Granular Per-Agent Speech Configuration

[qyuo]

Is your feature request related to a problem?
In multi-agent conversational applications, distinguishing between different agents is crucial for user experience. Assigning a unique voice to each agent provides clear audio cues, helping users identify which persona is speaking and making the interaction more intuitive.

Currently, the speech configuration is set globally for the entire session. This prevents developers from defining per-agent voices, which is a significant limitation, especially in bidirectional streaming contexts.

Describe the solution you'd like
I propose adding an optional speech_config parameter to the individual agent's configuration object. This would allow developers to specify unique voice and other speech settings on a per-agent basis.

Example Implementation:

llm_agent.LlmAgent(
    model=models.Gemini(
        model="gemini-live-2.5-flash-preview",
        retry_options=types.HttpRetryOptions(initial_delay=1, attempts=2),
    ),
    name="recommendation_agent",
    description="Recommends Fantasy Football Picks",
    instruction="Provide data-driven draft recommendations",
    speech_config=types.SpeechConfig(
        voice_config=types.VoiceConfig(
            prebuilt_voice_config=types.PrebuiltVoiceConfig(
                voice_name="Kore",
            )
        )
    ),
)