Spaces:
Sleeping
Sleeping
Update voice_agent.py
Browse files- voice_agent.py +8 -6
voice_agent.py
CHANGED
|
@@ -9,28 +9,31 @@ genai.configure(api_key=os.getenv("GEMINI_API_KEY"))
|
|
| 9 |
|
| 10 |
class CrewVoiceAgent:
|
| 11 |
def __init__(self):
|
|
|
|
|
|
|
| 12 |
self.agent = Agent(
|
| 13 |
role="Voice Assistant",
|
| 14 |
goal="Respond to user queries through voice with clarity and accuracy.",
|
| 15 |
-
backstory="You are a friendly
|
|
|
|
| 16 |
)
|
| 17 |
|
| 18 |
self.task = Task(
|
| 19 |
description="Interpret the user's speech and generate a clear, helpful spoken response.",
|
| 20 |
-
agent=self.agent
|
|
|
|
| 21 |
)
|
| 22 |
|
| 23 |
self.crew = Crew(
|
| 24 |
agents=[self.agent],
|
| 25 |
-
tasks=[self.task]
|
|
|
|
| 26 |
)
|
| 27 |
|
| 28 |
self.model = genai.GenerativeModel("gemini-2.5-flash-native-audio-dialog")
|
| 29 |
|
| 30 |
async def handle_audio(self, audio_bytes: bytes):
|
| 31 |
-
"""Convert microphone input β CrewAI β Gemini Audio β audio output"""
|
| 32 |
|
| 33 |
-
# Convert raw audio bytes to base64
|
| 34 |
audio_b64 = base64.b64encode(audio_bytes).decode()
|
| 35 |
|
| 36 |
response = self.model.generate_content(
|
|
@@ -42,7 +45,6 @@ class CrewVoiceAgent:
|
|
| 42 |
}
|
| 43 |
)
|
| 44 |
|
| 45 |
-
# Gemini returns audio chunks
|
| 46 |
audio_output = b""
|
| 47 |
for chunk in response.audio:
|
| 48 |
audio_output += chunk.data
|
|
|
|
| 9 |
|
| 10 |
class CrewVoiceAgent:
|
| 11 |
def __init__(self):
|
| 12 |
+
|
| 13 |
+
# IMPORTANT FIX: Disable CrewAI's default LLM selection
|
| 14 |
self.agent = Agent(
|
| 15 |
role="Voice Assistant",
|
| 16 |
goal="Respond to user queries through voice with clarity and accuracy.",
|
| 17 |
+
backstory="You are a friendly AI voice agent.",
|
| 18 |
+
llm=None # π THIS LINE IS CRITICAL
|
| 19 |
)
|
| 20 |
|
| 21 |
self.task = Task(
|
| 22 |
description="Interpret the user's speech and generate a clear, helpful spoken response.",
|
| 23 |
+
agent=self.agent,
|
| 24 |
+
llm=None # π Disable LLM here too
|
| 25 |
)
|
| 26 |
|
| 27 |
self.crew = Crew(
|
| 28 |
agents=[self.agent],
|
| 29 |
+
tasks=[self.task],
|
| 30 |
+
llm=None # π Prevent fallback to OpenAI everywhere
|
| 31 |
)
|
| 32 |
|
| 33 |
self.model = genai.GenerativeModel("gemini-2.5-flash-native-audio-dialog")
|
| 34 |
|
| 35 |
async def handle_audio(self, audio_bytes: bytes):
|
|
|
|
| 36 |
|
|
|
|
| 37 |
audio_b64 = base64.b64encode(audio_bytes).decode()
|
| 38 |
|
| 39 |
response = self.model.generate_content(
|
|
|
|
| 45 |
}
|
| 46 |
)
|
| 47 |
|
|
|
|
| 48 |
audio_output = b""
|
| 49 |
for chunk in response.audio:
|
| 50 |
audio_output += chunk.data
|