BerriAI/litellm · error · VertexAIError

Received non-JSON response from Google Speech-to-Text: {raw_

Error message

Received non-JSON response from Google Speech-to-Text: {raw_response.text}

What it means

Guard when parsing a non-JSON response from Google Speech-to-Text: the recognition response body could not be decoded as JSON, so the raw text is embedded in the error. Indicates the endpoint returned HTML/error text instead of a recognition result.

Source

Thrown at litellm/llms/vertex_ai/audio_transcription/transformation.py:167

        request_body: Final = VertexSpeechToTextRecognizeRequest(
            config=VertexSpeechToTextRecognitionConfig(
                model=model.removeprefix("vertex_ai/"),
                languageCodes=language_codes,
                features=VertexSpeechToTextRecognitionFeatures(enableAutomaticPunctuation=True),
                autoDecodingConfig=VertexSpeechToTextAutoDecodingConfig(),
            ),
            content=base64.b64encode(processed_audio.file_content).decode("utf-8"),
        )
        return AudioTranscriptionRequestData(data=dict(request_body))

    def transform_audio_transcription_response(
        self,
        raw_response: Response,
    ) -> TranscriptionResponse:
        try:
            response_json: Final = raw_response.json()
        except ValueError:
            raise VertexAIError(
                status_code=raw_response.status_code,
                message=f"Received non-JSON response from Google Speech-to-Text: {raw_response.text}",
            )
        parsed: Final = VertexSpeechToTextRecognizeResponse.model_validate(response_json)
        transcripts: Final = tuple(
            result.alternatives[0].transcript
            for result in parsed.results
            if result.alternatives and result.alternatives[0].transcript
        )
        response: Final = TranscriptionResponse(text=" ".join(transcripts))
        response["task"] = "transcribe"
        detected_language: Final = next((result.languageCode for result in parsed.results if result.languageCode), None)
        if detected_language is not None:
            response["language"] = detected_language
        billed_duration = _parse_duration_seconds(parsed.metadata.totalBilledDuration if parsed.metadata else None)
        if billed_duration is not None:
            response["duration"] = billed_duration
        response._hidden_params = response_json

View on GitHub (pinned to 77b7c6c40c)

Solutions

1. The Speech-to-Text endpoint returned non-JSON; inspect raw_response.text.
2. Check for proxy/endpoint misconfiguration or an HTML error page.

Example fix

# log raw_response.text to see the actual content returned.

When it happens

Trigger: Triggered when Google Speech-to-Text returns a non-JSON response body.

Common situations: See trigger scenarios.

AI-assisted analysis of BerriAI/litellm@77b7c6c40c (2026-08-18). Data as JSON: /api/errors/2ec9a6baae6f9dbe. Report an issue: GitHub.