Cost-efficient, real-timespeech-to-speech model

# The official OpenAI SDK — pointed at Higgs
client = AsyncOpenAI(
-   api_key=os.environ["OPENAI_API_KEY"],
+   api_key=os.environ["BOSON_API_KEY"],
+   base_url="https://api.boson.ai/v1",
)
 
-async with client.realtime.connect(model="gpt-realtime-2") as conn:
+async with client.realtime.connect(model="higgs-realtime") as conn:
    await conn.session.update(session={
+       "model": "higgs-realtime",  # required; the connect() arg is SDK-only
        "output_modalities": ["audio"],
        "audio": {
            "input": {"turn_detection": {"type": "server_vad"}},
            "output": {"voice": "default"},
        },
    })
 
    # stream mic audio in; server VAD handles turn-taking
    async for chunk in microphone():
        await conn.input_audio_buffer.append(
            audio=base64.b64encode(chunk).decode(),
        )
 
    # play assistant audio as it streams back
    async for event in conn:
        if event.type == "response.output_audio.delta":
            speaker.play(event.delta)