Support non-streaming Gemini 3.1 Flash TTS via google-genai

[Anrahya]

Problem Statement

Pipecat has GeminiTTSService, but it currently targets Google Cloud Text-to-Speech streaming models such as gemini-2.5-flash-tts / gemini-2.5-pro-tts via texttospeech_v1.StreamingSynthesize.

Google's Gemini speech generation docs now show Gemini TTS through google-genai generate_content with response_modalities=[AUDIO], including gemini-3.1-flash-tts-preview. The docs also state that TTS does not support streaming, so this seems like a different integration shape than the existing streaming service.

Proposed Solution

Would maintainers be open to a focused non-streaming Gemini TTS service for this path?

Possible shape:

• Use google-genai client.models.generate_content(...)
• Support Gemini API speech generation models such as gemini-3.1-flash-tts-preview
• Emit returned audio as TTSAudioRawFrame after the response completes
• Document that this is not intended as a low-latency realtime TTS path
• Include a minimal mocked test and example

Alternative Solutions

This may be related to #4351, but that PR is Vertex/aiplatform-focused and uses stream_generate_content. This issue is specifically about the non-streaming Gemini API TTS flow documented here:

https://ai.google.dev/gemini-api/docs/speech-generation

Additional Context

Happy to help implement this if the direction sounds reasonable.

Would you be willing to help implement this feature?

• Yes, I'd like to contribute

[haayhappen]

gemini-3.1-flash-tts-preview is supported by pipecats GoogleHttpTTSService - but it has absurd latencies and does not allow setting regional endpoints yet

[Anrahya]

I might be missing something, but I don’t think this is the same thing.

From what I can see, GoogleHttpTTSService is using Cloud Text-to-Speech synthesize_speech(...), not the Gemini 3.1 Flash TTS Preview API path with google-genai / models.generate_content / response_modalities: ["AUDIO"].

GeminiTTSService also seems to be the cloud TTS streaming path around Gemini 2.5 TTS, not the non-streaming 3.1 preview flow.

If gemini-3.1-flash-tts-preview really works through GoogleHttpTTSService, can you point me to the example or code path? I’d rather not duplicate existing work, but from the repo it does not look wired yet.