fix(audio): handle plain-text response body for response_format=text

litellm/llms/openai/transcriptions/whisper_transformation.py

@@ -1,3 +1,4 @@
+import json
 from typing import List, Optional, Union
 
 from httpx import Headers, Response
@@ -131,10 +132,8 @@ class OpenAIWhisperAudioTranscriptionConfig(BaseAudioTranscriptionConfig):
     ) -> TranscriptionResponse:
         try:
             raw_response_json = raw_response.json()
-        except Exception as e:
-            raise ValueError(
-                f"Error transforming response to json: {str(e)}\nResponse: {raw_response.text}"
-            )
+        except json.JSONDecodeError:
+            return TranscriptionResponse(text=raw_response.text)
 
         if any(
             key in raw_response_json

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

@@ -1,8 +1,11 @@
 """
-Tests for OpenAIWhisperAudioTranscriptionConfig.transform_audio_transcription_request.
+Tests for OpenAIWhisperAudioTranscriptionConfig.transform_audio_transcription_request
+and transform_audio_transcription_response.
 """
 
 import io
+import json
+from unittest.mock import MagicMock
 
 from litellm.llms.openai.transcriptions.whisper_transformation import (
     OpenAIWhisperAudioTranscriptionConfig,
@@ -41,3 +44,30 @@ class TestWhisperTransformRequestResponseFormat:
         """verbose_json explicitly set by the caller stays as-is."""
         data = self._transform({"response_format": "verbose_json"})
         assert data["response_format"] == "verbose_json"
+
+
+class TestWhisperTransformResponse:
+    def _make_response(self, *, text: str, is_json: bool):
+        mock = MagicMock()
+        if is_json:
+            mock.json.return_value = {"text": text}
+        else:
+            mock.json.side_effect = json.JSONDecodeError("", "", 0)
+            mock.text = text
+        return mock
+
+    def test_parses_json_response(self):
+        """JSON body (verbose_json or json format) is parsed into TranscriptionResponse."""
+        config = OpenAIWhisperAudioTranscriptionConfig()
+        result = config.transform_audio_transcription_response(
+            self._make_response(text="Hello world", is_json=True)
+        )
+        assert result.text == "Hello world"
+
+    def test_parses_plain_text_response(self):
+        """Plain-text body (response_format=text) is returned as TranscriptionResponse without error."""
+        config = OpenAIWhisperAudioTranscriptionConfig()
+        result = config.transform_audio_transcription_response(
+            self._make_response(text="Four score and seven years ago", is_json=False)
+        )
+        assert result.text == "Four score and seven years ago"