[Bug]: Gemini TTS model returning 400 with 'audio' parameters

[AyrennC]

What happened?

Using gemini-2.5-flash-preview-tts and gemini-2.5-pro-preview-tts through LiteLLM proxy always return 400 when an audio parameter is provided.

Here is the API body to /v1/chat/completions

{
  "model": "gemini/gemini-2.5-flash-preview-tts",
  "messages": [
    {
      "role": "user",
      "content": "You find yourself stranded on an island."
    }
  ],
  "stream": true,
  "temperature": "0.7",
  "frequency_penalty": "0",
  "modalities": [
    "audio"
  ],
  "audio": {
    "voice": "Zephyr",
    "format": "wav"
  }
}

Relevant log output

Failed to generate narration: 400 litellm.UnsupportedParamsError: gemini does not support parameters: ['audio'], for model=gemini-2.5-flash-preview-tts. To drop these, set `litellm.drop_params=True` or for proxy:

`litellm_settings:
 drop_params: true`
. 
 If you want to use these params dynamically send allowed_openai_params=['audio'] in your request.

Are you a ML Ops Team?

No

What LiteLLM version are you on ?

v1.71.1

Twitter / LinkedIn details

No response

[chipsandcake]

Also experiencing the same issue. Gemini TTS models only support the audio modality, not being able to provide audio parameters effectively renders these models completely unusable.

Would greatly appreciate it if anyone from the LiteLLM team is able to chime in!

[AyrennC]

Submitted PR to add audio parameter support to gemini tts models #11287