voice_agent.py

import os
from dotenv import load_dotenv

# Disable OpenAI fallback
os.environ["OPENAI_API_KEY"] = ""
os.environ["ANTHROPIC_API_KEY"] = ""
os.environ["COHERE_API_KEY"] = ""

from crewai import Agent, Task, Crew
from google import genai

load_dotenv()

client = genai.Client(api_key=os.getenv("GEMINI_API_KEY"))

class CrewVoiceAgent:
    def __init__(self):
        self.agent = Agent(
            role="Voice Assistant",
            goal="Respond clearly.",
            backstory="You are helpful.",
            llm=None
        )

        self.task = Task(
            description="Respond to speech.",
            expected_output="Audio",
            agent=self.agent,
            llm=None
        )

        self.crew = Crew(
            agents=[self.agent],
            tasks=[self.task],
            llm=None
        )

    async def handle_audio(self, audio_bytes: bytes):

        print("Received audio bytes:", len(audio_bytes))

        response = client.models.generate_content(
            model="gemini-2.5-flash-native-audio-dialog",
            contents=[
                {
                    "mime_type": "audio/webm",
                    "data": audio_bytes
                }
            ]
        )

        print("GEMINI RESPONSE FIELDS:", response)
        print("HAS AUDIO:", hasattr(response, "audio"))

        if hasattr(response, "audio"):
            print("AUDIO OBJECT:", response.audio)

        # Try direct
        try:
            if response.audio and response.audio.data:
                print("Returning audio bytes:", len(response.audio.data))
                return response.audio.data
        except Exception as e:
            print("Error reading audio:", e)

        print("NO AUDIO RETURNED")
        return b""