!IMPORTANT

I've used this project with openai-agents-python package source code customized.

So after i'm gonna provide code with fixed source code that you should directly use code with import.

Following below openai agent SDK

⟶ A companion to OpenAI_S2S, purpose-built to implement a chained architecture.

Setup

uv sync
uv run python main.py

PLEASE READ BELOW LINE (main.py 196)

# patch_stt_event_handler()         -> un-comment this : if you are using without modifying '.venv/lib/python3.12/site-packages/agents/voice/models/openai_stt.py' _handle_events

Summuries

1. Agent definition

Agent = name / instructions / handoff_description

math_tutor_agent = Agent(
    name="Math Tutor",
    handoff_description="Specialist agent for math questions",
    instructions="You provide help with math problems. Explain your reasoning at each step and include examples",
)

• Handoff ⊂ Tool
• Tool: function-calling, hosted tools(retrieval/web/computer use), agents-as-tools

2. Voice Agent

Voice angent workflow SDK : SingleAgentVoiceWorkflow

PCM : Pulse Code Modulation

Results

14:15:40 [LIFECYCLE] {"event": "turn_started", "type": "voice_stream_event_lifecycle"}
14:15:43 [LIFECYCLE] {"event": "turn_ended", "type": "voice_stream_event_lifecycle"}
14:15:43 [LIFECYCLE] {"event": "session_ended", "type": "voice_stream_event_lifecycle"}

References

• handoff : https://openai.github.io/openai-agents-python/ref/agent/?utm_source=chatgpt.com
• voice agent : https://openai.github.io/openai-agents-python/voice/quickstart/