Real-time meeting assistant

1
# pip install pyaudio websocket-client
2
import pyaudio
3
import websocket
4
import json
5
import threading
6
import time
7
from urllib.parse import urlencode
8
9
# ── Config ────────────────────────────────────────────────────
10
YOUR_API_KEY = "YOUR_API_KEY"
11
12
PROMPT = (
13
    "Summarize this speaker turn in one sentence, then list any "
14
    "action items mentioned.\n\nTranscript: {{turn}}"
15
)
16
17
LLM_GATEWAY_CONFIG = {
18
    "model": "claude-sonnet-4-5-20250929",
19
    "messages": [{"role": "user", "content": PROMPT}],
20
    "max_tokens": 500,
21
}
22
23
CONNECTION_PARAMS = {
24
    "sample_rate": 16000,
25
    "speech_model": "u3-rt-pro",
26
    "format_turns": True,
27
    "min_turn_silence": 560,   # Wait longer for natural meeting pauses
28
    "max_turn_silence": 2000,
29
    "llm_gateway": json.dumps(LLM_GATEWAY_CONFIG),
30
}
31
32
API_ENDPOINT = f"wss://streaming.assemblyai.com/v3/ws?{urlencode(CONNECTION_PARAMS)}"
33
34
# Audio settings
35
FRAMES_PER_BUFFER = 800
36
SAMPLE_RATE = 16000
37
stop_event = threading.Event()
38
39
def on_open(ws):
40
    print("Connected — speak into your microphone. Press Ctrl+C to stop.\n")
41
42
    def stream_audio():
43
        audio = pyaudio.PyAudio()
44
        stream = audio.open(
45
            input=True, frames_per_buffer=FRAMES_PER_BUFFER,
46
            channels=1, format=pyaudio.paInt16, rate=SAMPLE_RATE,
47
        )
48
        while not stop_event.is_set():
49
            try:
50
                data = stream.read(FRAMES_PER_BUFFER, exception_on_overflow=False)
51
                ws.send(data, websocket.ABNF.OPCODE_BINARY)
52
            except Exception:
53
                break
54
        stream.stop_stream()
55
        stream.close()
56
        audio.terminate()
57
58
    threading.Thread(target=stream_audio, daemon=True).start()
59
60
def on_message(ws, message):
61
    data = json.loads(message)
62
    msg_type = data.get("type")
63
64
    if msg_type == "Turn":
65
        transcript = data.get("transcript", "")
66
        if data.get("end_of_turn") and transcript:
67
            print(f"[Turn] {transcript}\n")
68
        elif transcript:
69
            print(f"\r  ... {transcript[-80:]}", end="", flush=True)
70
71
    elif msg_type == "LLMGatewayResponse":
72
        content = data.get("data", {}).get("choices", [{}])[0].get("message", {}).get("content", "")
73
        print(f"[Assistant] {content}\n")
74
75
    elif msg_type == "Termination":
76
        print(f"\nSession ended — {data.get('audio_duration_seconds', 0)}s of audio processed.")
77
78
def on_error(ws, error):
79
    print(f"Error: {error}")
80
    stop_event.set()
81
82
def on_close(ws, code, msg):
83
    stop_event.set()
84
85
ws_app = websocket.WebSocketApp(
86
    API_ENDPOINT,
87
    header={"Authorization": YOUR_API_KEY},
88
    on_open=on_open, on_message=on_message,
89
    on_error=on_error, on_close=on_close,
90
)
91
92
ws_thread = threading.Thread(target=ws_app.run_forever, daemon=True)
93
ws_thread.start()
94
95
try:
96
    while ws_thread.is_alive():
97
        time.sleep(0.1)
98
except KeyboardInterrupt:
99
    print("\nStopping...")
100
    stop_event.set()
101
    if ws_app.sock and ws_app.sock.connected:
102
        ws_app.send(json.dumps({"type": "Terminate"}))
103
        time.sleep(2)
104
    ws_app.close()