BerriAI/litellm · error · Exception

Unmapped response type. Got type: {type(result)}

Error message

Unmapped response type. Got type: {type(result)}

What it means

Raised in SpeechToCompletionBridgeHandler.speech (handler.py:123) when the internal litellm.completion() call returns something that is not a ModelResponse — e.g. a streaming iterator (CustomStreamWrapper) or an executor/future. The bridge can only convert a complete ModelResponse containing an audio part into HttpxBinaryResponseContent, so any other result type is treated as a programming/configuration error.

Source

Thrown at litellm/endpoints/speech/speech_to_completion_bridge/handler.py:123

            input=input,
            optional_params=optional_params,
            litellm_params=litellm_params,
            headers=headers,
            litellm_logging_obj=logging_obj,
            custom_llm_provider=custom_llm_provider,
            voice=voice,
        )

        result: Final = completion(
            **request_data,
        )

        if isinstance(result, ModelResponse):
            return self.transformation_handler.transform_response(
                model_response=result,
            )
        else:
            raise Exception(f"Unmapped response type. Got type: {type(result)}")


speech_to_completion_bridge_handler = SpeechToCompletionBridgeHandler()

View on GitHub (pinned to 6c2dcb801b)

Solutions

1. Remove stream=True from the speech/TTS call — the bridge requires a non-streaming response
2. Inspect optional_params and litellm_params forwarded to the bridge and strip any 'stream' key
3. If you patched litellm.completion in tests, make the mock return a ModelResponse instance

Example fix

# before
resp = litellm.audio_speech(model="gpt-4o-audio-preview", input="hi", voice="alloy", stream=True)

# after
resp = litellm.audio_speech(model="gpt-4o-audio-preview", input="hi", voice="alloy")

optional_params = {k: v for k, v in (optional_params or {}).items() if k != "stream"}
# ensure the bridge receives a non-streaming completion call

from litellm.types.utils import ModelResponse

def is_model_response(result: object) -> bool:
    return isinstance(result, ModelResponse)

try:
    audio = litellm.audio_speech(model=model, input=text, voice=voice)
except Exception as e:
    if "Unmapped response type" in str(e):
        logger.error("completion() returned %s; disable streaming for speech calls", e)
        raise

When it happens

Trigger: The transformed request data includes stream=True (optional_params leaking a stream flag into the completion call), so completion() returns a stream wrapper instead of ModelResponse; or a mocked/patched completion returning a different type.

Common situations: Passing stream=True (or a router/deployment default that sets stream) to an audio/speech call routed through the bridge; providers or custom callbacks that alter the completion return type; tests monkeypatching litellm.completion with a plain dict.

Related errors

• prompt_characters must be provided for tts calls. prompt_cha
• model is required
• input is required
• logging_obj is required
• No audio part found in the response

AI-assisted analysis of BerriAI/litellm@6c2dcb801b (2026-08-15). Data as JSON: /api/errors/2b75b5b5791ecf61. Report an issue: GitHub.