Google-adk live, audio pass to sub agents · AI Developer Accelerator

Google-adk live, audio pass to sub agents

Hello there,

Ive a question maybe some1 already tried:

I want to create root agent who will have agent as a tool (2x) and I wondering if there is possibility to pass input audio to these tools rather then only TEXT

```python

affect_alignment_agent = Agent(

model="gemini-2.0-flash-exp",

name="affect_alignment_agent",

description="Affect alignment agent",

instruction=affect_aligment,

)

engagement_agent = Agent(

model="gemini-2.0-flash-exp",

name="engagement_agent",

description="Engagement agent",

instruction=engagement,

affect_alignment_tool = agent_tool.AgentTool(agent=affect_alignment_agent) # Wrap the agent

engagement_tool = agent_tool.AgentTool(agent=engagement_agent) # Wrap the agent

root_agent = Agent(

model="gemini-2.0-flash-exp",

name="root_agent",

description="Root agent that coordinates the affect alignment and engagement agents",

instruction="""

You are responsible for coordinating the affect alignment and engagement agents tools.

Send user input to the tools and wait for their responses.

Combine the responses from the tools and send the final response to the user.

""",

tools=[affect_alignment_tool, engagement_tool],

)

```

what I can see from the log:

```

Raw response:

{"candidates":[{"content":{"parts":[{"text":"```text\nDetected Emotional State: Neutral (valence = 0.5, arousal = 0.5). Delta from baseline: No significant change.\nStrategy: Maintain a neutral and friendly tone as the user's emotional state is currently stable.\n\nResponse: \"Hello! I'm doing well, thank you for asking. How can I assist you today?\"\n```\n"}],"role":"model"},"finish_reason":"STOP","avg_logprobs":-0.13121765087812376}],"model_version":"gemini-2.0-flash-exp","usage_metadata":{"candidates_token_count":78,"candidates_tokens_details":[{"modality":"TEXT","token_count":78}],"prompt_token_count":1111,"prompt_tokens_details":[{"modality":"TEXT","token_count":1111}],"total_token_count":1189},"automatic_function_calling_history":[]}

```

Im not pretty sure that these response was base on AUDIO. It looks like it is base on TEXT only.

Is this possible to create custom runner with `artifact_service=artifact_service` and base on that other agents will be able to access AUDIO ?

2 comments