Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,203 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
import mediapipe as mp
|
3 |
+
import numpy as np
|
4 |
+
import base64
|
5 |
+
import io
|
6 |
+
import PIL.Image
|
7 |
+
import asyncio
|
8 |
+
import os
|
9 |
+
from google import genai
|
10 |
+
from streamlit_webrtc import webrtc_streamer
|
11 |
+
import av
|
12 |
+
import pyaudio
|
13 |
+
from mediapipe.tasks import python
|
14 |
+
from mediapipe.tasks.python import vision
|
15 |
+
|
16 |
+
# Configuration
|
17 |
+
FORMAT = pyaudio.paInt16
|
18 |
+
CHANNELS = 1
|
19 |
+
SEND_SAMPLE_RATE = 16000
|
20 |
+
RECEIVE_SAMPLE_RATE = 24000
|
21 |
+
CHUNK_SIZE = 1024
|
22 |
+
|
23 |
+
# Initialize Genai client
|
24 |
+
genai.configure(api_key=os.getenv("GOOGLE_API_KEY"))
|
25 |
+
client = genai.Client(http_options={"api_version": "v1alpha"})
|
26 |
+
MODEL = "models/gemini-2.0-flash-exp"
|
27 |
+
CONFIG = {"generation_config": {"response_modalities": ["AUDIO"]}}
|
28 |
+
|
29 |
+
class AudioProcessor:
|
30 |
+
def __init__(self):
|
31 |
+
self.audio = pyaudio.PyAudio()
|
32 |
+
self.stream = None
|
33 |
+
self.audio_queue = asyncio.Queue()
|
34 |
+
|
35 |
+
def start_stream(self):
|
36 |
+
mic_info = self.audio.get_default_input_device_info()
|
37 |
+
self.stream = self.audio.open(
|
38 |
+
format=FORMAT,
|
39 |
+
channels=CHANNELS,
|
40 |
+
rate=SEND_SAMPLE_RATE,
|
41 |
+
input=True,
|
42 |
+
input_device_index=mic_info["index"],
|
43 |
+
frames_per_buffer=CHUNK_SIZE,
|
44 |
+
)
|
45 |
+
|
46 |
+
def stop_stream(self):
|
47 |
+
if self.stream:
|
48 |
+
self.stream.stop_stream()
|
49 |
+
self.stream.close()
|
50 |
+
self.stream = None
|
51 |
+
|
52 |
+
class VideoProcessor:
|
53 |
+
def __init__(self):
|
54 |
+
self.frame_queue = asyncio.Queue(maxsize=5)
|
55 |
+
self.mp_draw = mp.solutions.drawing_utils
|
56 |
+
self.mp_face_detection = mp.solutions.face_detection
|
57 |
+
self.face_detection = self.mp_face_detection.FaceDetection(
|
58 |
+
min_detection_confidence=0.5)
|
59 |
+
|
60 |
+
def video_frame_callback(self, frame):
|
61 |
+
# Convert the frame to RGB
|
62 |
+
img = frame.to_ndarray(format="rgb24")
|
63 |
+
|
64 |
+
# Process the frame with MediaPipe
|
65 |
+
results = self.face_detection.process(img)
|
66 |
+
|
67 |
+
# Draw face detection annotations if faces are detected
|
68 |
+
if results.detections:
|
69 |
+
for detection in results.detections:
|
70 |
+
self.mp_draw.draw_detection(img, detection)
|
71 |
+
|
72 |
+
# Convert to PIL Image
|
73 |
+
pil_img = PIL.Image.fromarray(img)
|
74 |
+
pil_img.thumbnail([1024, 1024])
|
75 |
+
|
76 |
+
# Prepare frame data for Gemini
|
77 |
+
image_io = io.BytesIO()
|
78 |
+
pil_img.save(image_io, format="jpeg")
|
79 |
+
image_io.seek(0)
|
80 |
+
|
81 |
+
frame_data = {
|
82 |
+
"mime_type": "image/jpeg",
|
83 |
+
"data": base64.b64encode(image_io.read()).decode()
|
84 |
+
}
|
85 |
+
|
86 |
+
try:
|
87 |
+
self.frame_queue.put_nowait(frame_data)
|
88 |
+
except asyncio.QueueFull:
|
89 |
+
pass
|
90 |
+
|
91 |
+
return av.VideoFrame.from_ndarray(img, format="rgb24")
|
92 |
+
|
93 |
+
def __del__(self):
|
94 |
+
# Cleanup MediaPipe resources
|
95 |
+
if hasattr(self, 'face_detection'):
|
96 |
+
self.face_detection.close()
|
97 |
+
|
98 |
+
def initialize_session_state():
|
99 |
+
if 'audio_processor' not in st.session_state:
|
100 |
+
st.session_state.audio_processor = AudioProcessor()
|
101 |
+
if 'video_processor' not in st.session_state:
|
102 |
+
st.session_state.video_processor = VideoProcessor()
|
103 |
+
if 'session' not in st.session_state:
|
104 |
+
st.session_state.session = None
|
105 |
+
if 'messages' not in st.session_state:
|
106 |
+
st.session_state.messages = []
|
107 |
+
|
108 |
+
def display_chat_messages():
|
109 |
+
for message in st.session_state.messages:
|
110 |
+
with st.chat_message(message["role"]):
|
111 |
+
st.markdown(message["content"])
|
112 |
+
|
113 |
+
def main():
|
114 |
+
st.title("Gemini Interactive Assistant")
|
115 |
+
|
116 |
+
# Initialize session state
|
117 |
+
initialize_session_state()
|
118 |
+
|
119 |
+
# Sidebar configuration
|
120 |
+
st.sidebar.title("Settings")
|
121 |
+
input_mode = st.sidebar.radio(
|
122 |
+
"Input Mode",
|
123 |
+
["Text Only", "Audio + Video", "Audio Only"]
|
124 |
+
)
|
125 |
+
|
126 |
+
# Enable face detection option
|
127 |
+
enable_face_detection = st.sidebar.checkbox("Enable Face Detection", value=True)
|
128 |
+
|
129 |
+
if enable_face_detection:
|
130 |
+
detection_confidence = st.sidebar.slider(
|
131 |
+
"Face Detection Confidence",
|
132 |
+
min_value=0.0,
|
133 |
+
max_value=1.0,
|
134 |
+
value=0.5,
|
135 |
+
step=0.1
|
136 |
+
)
|
137 |
+
st.session_state.video_processor.face_detection = (
|
138 |
+
st.session_state.video_processor.mp_face_detection.FaceDetection(
|
139 |
+
min_detection_confidence=detection_confidence
|
140 |
+
)
|
141 |
+
)
|
142 |
+
|
143 |
+
# Display chat history
|
144 |
+
display_chat_messages()
|
145 |
+
|
146 |
+
# Main interaction area
|
147 |
+
if input_mode == "Text Only":
|
148 |
+
user_input = st.chat_input("Your message")
|
149 |
+
if user_input:
|
150 |
+
# Add user message to chat
|
151 |
+
st.session_state.messages.append({"role": "user", "content": user_input})
|
152 |
+
with st.chat_message("user"):
|
153 |
+
st.markdown(user_input)
|
154 |
+
|
155 |
+
async def send_message():
|
156 |
+
async with client.aio.live.connect(model=MODEL, config=CONFIG) as session:
|
157 |
+
await session.send(user_input, end_of_turn=True)
|
158 |
+
turn = session.receive()
|
159 |
+
async for response in turn:
|
160 |
+
if text := response.text:
|
161 |
+
# Add assistant response to chat
|
162 |
+
st.session_state.messages.append(
|
163 |
+
{"role": "assistant", "content": text}
|
164 |
+
)
|
165 |
+
with st.chat_message("assistant"):
|
166 |
+
st.markdown(text)
|
167 |
+
|
168 |
+
asyncio.run(send_message())
|
169 |
+
|
170 |
+
else:
|
171 |
+
# Video stream setup
|
172 |
+
if input_mode == "Audio + Video":
|
173 |
+
ctx = webrtc_streamer(
|
174 |
+
key="gemini-stream",
|
175 |
+
video_frame_callback=st.session_state.video_processor.video_frame_callback,
|
176 |
+
rtc_configuration={"iceServers": [{"urls": ["stun:stun.l.google.com:19302"]}]},
|
177 |
+
media_stream_constraints={"video": True, "audio": True},
|
178 |
+
)
|
179 |
+
|
180 |
+
# Audio controls
|
181 |
+
col1, col2 = st.columns(2)
|
182 |
+
with col1:
|
183 |
+
if st.button("Start Recording", type="primary"):
|
184 |
+
st.session_state.audio_processor.start_stream()
|
185 |
+
st.session_state['recording'] = True
|
186 |
+
|
187 |
+
with col2:
|
188 |
+
if st.button("Stop Recording", type="secondary"):
|
189 |
+
st.session_state.audio_processor.stop_stream()
|
190 |
+
st.session_state['recording'] = False
|
191 |
+
|
192 |
+
async def process_audio_stream():
|
193 |
+
while st.session_state.get('recording', False):
|
194 |
+
if st.session_state.audio_processor.stream:
|
195 |
+
data = st.session_state.audio_processor.stream.read(CHUNK_SIZE)
|
196 |
+
await st.session_state.audio_processor.audio_queue.put({
|
197 |
+
"data": data,
|
198 |
+
"mime_type": "audio/pcm"
|
199 |
+
})
|
200 |
+
await asyncio.sleep(0.1)
|
201 |
+
|
202 |
+
if __name__ == "__main__":
|
203 |
+
main()
|