fix(audio-transcription): keep reply bodies out of watsonx, whisper, xai and validation errors

Watsonx and whisper no longer print the parsed reply after Got: when it has no text field. TranscriptionResponse and the xAI response model hide input in validation errors, so a malformed field cannot quote transcript fragments. Deepgram and ElevenLabs keep the original error text, which is now safe.

litellm/llms/deepgram/audio_transcription/transformation.py

@@ -158,7 +158,7 @@ class DeepgramAudioTranscriptionConfig(BaseAudioTranscriptionConfig):
             return response
 
         except Exception as e:
-            raise ValueError(f"Error transforming Deepgram response: {type(e).__name__}") from None
+            raise ValueError(f"Error transforming Deepgram response: {e}") from None
 
     def _reconstruct_diarized_transcript(self, words: list) -> str:
         """

litellm/llms/elevenlabs/audio_transcription/transformation.py

@@ -160,7 +160,7 @@ class ElevenLabsAudioTranscriptionConfig(BaseAudioTranscriptionConfig):
             return response
 
         except Exception as e:
-            raise ValueError(f"Error transforming ElevenLabs response: {type(e).__name__}") from None
+            raise ValueError(f"Error transforming ElevenLabs response: {e}") from None
 
     def get_complete_url(
         self,

litellm/llms/openai/transcriptions/whisper_transformation.py

@@ -141,7 +141,4 @@ class OpenAIWhisperAudioTranscriptionConfig(BaseAudioTranscriptionConfig):
         if any(key in raw_response_json for key in TranscriptionResponse.model_fields):
             return TranscriptionResponse(**raw_response_json)
         else:
-            raise ValueError(
-                "Invalid response format. Received response does not match the expected format. Got: ",
-                raw_response_json,
-            )
+            raise ValueError("Invalid response format. Received response does not match the expected format.")

litellm/llms/watsonx/audio_transcription/transformation.py

@@ -191,10 +191,7 @@ class IBMWatsonXAudioTranscriptionConfig(IBMWatsonXMixin, OpenAIWhisperAudioTran
             response_kwargs["usage"] = usage
 
         if not response_kwargs:
-            raise ValueError(
-                "Invalid response format. Received response does not match the expected format. Got: ",
-                response_object,
-            )
+            raise ValueError("Invalid response format. Received response does not match the expected format.")
 
         response: Final = TranscriptionResponse(**response_kwargs)
 

litellm/llms/xai/audio_transcription/transformation.py

@@ -39,7 +39,7 @@ class _XAISttWord(LiteLLMBaseModel):
 
 
 class _XAISttResponse(LiteLLMBaseModel):
-    model_config = ConfigDict(extra="allow")
+    model_config = ConfigDict(extra="allow", hide_input_in_errors=True)
     text: str = ""
     language: str = "unknown"
     duration: float | None = None

litellm/types/utils.py

@@ -2984,6 +2984,8 @@ _DETECTED_LANGUAGES: Final = TypeAdapter(tuple[TranscriptionDetectedLanguage, ..
 
 
 class TranscriptionResponse(OpenAIObject):
+    model_config = ConfigDict(hide_input_in_errors=True)
+
     text: str | None = None
     usage: TranscriptionUsageDurationObject | TranscriptionUsageTokensObject | None = None
     languages: Sequence[TranscriptionDetectedLanguage] | None = None

tests/unit/llms/deepgram/audio_transcription/test_deepgram_audio_transcription_transformation.py

@@ -601,18 +601,21 @@ def test_transform_response_keeps_the_transcript_out_of_malformed_words_errors(w
 @pytest.mark.parametrize(
     ("payload", "error"),
     [
-        ({"results": None}, "TypeError"),
-        (_deepgram_payload({"transcript": "hello", "words": 5}, {}), "TypeError"),
-        ({**_deepgram_payload({"transcript": "hello"}, {}), "metadata": None}, "TypeError"),
-        (_deepgram_payload({"transcript": ["hello"]}, {}), "ValidationError"),
+        ({"results": None}, "'NoneType' object is not subscriptable"),
+        (_deepgram_payload({"transcript": "hello", "words": 5}, {}), "object of type 'int' has no len()"),
+        (
+            {**_deepgram_payload({"transcript": "hello"}, {}), "metadata": None},
+            "'NoneType' object is not subscriptable",
+        ),
+        (_deepgram_payload({"transcript": ["hello"]}, {}), "1 validation error for TranscriptionResponse\ntext\n"),
     ],
 )
-def test_transform_response_names_only_the_failure_type_of_a_malformed_body(payload: object, error: str):
+def test_transform_response_wraps_the_error_of_the_failing_lookup_without_the_body(payload: object, error: str):
     raw_response: Final = httpx.Response(200, json=payload)
 
     with pytest.raises(ValueError, match="Error transforming Deepgram response: ") as exc_info:
         DeepgramAudioTranscriptionConfig().transform_audio_transcription_response(raw_response)
 
     assert exc_info.type is ValueError
-    assert str(exc_info.value) == f"Error transforming Deepgram response: {error}"
+    assert str(exc_info.value).startswith(f"Error transforming Deepgram response: {error}")
     assert "hello" not in "".join(traceback.format_exception(exc_info.value))

tests/unit/llms/elevenlabs/audio_transcription/test_transformation.py

@@ -129,6 +129,15 @@ def test_transform_audio_transcription_response_keeps_the_body_out_of_malformed_
     assert "sensitive" not in "".join(traceback.format_exception(exc_info.value))
 
 
+def test_transform_audio_transcription_response_names_the_failing_lookup_of_a_malformed_payload():
+    raw_response = httpx.Response(200, json={"text": "sensitive", "words": [None]})
+
+    with pytest.raises(ValueError, match=r"^Error transforming ElevenLabs response: 1 validation error for ") as exc_info:
+        ElevenLabsAudioTranscriptionConfig().transform_audio_transcription_response(raw_response=raw_response)
+
+    assert "sensitive" not in "".join(traceback.format_exception(exc_info.value))
+
+
 class TestElevenLabsAudioTranscription:
     @pytest.mark.usefixtures("elevenlabs_api_key_env")
     def test_elevenlabs_diarize_parameter_passthrough(self):

tests/unit/llms/mistral/audio_transcription/test_mistral_audio_transcription_transformation.py

@@ -4,6 +4,7 @@ from unittest.mock import MagicMock
 import httpx
 import litellm
 import pytest
+from pydantic import ValidationError
 
 from litellm.llms.base_llm.audio_transcription.transformation import (
     BaseAudioTranscriptionConfig,
@@ -202,3 +203,13 @@ def test_mistral_audio_transcription_response_transform_keeps_non_json_body_out_
         MistralAudioTranscriptionConfig().transform_audio_transcription_response(raw_response)
 
     assert "sensitive" not in str(exc_info.value)
+
+
+@pytest.mark.parametrize("text", [["sensitive"], {"value": "sensitive"}])
+def test_mistral_audio_transcription_response_transform_keeps_a_malformed_text_field_out_of_the_error(text: object):
+    raw_response = httpx.Response(200, json={"text": text, "segments": [{"text": "sensitive"}]})
+
+    with pytest.raises(ValidationError) as exc_info:
+        MistralAudioTranscriptionConfig().transform_audio_transcription_response(raw_response)
+
+    assert "sensitive" not in str(exc_info.value)

tests/unit/llms/openai/transcriptions/test_whisper_transformation.py

@@ -5,8 +5,10 @@ and transform_audio_transcription_response.
 
 import io
 import json
+import traceback
 from unittest.mock import MagicMock
 
+import httpx
 import pytest
 
 from litellm.llms.openai.transcriptions.whisper_transformation import (
@@ -104,3 +106,10 @@ class TestWhisperTransformResponse:
                     is_json=False,
                 )
             )
+
+    def test_json_object_without_transcription_fields_keeps_the_body_out_of_the_error(self):
+        config = OpenAIWhisperAudioTranscriptionConfig()
+        raw_response = httpx.Response(200, json={"segments": [{"text": "sensitive"}]})
+        with pytest.raises(ValueError, match="Invalid response format") as exc_info:
+            config.transform_audio_transcription_response(raw_response)
+        assert "sensitive" not in "".join(traceback.format_exception(exc_info.value))

tests/unit/llms/watsonx/audio_transcription/test_watsonx_audio_transcription_transformation.py

@@ -4,6 +4,7 @@ Tests for IBM WatsonX Audio Transcription.
 Validates the WatsonX transcription response transformation.
 """
 
+import traceback
 from unittest.mock import MagicMock
 
 import httpx
@@ -120,17 +121,21 @@ def test_transform_audio_transcription_response_copies_every_field_except_model(
     assert {key: response[key] for key in expected_extras} == expected_extras
 
 
-@pytest.mark.parametrize("payload", [{}, {"model": "whisper"}, {"duration": 2.0}, {"text": None, "usage": None}])
-def test_transform_audio_transcription_response_without_text_or_usage_reports_the_body(
+@pytest.mark.parametrize(
+    "payload",
+    [
+        {"model": "sensitive"},
+        {"duration": 2.0, "segments": [{"text": "sensitive"}]},
+        {"text": None, "usage": None, "note": "sensitive"},
+    ],
+)
+def test_transform_audio_transcription_response_without_text_or_usage_keeps_the_body_out_of_the_error(
     payload: dict[str, object],
 ) -> None:
     with pytest.raises(ValueError, match="Invalid response format") as exc_info:
         _transform_transcription_response(payload)
 
-    assert exc_info.value.args == (
-        "Invalid response format. Received response does not match the expected format. Got: ",
-        payload,
-    )
+    assert "sensitive" not in "".join(traceback.format_exception(exc_info.value))
 
 
 @pytest.mark.parametrize("payload", [["not", "an", "object"], "plain text", 7, True])

tests/unit/llms/xai/test_xai_audio_transcription_transformation.py

@@ -1,3 +1,5 @@
+import traceback
+
 import httpx
 import pytest
 
@@ -146,6 +148,23 @@ def test_transform_response_raises_on_error_status():
     assert "Incorrect API key provided" in exc.value.message
 
 
+@pytest.mark.parametrize(
+    "content",
+    [
+        b'{"text": ["sensitive"]}',
+        b'{"text": "ok", "words": [{"text": ["sensitive"]}]}',
+        b'{"text": "sensitive',
+    ],
+)
+def test_transform_response_keeps_a_malformed_body_out_of_the_error(content):
+    raw = httpx.Response(200, content=content, request=httpx.Request("POST", "https://api.x.ai/v1/stt"))
+
+    with pytest.raises(XAIAudioTranscriptionError, match="Error parsing xAI response") as exc:
+        CONFIG.transform_audio_transcription_response(raw_response=raw)
+
+    assert "sensitive" not in "".join(traceback.format_exception(exc.value))
+
+
 def test_transcription_routes_to_xai_stt(monkeypatch):
     monkeypatch.delenv("XAI_API_KEY", raising=False)
     monkeypatch.setattr(litellm, "xai_key", None)