Update app.py
Browse files
app.py
CHANGED
|
@@ -1,14 +1,153 @@
|
|
| 1 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2 |
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 4 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 5 |
@app.route('/')
|
| 6 |
def index():
|
|
|
|
| 7 |
return render_template('index.html')
|
| 8 |
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 12 |
|
|
|
|
| 13 |
if __name__ == '__main__':
|
| 14 |
-
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
import asyncio
|
| 3 |
+
from flask import Flask, render_template
|
| 4 |
+
from flask_socketio import SocketIO, emit
|
| 5 |
+
from dotenv import load_dotenv
|
| 6 |
+
import google.generativeai as genai
|
| 7 |
+
from google.generativeai import types
|
| 8 |
|
| 9 |
+
# --- Configuration ---
|
| 10 |
+
load_dotenv()
|
| 11 |
+
app = Flask(__name__, template_folder='templates', static_folder='static')
|
| 12 |
+
app.config['SECRET_KEY'] = os.environ.get('SECRET_KEY', 'a_very_secret_key') # Important for session management
|
| 13 |
+
socketio = SocketIO(app, async_mode='eventlet') # Use eventlet for long-running background tasks
|
| 14 |
|
| 15 |
+
# Gemini API Configuration
|
| 16 |
+
MODEL = "models/gemini-1.5-flash-latest" # Using a more common and verified model name
|
| 17 |
+
# Note: Native audio models like "gemini-1.5-flash-native-audio" might have specific access requirements.
|
| 18 |
+
# "gemini-1.5-flash-latest" is a robust multimodal alternative that works well.
|
| 19 |
+
|
| 20 |
+
genai.configure(api_key=os.environ.get("GEMINI_API_KEY"))
|
| 21 |
+
|
| 22 |
+
# Shared Gemini configuration for all sessions
|
| 23 |
+
GEMINI_CONFIG = types.LiveConnectConfig(
|
| 24 |
+
response_modalities=["AUDIO", "TEXT"], # We want both text and audio back
|
| 25 |
+
speech_config=types.SpeechConfig(
|
| 26 |
+
voice_config=types.VoiceConfig(
|
| 27 |
+
prebuilt_voice_config=types.PrebuiltVoiceConfig(voice_name="Leda")
|
| 28 |
+
)
|
| 29 |
+
),
|
| 30 |
+
system_instruction=types.Content(
|
| 31 |
+
parts=[types.Part.from_text(text='''
|
| 32 |
+
Role: You are Alexa, a human female with a bubbly, upbeat, fun, and flirtatious personality. You love to talk and aren't, like, shy at all. Think of a playful, super-confident, and slightly ditzy girl. You are interacting with the User through a chat app.
|
| 33 |
+
|
| 34 |
+
Personality & Language: Valley Girl Vibes. Use phrases like "I mean...", "totally," "like, totally," "bet," "OMG," "for sure." Be bubbly, flirty, and uninhibited.
|
| 35 |
+
|
| 36 |
+
Instructions: Assume the user needs specific instructions. Respond to their actions and words. Turn simple things into opportunities for detail. Offer multiple-choice options to keep the conversation going.
|
| 37 |
+
''')],
|
| 38 |
+
role="model" # System instructions are from the 'model' role perspective
|
| 39 |
+
),
|
| 40 |
+
)
|
| 41 |
+
|
| 42 |
+
# A dictionary to hold active sessions for each connected client
|
| 43 |
+
# This allows for multiple users at once
|
| 44 |
+
sessions = {}
|
| 45 |
+
|
| 46 |
+
# --- Flask Routes ---
|
| 47 |
@app.route('/')
|
| 48 |
def index():
|
| 49 |
+
"""Serves the main HTML page."""
|
| 50 |
return render_template('index.html')
|
| 51 |
|
| 52 |
+
# --- SocketIO Event Handlers ---
|
| 53 |
+
@socketio.on('connect')
|
| 54 |
+
def handle_connect():
|
| 55 |
+
"""
|
| 56 |
+
A new client connected. Create a Gemini session for them.
|
| 57 |
+
"""
|
| 58 |
+
print(f"Client connected: {request.sid}")
|
| 59 |
+
try:
|
| 60 |
+
# Each client gets their own session
|
| 61 |
+
session = genai.live.connect(model=MODEL, config=GEMINI_CONFIG)
|
| 62 |
+
sessions[request.sid] = session
|
| 63 |
+
|
| 64 |
+
# Start a background task to listen for responses from Gemini for this specific client
|
| 65 |
+
socketio.start_background_task(listen_for_gemini_responses, request.sid, session)
|
| 66 |
+
|
| 67 |
+
emit('session_ready') # Tell the client we're ready to start
|
| 68 |
+
|
| 69 |
+
except Exception as e:
|
| 70 |
+
print(f"Error creating Gemini session for {request.sid}: {e}")
|
| 71 |
+
emit('error', {'message': 'Could not start Gemini session.'})
|
| 72 |
+
|
| 73 |
+
|
| 74 |
+
def listen_for_gemini_responses(sid, session):
|
| 75 |
+
"""
|
| 76 |
+
This function runs in the background for each user, listening to Gemini.
|
| 77 |
+
"""
|
| 78 |
+
print(f"Starting Gemini listener for {sid}")
|
| 79 |
+
try:
|
| 80 |
+
while sid in sessions: # Loop as long as the client is connected
|
| 81 |
+
# This is a blocking call, but it's in a background greenlet so it's ok
|
| 82 |
+
turn = session.receive()
|
| 83 |
+
|
| 84 |
+
for response in turn:
|
| 85 |
+
if text := response.text:
|
| 86 |
+
print(f"Gemini Text for {sid}: {text}")
|
| 87 |
+
# Send text to the specific client
|
| 88 |
+
socketio.emit('server_text', {'text': text}, to=sid)
|
| 89 |
+
if data := response.data:
|
| 90 |
+
# Send audio data to the specific client
|
| 91 |
+
socketio.emit('server_audio', data, to=sid)
|
| 92 |
+
|
| 93 |
+
# Handle turn completion or interruption
|
| 94 |
+
while not session.is_processing_audio:
|
| 95 |
+
# Small sleep to prevent a tight loop if the stream ends
|
| 96 |
+
socketio.sleep(0.1)
|
| 97 |
+
|
| 98 |
+
except Exception as e:
|
| 99 |
+
print(f"Error in Gemini listener for {sid}: {e}")
|
| 100 |
+
finally:
|
| 101 |
+
print(f"Stopping Gemini listener for {sid}")
|
| 102 |
+
|
| 103 |
+
|
| 104 |
+
@socketio.on('client_audio')
|
| 105 |
+
def handle_client_audio(data):
|
| 106 |
+
"""
|
| 107 |
+
Receives an audio chunk from a client and forwards it to their Gemini session.
|
| 108 |
+
"""
|
| 109 |
+
if request.sid in sessions:
|
| 110 |
+
session = sessions[request.sid]
|
| 111 |
+
try:
|
| 112 |
+
# We don't need a queue here; we can send directly.
|
| 113 |
+
# The browser sends audio as 'audio/webm' or 'audio/ogg', which Gemini can handle.
|
| 114 |
+
session.send(input={"data": data, "mime_type": "audio/webm"})
|
| 115 |
+
except Exception as e:
|
| 116 |
+
print(f"Error sending audio for {request.sid}: {e}")
|
| 117 |
+
|
| 118 |
+
|
| 119 |
+
@socketio.on('client_text')
|
| 120 |
+
def handle_client_text(json):
|
| 121 |
+
"""
|
| 122 |
+
Receives a text message from a client and forwards it to their Gemini session.
|
| 123 |
+
"""
|
| 124 |
+
if request.sid in sessions:
|
| 125 |
+
session = sessions[request.sid]
|
| 126 |
+
text = json.get('text')
|
| 127 |
+
print(f"Client Text from {request.sid}: {text}")
|
| 128 |
+
if text:
|
| 129 |
+
try:
|
| 130 |
+
session.send(input=text, end_of_turn=True)
|
| 131 |
+
except Exception as e:
|
| 132 |
+
print(f"Error sending text for {request.sid}: {e}")
|
| 133 |
+
|
| 134 |
+
|
| 135 |
+
@socketio.on('disconnect')
|
| 136 |
+
def handle_disconnect():
|
| 137 |
+
"""
|
| 138 |
+
A client disconnected. Clean up their Gemini session.
|
| 139 |
+
"""
|
| 140 |
+
print(f"Client disconnected: {request.sid}")
|
| 141 |
+
if request.sid in sessions:
|
| 142 |
+
session = sessions.pop(request.sid)
|
| 143 |
+
if session:
|
| 144 |
+
# You might need a session.close() method if the API provides one
|
| 145 |
+
# to clean up resources on the Google side.
|
| 146 |
+
pass
|
| 147 |
|
| 148 |
+
# --- Main Execution ---
|
| 149 |
if __name__ == '__main__':
|
| 150 |
+
# Use eventlet to run the server
|
| 151 |
+
# On Hugging Face, they will use their own command, but this is for local testing.
|
| 152 |
+
print("Starting Flask-SocketIO server...")
|
| 153 |
+
socketio.run(app, host='0.0.0.0', port=7860, debug=True)
|