test(audio): cover Gemini transcription conflicts

This commit is contained in:
Marcin Juraszek 2026-09-22 23:21:07 +02:00
parent 2ab4fb33b2
commit 0848080432

View file

@ -220,6 +220,22 @@ class TestTransformRequest:
"diarization_mode": "speaker",
}
def test_global_drop_params_removes_keywords_with_word_timestamps(self, config, monkeypatch):
monkeypatch.setattr(litellm, "drop_params", True)
request_data = config.transform_audio_transcription_request(
model="gemini-3.5-transcribe",
audio_file=("sample.wav", AUDIO_BYTES, "audio/wav"),
optional_params={
"keywords": ["alpha"],
"timestamp_granularities": ["word"],
},
litellm_params={},
)
transcription_config = request_data.data["generation_config"]["transcription_config"]
assert "custom_vocabulary" not in transcription_config
assert transcription_config["mode"]["timestamp_granularities"] == ("word",)
def test_word_timestamp_granularity_maps_to_verbatim_diarization_mode(self, config):
request_data = config.transform_audio_transcription_request(
model="gemini-3.5-transcribe",
@ -253,6 +269,18 @@ class TestTransformRequest:
}
}
@pytest.mark.parametrize("response_format", ["srt", "vtt"])
def test_keywords_with_subtitle_response_format_raise(self, config, response_format, monkeypatch):
monkeypatch.setattr(litellm, "drop_params", False)
with pytest.raises(litellm.UnsupportedParamsError, match="custom vocabulary with word timestamps"):
config.transform_audio_transcription_request(
model="gemini-3.5-transcribe",
audio_file=("sample.wav", AUDIO_BYTES, "audio/wav"),
optional_params={"keywords": ["alpha"], "response_format": response_format},
litellm_params={},
)
@pytest.mark.parametrize("response_format", ["json", "text", "verbose_json"])
def test_non_subtitle_response_format_sends_no_mode(self, config, response_format):
request_data = config.transform_audio_transcription_request(