Gemini 2 flash Realtime Can't generate audio in Russian (generates some noise instead)

Google AI Studio
1

Gemini 2 Realtime Can’t generate audio in Russian (generates some noise instead - like words reversed or something)
I tried
gemini-2.0-flash-exp (in Ai Studio or Vertex , web or api)
gemini-2.0-flash-001 (in Vertex)
gemini-2.0-flash (in Ai Studio or Vertex , web or api)

2

Gemini 2.0 currently does not support any language other than English.

4

Hm… no, it speaks Spanish, Italiano, Thai - all fine. But no luck with Russian or Ukrainian

1 Like
6

I think it’s because Russian and Ukrainian languages are Slavic. Maybe the Gemini team doesn’t gather enough data for proper training for this group of languages ._.
Perhaps it’s also worth trying Polish, Czech, Bulgarian, and other Slavic languages.

8

Hello,
I would like to set the language for the gemini-2.0-flash-001 realtime LLM (only audio, not images).

by passing this websocket message


    setup_message = {
        "setup": {
            "model": "models/gemini-2.0-flash-live-001",
            # "language_codes": ["es-ES"],
            # "languages": ["es-ES"],
            "system_instruction": {
                "role": "system",
                "parts": [{"text": SYSTEM_MESSAGE}],
            },
        }
    }

    await llm_ws.send(json.dumps(setup_message))

but “language_codes” or “languages” paramethers are not accepted

I can I force
gemini-2.0-flash-001 to use Spanish?

With out this setting the model (gemini-2.0-flash-001) can still interactuate with the model, but the accent it has is terrible.

In UI of gemini-2.0-flash-001 https://aistudio.google.com/live the language can be selected, but it is not the case in the code ( Get code section does NOT reflect the selected language :frowning: ).