fix: end ElevenLabs server VAD turns#5998
Conversation
![devin-ai-integration[bot]](https://avatars.githubusercontent.com/in/811515?s=60&v=4){kind=small}
| from livekit.plugins.elevenlabs import STT | ||
| from livekit.plugins.elevenlabs.stt import SpeechStream, VADOptions | ||
|
|
||
| pytestmark = [pytest.mark.unit, pytest.mark.plugin("elevenlabs")] |
There was a problem hiding this comment.
🟡 Test module declares two category markers, violating "exactly one category" rule
The test file uses pytestmark = [pytest.mark.unit, pytest.mark.plugin("elevenlabs")] which declares two categories. AGENTS.md explicitly states: "Every test module declares exactly one category via a module-level marker." All other plugin tests (e.g., tests/test_plugin_elevenlabs_tts.py:13, tests/test_plugin_assemblyai_stt.py:13) use only pytest.mark.plugin("..."). Having both unit and plugin means this test will be collected during --unit runs (the CI unit gate), even though it imports from livekit.plugins.elevenlabs — the unit category is defined as "fast, hermetic, no external providers/credentials/network." The correct marker should be pytestmark = pytest.mark.plugin("elevenlabs") only, consistent with all other plugin test files.
| pytestmark = [pytest.mark.unit, pytest.mark.plugin("elevenlabs")] | |
| pytestmark = pytest.mark.plugin("elevenlabs") |
Was this helpful? React with 👍 or 👎 to provide feedback.
| if is_given(self._opts.server_vad) and self._opts.server_vad is not None: | ||
| self._event_ch.send_nowait(stt.SpeechEvent(type=SpeechEventType.END_OF_SPEECH)) | ||
| self._speaking = False |
There was a problem hiding this comment.
🚩 Pre-existing inconsistency: _connect_ws commit_strategy check vs new _process_stream_event check
The new code at livekit-plugins/livekit-plugins-elevenlabs/livekit/plugins/elevenlabs/stt.py:600 correctly uses is_given(self._opts.server_vad) and self._opts.server_vad is not None to determine whether server VAD is active. However, the pre-existing code at livekit-plugins/livekit-plugins-elevenlabs/livekit/plugins/elevenlabs/stt.py:484 uses self._opts.server_vad is None to determine commit_strategy. Since NotGiven.__bool__ returns False but NOT_GIVEN is None is False, when server_vad is NOT_GIVEN (the default), _connect_ws incorrectly sets commit_strategy="vad" instead of "manual". This means the WebSocket connects in VAD mode even when no server VAD was requested. Line 494's walrus assignment if server_vad := self._opts.server_vad: is saved by NotGiven.__bool__ returning False, so no crash occurs on the .get() calls. The fix would be to change line 484 to something like commit_strategy = "vad" if is_given(self._opts.server_vad) and self._opts.server_vad is not None else "manual".
Was this helpful? React with 👍 or 👎 to provide feedback.
1 participant
Summary
Fixes #5849.
To verify
python -m py_compile livekit-plugins\livekit-plugins-elevenlabs\livekit\plugins\elevenlabs\stt.py tests\test_plugin_elevenlabs_stt.pypython -m ruff check livekit-plugins\livekit-plugins-elevenlabs\livekit\plugins\elevenlabs\stt.py tests\test_plugin_elevenlabs_stt.pypython -m ruff format --check livekit-plugins\livekit-plugins-elevenlabs\livekit\plugins\elevenlabs\stt.py tests\test_plugin_elevenlabs_stt.pyPYTHONPATH=livekit-agents;livekit-plugins\livekit-plugins-elevenlabs python -m pytest tests\test_plugin_elevenlabs_stt.py -q