import gradio as gr import os, time, re, json, base64, asyncio, threading, uuid, io import numpy as np import soundfile as sf from pydub import AudioSegment from openai import OpenAI from websockets import connect from dotenv import load_dotenv # Load secrets load_dotenv() OPENAI_API_KEY = os.getenv("OPENAI_API_KEY") ASSISTANT_ID = os.getenv("ASSISTANT_ID") client = OpenAI(api_key=OPENAI_API_KEY) HEADERS = {"Authorization": f"Bearer {OPENAI_API_KEY}", "OpenAI-Beta": "realtime=v1"} WS_URI = "wss://api.openai.com/v1/realtime?intent=transcription" connections = {} class WebSocketClient: def __init__(self, uri, headers, client_id): self.uri = uri self.headers = headers self.client_id = client_id self.websocket = None self.queue = asyncio.Queue(maxsize=10) self.transcript = "" self.loop = asyncio.new_event_loop() async def connect(self): try: self.websocket = await connect(self.uri, additional_headers=self.headers) with open("openai_transcription_settings.json", "r") as f: await self.websocket.send(f.read()) await asyncio.gather(self.receive_messages(), self.send_audio_chunks()) except Exception as e: print(f"\U0001F534 WebSocket Connection Failed: {e}") def run(self): asyncio.set_event_loop(self.loop) self.loop.run_until_complete(self.connect()) def enqueue_audio_chunk(self, sr, arr): if not self.queue.full(): asyncio.run_coroutine_threadsafe(self.queue.put((sr, arr)), self.loop) async def send_audio_chunks(self): while True: sr, arr = await self.queue.get() if arr.ndim > 1: arr = arr.mean(axis=1) if np.max(np.abs(arr)) > 0: arr = arr / np.max(np.abs(arr)) int16 = (arr * 32767).astype(np.int16) buf = io.BytesIO() sf.write(buf, int16, sr, format='WAV', subtype='PCM_16') audio = AudioSegment.from_file(buf, format="wav").set_frame_rate(24000) out = io.BytesIO() audio.export(out, format="wav") out.seek(0) await self.websocket.send(json.dumps({ "type": "input_audio_buffer.append", "audio": base64.b64encode(out.read()).decode() })) async def receive_messages(self): async for msg in self.websocket: data = json.loads(msg) if data["type"] == "conversation.item.input_audio_transcription.delta": self.transcript += data["delta"] def create_ws(): cid = str(uuid.uuid4()) client = WebSocketClient(WS_URI, HEADERS, cid) threading.Thread(target=client.run, daemon=True).start() connections[cid] = client return cid def send_audio(chunk, cid): if not cid or cid not in connections: return "Connecting..." sr, arr = chunk connections[cid].enqueue_audio_chunk(sr, arr) return connections[cid].transcript.strip() def clear_transcript_only(cid): if cid in connections: connections[cid].transcript = "" return "" def clear_chat_only(): return [], None, None def handle_chat(user_input, history, thread_id, image_url): if not OPENAI_API_KEY or not ASSISTANT_ID: return "โŒ Missing secrets!", history, thread_id, image_url try: if thread_id is None: thread = client.beta.threads.create() thread_id = thread.id client.beta.threads.messages.create(thread_id=thread_id, role="user", content=user_input) run = client.beta.threads.runs.create(thread_id=thread_id, assistant_id=ASSISTANT_ID) while True: status = client.beta.threads.runs.retrieve(thread_id=thread_id, run_id=run.id) if status.status == "completed": break time.sleep(1) msgs = client.beta.threads.messages.list(thread_id=thread_id) for msg in reversed(msgs.data): if msg.role == "assistant": content = msg.content[0].text.value history.append((user_input, content)) match = re.search( r'https://raw\.githubusercontent\.com/AndrewLORTech/surgical-pathology-manual/main/[\w\-/]*\.png', content) if match: image_url = match.group(0) break return "", history, thread_id, image_url except Exception as e: return f"โŒ {e}", history, thread_id, image_url def feed_transcript(transcript, history, thread_id, image_url, cid): if not transcript.strip(): return gr.update(), history, thread_id, image_url if cid in connections: connections[cid].transcript = "" return handle_chat(transcript, history, thread_id, image_url) def update_image_display(image_url): if image_url and isinstance(image_url, str) and image_url.startswith("http"): return image_url return None with gr.Blocks(theme=gr.themes.Soft()) as app: gr.Markdown("# ๐Ÿ“„ Document AI Assistant") chat_state = gr.State([]) thread_state = gr.State() image_state = gr.State() client_id = gr.State() # Hidden toggles as backend state show_left = gr.Checkbox(value=True, visible=False) show_right = gr.Checkbox(value=True, visible=False) with gr.Row(): toggle_left_btn = gr.Button("๐Ÿ“„ Toggle Document Panel") toggle_right_btn = gr.Button("๐ŸŽ™๏ธ Toggle Voice Panel") with gr.Row(equal_height=True) as layout: left_col = gr.Column(visible=True, scale=1) with left_col: image_display = gr.Image(label="๐Ÿ–ผ๏ธ Document", type="filepath", show_download_button=False, height=600) center_col = gr.Column(scale=2) with center_col: chat = gr.Chatbot(label="๐Ÿ’ฌ Chat", height=600) with gr.Row(): user_prompt = gr.Textbox(placeholder="Ask your question...", show_label=False, scale=8) send_btn = gr.Button("Send", variant="primary", scale=2) with gr.Row(): clear_chat_btn = gr.Button("๐Ÿ—‘๏ธ Clear Chat") right_col = gr.Column(visible=True, scale=1) with right_col: gr.Markdown("### ๐ŸŽ™๏ธ Voice Input") voice_input = gr.Audio(label="Tap to Record", streaming=True, type="numpy", show_label=True) voice_transcript = gr.Textbox(label="Transcript", lines=2, interactive=False) with gr.Row(): voice_send_btn = gr.Button("๐ŸŸข Send Voice to Assistant") clear_transcript_btn = gr.Button("๐Ÿงน Clear Transcript") def update_layout(show_left_val, show_right_val): center_scale = 2 if not show_left_val and not show_right_val: center_scale = 4 elif not show_left_val or not show_right_val: center_scale = 3 return ( gr.update(visible=show_left_val), gr.update(scale=center_scale), gr.update(visible=show_right_val) ) def toggle_left_fn(current): return not current def toggle_right_fn(current): return not current toggle_left_btn.click(fn=toggle_left_fn, inputs=show_left, outputs=show_left) toggle_right_btn.click(fn=toggle_right_fn, inputs=show_right, outputs=show_right) show_left.change(fn=update_layout, inputs=[show_left, show_right], outputs=[left_col, center_col, right_col]) show_right.change(fn=update_layout, inputs=[show_left, show_right], outputs=[left_col, center_col, right_col]) send_btn.click(fn=handle_chat, inputs=[user_prompt, chat_state, thread_state, image_state], outputs=[user_prompt, chat, thread_state, image_state]) voice_input.stream(fn=send_audio, inputs=[voice_input, client_id], outputs=voice_transcript, stream_every=0.5) voice_send_btn.click(fn=feed_transcript, inputs=[voice_transcript, chat_state, thread_state, image_state, client_id], outputs=[user_prompt, chat, thread_state, image_state]) clear_transcript_btn.click(fn=clear_transcript_only, inputs=[client_id], outputs=voice_transcript) clear_chat_btn.click(fn=clear_chat_only, outputs=[chat, thread_state, image_state]) image_state.change(fn=update_image_display, inputs=image_state, outputs=image_display) app.load(fn=create_ws, outputs=[client_id]) app.launch()