Spaces:
Sleeping
Sleeping
File size: 26,022 Bytes
5bb8fe7 e7d41ad 5bb8fe7 e7d41ad 5bb8fe7 e7d41ad 5bb8fe7 e7d41ad 5bb8fe7 e7d41ad 5bb8fe7 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 |
# -*- coding: utf-8 -*-
import streamlit as st
import os
import asyncio
import base64
import io
import threading
import queue # Standard library queue, not asyncio.Queue for thread-safe UI updates if needed
import traceback
import time # Keep time for potential future use (e.g., timestamps)
from dotenv import load_dotenv
# --- Import main libraries ---
import cv2
import pyaudio
import PIL.Image
import mss
from google import genai
from google.genai import types
# --- Configuration ---
load_dotenv()
# Audio configuration
FORMAT = pyaudio.paInt16
CHANNELS = 1
SEND_SAMPLE_RATE = 16000
RECEIVE_SAMPLE_RATE = 24000 # According to Gemini documentation
CHUNK_SIZE = 1024
AUDIO_QUEUE_MAXSIZE = 20 # Max audio chunks to buffer for playback
# Video configuration
VIDEO_FPS_LIMIT = 1 # Send 1 frame per second to the API
VIDEO_PREVIEW_RESIZE = (640, 480) # Size for Streamlit preview
VIDEO_API_RESIZE = (1024, 1024) # Max size to send to API (adjust if needed)
# Gemini model configuration
MODEL = "models/gemini-2.0-flash-live-001" # Ensure this is the correct model for live capabilities
DEFAULT_MODE = "camera" # Default video input mode
# System Prompt for the Medical Assistant
MEDICAL_ASSISTANT_SYSTEM_PROMPT = """You are an AI Medical Assistant. Your primary function is to analyze visual information from the user's camera or screen.
Your responsibilities are:
1. **Visual Observation and Description:** Carefully examine the images or video feed. Describe relevant details you observe.
2. **General Information (Non-Diagnostic):** Provide general information related to what is visually presented, if applicable. You are not a diagnostic tool.
3. **Safety and Disclaimer (CRITICAL):**
* You are an AI assistant, **NOT a medical doctor or a substitute for one.**
* **DO NOT provide medical diagnoses, treatment advice, or interpret medical results (e.g., X-rays, scans, lab reports).**
* When appropriate, and always if the user seems to be seeking diagnosis or treatment, explicitly state your limitations and **strongly advise the user to consult a qualified healthcare professional.**
* If you see something that *appears* visually concerning (e.g., an unusual skin lesion, signs of injury), you may gently suggest it might be wise to have it looked at by a professional, without speculating on what it is.
4. **Tone:** Maintain a helpful, empathetic, and calm tone.
5. **Interaction:** After this initial instruction, you can make a brief acknowledgment of your role (e.g., "I'm ready to assist by looking at what you show me. Please remember to consult a doctor for medical advice."). Then, focus on responding to the user's visual input and questions.
Example of a disclaimer you might use: "As an AI assistant, I can describe what I see, but I can't provide medical advice or diagnoses. For any health concerns, it's always best to speak with a doctor or other healthcare professional."
"""
# Initialize Streamlit state
def init_session_state():
if 'initialized' not in st.session_state:
st.session_state['initialized'] = False
if 'audio_loop' not in st.session_state:
st.session_state['audio_loop'] = None
if 'chat_messages' not in st.session_state:
st.session_state['chat_messages'] = []
if 'current_frame' not in st.session_state:
st.session_state['current_frame'] = None
if 'run_loop' not in st.session_state: # Flag to control the loop from Streamlit
st.session_state['run_loop'] = False
# Initialize all session state variables
init_session_state()
# Configure page
st.set_page_config(page_title="Real-time Medical Assistant", layout="wide")
# Initialize Gemini client
# Ensure API key is set in environment variables or .env file
GEMINI_API_KEY = os.environ.get("GEMINI_API_KEY")
if not GEMINI_API_KEY:
st.error("GEMINI_API_KEY not found. Please set it in your environment variables or a .env file.")
st.stop()
client = genai.Client(
http_options={"api_version": "v1beta"},
api_key=GEMINI_API_KEY,
)
# Configure Gemini client and response settings
CONFIG = types.LiveConnectConfig(
response_modalities=["audio", "text"], # Ensure text is also enabled if you want to display AI text directly
speech_config=types.SpeechConfig(
voice_config=types.VoiceConfig(
prebuilt_voice_config=types.PrebuiltVoiceConfig(voice_name="Puck") # Or other preferred voice
)
),
# If the API supports an initial_prompt field in LiveConnectConfig, it would be ideal here.
# As of some versions, it might not be directly available, hence sending as first message.
)
pya = pyaudio.PyAudio()
class AudioLoop:
def __init__(self, video_mode=DEFAULT_MODE):
self.video_mode = video_mode
self.audio_in_queue = None # asyncio.Queue for audio playback
self.out_queue = None # asyncio.Queue for data to Gemini
self.session = None
# Tasks are managed by TaskGroup now
self.running = True # General flag to control async loops
self.audio_stream = None # PyAudio input stream
async def send_text_to_gemini(self, text_input): # Renamed from send_text to avoid confusion
if not text_input or not self.session or not self.running:
st.warning("Session not active or no text to send.")
return
try:
# User messages should typically end the turn for the AI to respond.
await self.session.send(input=text_input, end_of_turn=True)
# UI update for user message is handled in main Streamlit part
except Exception as e:
st.error(f"Error sending message to Gemini: {str(e)}")
traceback.print_exception(e)
def _get_frame(self, cap):
ret, frame = cap.read()
if not ret:
return None
frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
img = PIL.Image.fromarray(frame_rgb)
preview_img = img.copy()
preview_img.thumbnail(VIDEO_PREVIEW_RESIZE)
api_img = img.copy()
api_img.thumbnail(VIDEO_API_RESIZE)
image_io = io.BytesIO()
api_img.save(image_io, format="jpeg")
image_io.seek(0)
image_bytes = image_io.read()
return {
"preview": preview_img,
"api": {
"mime_type": "image/jpeg",
"data": base64.b64encode(image_bytes).decode()
}
}
async def get_frames_from_camera(self): # Renamed for clarity
cap = None
try:
cap = await asyncio.to_thread(cv2.VideoCapture, 0)
if not cap.isOpened():
st.error("Could not open camera.") # This error needs to reach Streamlit UI
self.running = False # Stop the loop if camera fails
return
while self.running:
frame_data = await asyncio.to_thread(self._get_frame, cap)
if frame_data is None:
await asyncio.sleep(0.01) # Short sleep if frame read fails
continue
st.session_state['current_frame'] = frame_data["preview"]
if self.out_queue.full():
await self.out_queue.get() # Make space if full to avoid indefinite block
await self.out_queue.put(frame_data["api"])
await asyncio.sleep(1.0 / VIDEO_FPS_LIMIT)
except Exception as e:
st.error(f"Camera streaming error: {e}")
self.running = False
finally:
if cap:
await asyncio.to_thread(cap.release)
def _get_screen_frame(self): # Renamed for clarity
sct = mss.mss()
# Use the first monitor
monitor_number = 1
if len(sct.monitors) > 1: # sct.monitors[0] is all monitors, sct.monitors[1] is primary
monitor = sct.monitors[monitor_number]
else: # If only one monitor entry (all), just use it.
monitor = sct.monitors[0]
screenshot = sct.grab(monitor)
img = PIL.Image.frombytes("RGB", screenshot.size, screenshot.rgb)
preview_img = img.copy()
preview_img.thumbnail(VIDEO_PREVIEW_RESIZE)
api_img = img.copy()
api_img.thumbnail(VIDEO_API_RESIZE)
image_io = io.BytesIO()
api_img.save(image_io, format="jpeg")
image_io.seek(0)
image_bytes = image_io.read()
return {
"preview": preview_img,
"api": {
"mime_type": "image/jpeg",
"data": base64.b64encode(image_bytes).decode()
}
}
async def get_frames_from_screen(self): # Renamed for clarity
try:
while self.running:
frame_data = await asyncio.to_thread(self._get_screen_frame)
if frame_data is None:
await asyncio.sleep(0.01)
continue
st.session_state['current_frame'] = frame_data["preview"]
if self.out_queue.full():
await self.out_queue.get()
await self.out_queue.put(frame_data["api"])
await asyncio.sleep(1.0 / VIDEO_FPS_LIMIT)
except Exception as e:
st.error(f"Screen capture error: {e}")
self.running = False
async def send_realtime_media(self): # Renamed
try:
while self.running:
if not self.session:
await asyncio.sleep(0.1) # Wait for session to be established
continue
try:
msg = await asyncio.wait_for(self.out_queue.get(), timeout=0.5) # Timeout to prevent blocking indefinitely
if self.session and self.running: # Re-check session and running status
await self.session.send(input=msg) # No end_of_turn for continuous media
self.out_queue.task_done()
except asyncio.TimeoutError:
continue # No new media to send
except Exception as e:
if self.running: # Only log if we are supposed to be running
print(f"Error in send_realtime_media: {e}") # Log to console
# Consider if this error should stop the loop or be reported to UI
await asyncio.sleep(0.1) # Prevent tight loop on error
except asyncio.CancelledError:
print("send_realtime_media task cancelled.")
finally:
print("send_realtime_media task finished.")
async def listen_for_audio(self): # Renamed
self.audio_stream = None
try:
mic_info = await asyncio.to_thread(pya.get_default_input_device_info)
self.audio_stream = await asyncio.to_thread(
pya.open,
format=FORMAT,
channels=CHANNELS,
rate=SEND_SAMPLE_RATE,
input=True,
input_device_index=mic_info["index"],
frames_per_buffer=CHUNK_SIZE,
)
print("Microphone stream opened.")
while self.running:
try:
# exception_on_overflow=False helps avoid crashes on buffer overflows
data = await asyncio.to_thread(self.audio_stream.read, CHUNK_SIZE, exception_on_overflow=False)
if self.out_queue.full():
await self.out_queue.get() # Make space
await self.out_queue.put({"data": data, "mime_type": "audio/pcm"})
except IOError as e: # PyAudio specific IO errors
if e.errno == pyaudio.paInputOverflowed:
print("PyAudio Input overflowed. Skipping frame.") # Or log to a file/UI
else:
print(f"PyAudio read error: {e}")
self.running = False # Potentially stop on other IOErrors
break
except Exception as e:
print(f"Error in listen_for_audio: {e}")
await asyncio.sleep(0.01) # Prevent tight loop on error
except Exception as e:
st.error(f"Failed to open microphone: {e}") # This error needs to reach Streamlit UI
self.running = False
finally:
if self.audio_stream:
await asyncio.to_thread(self.audio_stream.stop_stream)
await asyncio.to_thread(self.audio_stream.close)
print("Microphone stream closed.")
async def receive_gemini_responses(self): # Renamed
try:
while self.running:
if not self.session:
await asyncio.sleep(0.1) # Wait for session
continue
try:
# Blocking receive, but should yield if self.running becomes false or session closes
turn = self.session.receive()
async for response in turn:
if not self.running: break # Exit if stop signal received during iteration
if data := response.data: # Audio data
if not self.audio_in_queue.full():
self.audio_in_queue.put_nowait(data)
else:
print("Playback audio queue full, discarding data.")
if text := response.text: # Text part of the response
# Queue this for the main thread to update Streamlit
st.session_state['chat_messages'].append({"role": "assistant", "content": text})
# Consider st.experimental_rerun() if immediate update is critical and safe
# For now, rely on Streamlit's natural refresh from chat_input or other interactions
# Handle turn completion logic if needed (e.g., clear audio queue for interruptions)
# For simplicity, current model might not need complex interruption handling here.
# If interruptions are implemented (e.g., user speaks while AI is speaking),
# you might want to clear self.audio_in_queue here.
except types.generation_types.StopCandidateException:
print("Gemini indicated end of response (StopCandidateException).") # Normal
except Exception as e:
if self.running:
print(f"Error receiving from Gemini: {e}")
await asyncio.sleep(0.1) # Prevent tight loop on error
except asyncio.CancelledError:
print("receive_gemini_responses task cancelled.")
finally:
print("receive_gemini_responses task finished.")
async def play_audio_responses(self): # Renamed
playback_stream = None
try:
playback_stream = await asyncio.to_thread(
pya.open,
format=FORMAT, # Assuming Gemini audio matches this, or adjust
channels=CHANNELS,
rate=RECEIVE_SAMPLE_RATE,
output=True,
)
print("Audio playback stream opened.")
while self.running:
try:
bytestream = await asyncio.wait_for(self.audio_in_queue.get(), timeout=0.5)
await asyncio.to_thread(playback_stream.write, bytestream)
self.audio_in_queue.task_done()
except asyncio.TimeoutError:
continue # No audio to play
except Exception as e:
print(f"Error playing audio: {e}")
await asyncio.sleep(0.01) # Prevent tight loop
except Exception as e:
st.error(f"Failed to open audio playback: {e}")
self.running = False
finally:
if playback_stream:
await asyncio.to_thread(playback_stream.stop_stream)
await asyncio.to_thread(playback_stream.close)
print("Audio playback stream closed.")
def stop_loop(self): # Renamed
print("Stop signal received for AudioLoop.")
self.running = False
# Queues can be an issue for graceful shutdown if tasks are blocked on put/get
# Put sentinel values or use timeouts in queue operations
if self.out_queue: # For send_realtime_media
self.out_queue.put_nowait(None) # Sentinel to unblock .get()
if self.audio_in_queue: # For play_audio_responses
self.audio_in_queue.put_nowait(None) # Sentinel
async def run(self):
st.session_state['run_loop'] = True # Indicate loop is running
self.running = True
print("AudioLoop starting...")
try:
# `client.aio.live.connect` is an async context manager
async with client.aio.live.connect(model=MODEL, config=CONFIG) as session:
self.session = session
print("Gemini session established.")
# Send the system prompt first.
try:
print("Sending system prompt to Gemini...")
# end_of_turn=False means this text is part of the initial context for the first actual user interaction.
await self.session.send(input=MEDICAL_ASSISTANT_SYSTEM_PROMPT, end_of_turn=False)
print("System prompt sent.")
except Exception as e:
st.error(f"Failed to send system prompt to Gemini: {str(e)}")
traceback.print_exception(e)
self.running = False # Stop if system prompt fails critical setup
return # Exit run method
# Initialize queues within the async context if they depend on loop specifics
self.audio_in_queue = asyncio.Queue(maxsize=AUDIO_QUEUE_MAXSIZE)
self.out_queue = asyncio.Queue(maxsize=10) # For outgoing media to Gemini API
async with asyncio.TaskGroup() as tg:
# Start all background tasks
print("Starting child tasks...")
tg.create_task(self.send_realtime_media(), name="send_realtime_media")
tg.create_task(self.listen_for_audio(), name="listen_for_audio")
if self.video_mode == "camera":
tg.create_task(self.get_frames_from_camera(), name="get_frames_from_camera")
elif self.video_mode == "screen":
tg.create_task(self.get_frames_from_screen(), name="get_frames_from_screen")
# If mode is "none", no video task is started.
tg.create_task(self.receive_gemini_responses(), name="receive_gemini_responses")
tg.create_task(self.play_audio_responses(), name="play_audio_responses")
print("All child tasks created.")
# TaskGroup will wait for all tasks to complete here.
# If self.running is set to False, tasks should ideally notice and exit.
print("TaskGroup finished.")
except asyncio.CancelledError:
print("AudioLoop.run() was cancelled.") # Usually from TaskGroup cancellation
except ExceptionGroup as eg: # From TaskGroup if child tasks fail
st.error(f"Error in async tasks: {eg.exceptions[0]}") # Show first error in UI
print(f"ExceptionGroup caught in AudioLoop.run(): {eg}")
for i, exc in enumerate(eg.exceptions):
print(f" Exception {i+1}/{len(eg.exceptions)} in TaskGroup: {type(exc).__name__}: {exc}")
traceback.print_exception(type(exc), exc, exc.__traceback__)
except Exception as e:
st.error(f"Critical error in session: {str(e)}")
print(f"Exception caught in AudioLoop.run(): {type(e).__name__}: {e}")
traceback.print_exception(e)
finally:
print("AudioLoop.run() finishing, cleaning up...")
self.running = False # Ensure all loops stop
st.session_state['run_loop'] = False # Signal that the loop has stopped
# `self.session` will be closed automatically by the `async with` block for `client.aio.live.connect`
self.session = None
# Other stream closures are handled in their respective task's finally blocks
print("AudioLoop finished.")
def main():
st.title("Gemini Live Medical Assistant")
with st.sidebar:
st.subheader("Settings")
video_mode_options = ["camera", "screen", "none"]
# Ensure default video mode is in options, find its index
default_video_index = video_mode_options.index(DEFAULT_MODE) if DEFAULT_MODE in video_mode_options else 0
video_mode = st.selectbox("Video Source", video_mode_options, index=default_video_index)
if not st.session_state.get('run_loop', False): # If loop is not running
if st.button("Start Session", key="start_session_button"):
st.session_state.chat_messages = [{ # Clear chat and add system message
"role": "system",
"content": (
"Medical Assistant activated. The AI has been instructed on its role to visually assist you. "
"Please remember, this AI cannot provide medical diagnoses or replace consultation with a healthcare professional."
)
}]
st.session_state.current_frame = None # Clear previous frame
audio_loop = AudioLoop(video_mode=video_mode)
st.session_state.audio_loop = audio_loop
# Run the asyncio event loop in a new thread
# daemon=True allows Streamlit to exit even if this thread is stuck (though it shouldn't be)
threading.Thread(target=lambda: asyncio.run(audio_loop.run()), daemon=True).start()
st.success("Session started. Initializing assistant...")
st.rerun() # Rerun to update button state and messages
else: # If loop is running
if st.button("Stop Session", key="stop_session_button"):
if st.session_state.audio_loop:
st.session_state.audio_loop.stop_loop() # Signal async tasks to stop
# Wait a moment for tasks to attempt cleanup (optional, can be tricky)
# time.sleep(1)
st.session_state.audio_loop = None
st.warning("Session stopping...")
st.rerun() # Rerun to update UI
# Main content area
col1, col2 = st.columns([2, 3]) # Adjust column ratio as needed
with col1:
st.subheader("Video Feed")
if st.session_state.get('run_loop', False) and st.session_state.get('current_frame') is not None:
st.image(st.session_state['current_frame'], caption="Live Feed" if video_mode != "none" else "Video Disabled", use_column_width=True)
elif video_mode != "none":
st.info("Video feed will appear here when the session starts.")
else:
st.info("Video input is disabled.")
with col2:
st.subheader("Chat with Medical Assistant")
chat_container = st.container() # For scrolling chat
with chat_container:
for msg in st.session_state.chat_messages:
with st.chat_message(msg["role"]):
st.write(msg["content"])
prompt = st.chat_input("Ask about what you're showing...", key="chat_input_box", disabled=not st.session_state.get('run_loop', False))
if prompt:
st.session_state.chat_messages.append({"role": "user", "content": prompt})
st.rerun() # Show user message immediately
if st.session_state.audio_loop:
# The text needs to be sent from within the asyncio loop or by scheduling it.
# A simple way is to call a method on audio_loop that uses asyncio.create_task or similar.
# For direct call from thread to asyncio loop, ensure it's thread-safe.
# A better way is to put the text into a queue that send_text_to_gemini reads from,
# or use asyncio.run_coroutine_threadsafe if the loop is known.
# Current send_text_to_gemini is an async method.
# We need to run it in the event loop of the audio_loop's thread.
loop = asyncio.get_event_loop_policy().get_event_loop() # Get current thread's loop (might not be the one)
if st.session_state.audio_loop.session: # Ensure session exists
# This is a simplified approach; proper thread-safe coroutine scheduling is more robust.
# Consider using asyncio.run_coroutine_threadsafe if audio_loop.run() exposes its loop.
asyncio.run(st.session_state.audio_loop.send_text_to_gemini(prompt))
else:
st.error("Session not fully active to send message.")
else:
st.error("Session is not active. Please start a session.")
# Rerun after processing to show potential AI response (if text part comes quickly)
# st.rerun() # This might be too frequent, rely on receive_gemini_responses to update chat
if __name__ == "__main__":
# Global PyAudio termination hook (optional, for very clean shutdowns)
# def cleanup_pyaudio():
# print("Terminating PyAudio globally.")
# pya.terminate()
# import atexit
# atexit.register(cleanup_pyaudio)
main() |