=0):
+        self.size = size
+
+
+class _FakePcmData:
+    def __init__(
+        self,
+        *,
+        samples=None,
+        sample_rate=16000,
+        channels=1,
+        format=None,
+        duration_ms=0,
+    ):
+        self.samples = samples if samples is not None else _FakeSamples()
+        self.sample_rate = sample_rate
+        self.channels = channels
+        self.format = format
+        self.duration_ms = duration_ms
+
+    def resample(self, sample_rate):
+        self.sample_rate = sample_rate
+        return self
+
+    def to_float32(self):
+        return self
+
+    def append(self, audio_data):
+        return audio_data
+
+
+class _ExplodingPcmData:
+    samples = _FakeSamples(size=1)
+
+    def resample(self, sample_rate):
+        raise AssertionError("unexpected audio bug")
+
+
 class TestSTT:
     """Unit tests for the FunASR STT plugin (no model loaded)."""
 
@@ -24,3 +71,24 @@ def test_custom_params(self):
         assert stt.language == "zh"
         assert stt.device == "cuda"
         assert stt.use_itn is False
+
+    def test_process_audio_raises_unexpected_buffer_errors(self):
+        stt = funasr.STT(client=_FakeAutoModel())
+
+        with pytest.raises(AssertionError, match="unexpected audio bug"):
+            asyncio.run(stt.process_audio(_ExplodingPcmData(), participant=object()))
+
+    def test_process_buffer_raises_unexpected_transcription_errors(self):
+        stt = funasr.STT(client=_FakeAutoModel())
+        stt._audio_buffer = _FakePcmData(
+            samples=_FakeSamples(size=16000),
+            duration_ms=1000,
+        )
+
+        async def raise_unexpected_error(*, audio_array):
+            raise AssertionError("unexpected transcription bug")
+
+        stt._transcribe = raise_unexpected_error
+
+        with pytest.raises(AssertionError, match="unexpected transcription bug"):
+            asyncio.run(stt._process_buffer(participant=object()))

plugins/funasr/vision_agents/plugins/funasr/stt.py

@@ -136,8 +136,10 @@ async def process_audio(
             if should_process:
                 await self._process_buffer(participant)
 
-        except Exception:
-            logger.exception("Error buffering audio for FunASR")
+        except ValueError:
+            logger.exception("Invalid PCM audio while buffering for FunASR")
+        except RuntimeError:
+            logger.exception("Audio buffering/runtime error for FunASR")
 
     async def _process_buffer(self, participant: Participant):
         """Process the current audio buffer through FunASR."""
@@ -158,7 +160,7 @@ async def _process_buffer(self, participant: Participant):
 
         try:
             text = await self._transcribe(audio_array=audio_array)
-        except Exception:
+        except (RuntimeError, ValueError, KeyError, IndexError, TypeError):
             logger.exception("Error processing audio buffer with FunASR")
             return