fix(audio): preserve transcription config compatibility

This commit is contained in:
Marcin Juraszek 2026-09-22 20:03:19 +00:00
parent fd023cc46a
commit b646791578
4 changed files with 62 additions and 3 deletions

View file

@ -122,10 +122,9 @@ def get_supported_openai_params(
transcription_provider_config = litellm.ProviderConfigManager.get_provider_audio_transcription_config(
model=model, provider=LlmProviders.OPENAI
)
if isinstance(transcription_provider_config, litellm.OpenAIGPTAudioTranscriptionConfig):
if transcription_provider_config is not None:
return transcription_provider_config.get_supported_openai_params(model=model)
else:
raise ValueError(f"Unsupported provider config: {transcription_provider_config} for model: {model}")
raise ValueError(f"Unsupported provider config: {transcription_provider_config} for model: {model}")
return litellm.OpenAIConfig().get_supported_openai_params(model=model)
elif custom_llm_provider == "sap":
if request_type == "chat_completion":

View file

@ -2148,6 +2148,20 @@ def test_gemini_model_info_reports_transcription_params(monkeypatch):
assert "keywords" in model_info["supported_openai_params"]
def test_gemini_live_transcription_dispatch_has_no_batch_params():
from litellm.litellm_core_utils.get_supported_openai_params import (
get_supported_openai_params,
)
params = get_supported_openai_params(
model="gemini-3.5-transcribe-live",
custom_llm_provider="gemini",
request_type="transcription",
)
assert params is None
@pytest.mark.parametrize(
"model, expected_bool",
[

View file

@ -15,6 +15,7 @@ from litellm.litellm_core_utils.audio_utils.utils import (
get_audio_file_content_hash,
get_audio_file_for_health_check,
get_audio_file_name,
normalize_transcription_keywords,
process_audio_file,
)
@ -349,6 +350,19 @@ class TestNormalizeTranscriptionLanguageToBcp47:
assert normalize_transcription_language_to_bcp47(language) == expected
class TestNormalizeTranscriptionKeywords:
@pytest.mark.parametrize(
("keywords", "expected"),
[
(["alpha", "", 42, "beta"], ("alpha", "beta")),
("alpha", ()),
(None, ()),
],
)
def test_normalization(self, keywords, expected):
assert normalize_transcription_keywords(keywords) == expected
class TestResolveSpeechMediaType:
@pytest.mark.parametrize(
("upstream_content_type", "response_format", "expected"),

View file

@ -864,6 +864,38 @@ def test_gemini_session_update_maps_input_audio_transcription_settings():
}
@pytest.mark.parametrize(
("transcription", "expected"),
[
("invalid", {}),
({}, {}),
({"language": "", "keywords": ["", 42]}, {}),
(
{"language": "pl", "keywords": ["", 42, "term"]},
{
"languageCodes": ["pl-PL"],
"customVocabulary": ["term"],
},
),
],
)
def test_gemini_session_update_handles_input_audio_transcription_values(transcription, expected):
config = GeminiRealtimeConfig()
session_update = {
"type": "session.update",
"session": {"input_audio_transcription": transcription},
}
result = config.transform_realtime_request(
json.dumps(session_update),
"gemini-3.5-transcribe-live",
session_configuration_request=None,
)
setup = json.loads(result[0])
assert setup["setup"]["inputAudioTranscription"] == expected
def test_gemini_session_update_includes_input_audio_transcription_default():
"""Verify _handle_session_update includes inputAudioTranscription default."""
config = GeminiRealtimeConfig()