STT quickstart

1
import asyncio
2
import base64
3
import json
4
import os
5
from pathlib import Path
6
7
from websockets.asyncio.client import connect
8
9
URL = "wss://api.narilabs.com/v1/realtime?intent=transcription"
10
11
12
async def receive(socket, expected=None):
13
    event = json.loads(await socket.recv())
14
    if event["type"] == "error":
15
        raise RuntimeError(event["error"]["code"] + ": " + event["error"]["message"])
16
    if event["type"] == "input_audio_buffer.commit_empty":
17
        raise RuntimeError("No pending audio to transcribe")
18
    if expected and event["type"] != expected:
19
        raise RuntimeError(f"Expected {expected}, received {event['type']}")
20
    return event
21
22
23
async def main():
24
    pcm = Path("hello.pcm").read_bytes()
25
26
    async with connect(
27
        URL,
28
        additional_headers={"Authorization": f"Bearer {os.environ['NARI_API_KEY']}"},
29
    ) as socket:
30
        await receive(socket, "session.created")
31
        await socket.send(json.dumps({
32
            "type": "session.configure",
33
            "session": {"model": "qwen3-asr:free", "language": "en", "turn_detection": None},
34
        }))
35
        await receive(socket, "session.configured")
36
37
        async def send_audio():
38
            for offset in range(0, len(pcm), 3200):  # 100 ms
39
                chunk = pcm[offset:offset + 3200]
40
                await socket.send(json.dumps({
41
                    "type": "input_audio_buffer.append",
42
                    "audio": base64.b64encode(chunk).decode("ascii"),
43
                }))
44
                await asyncio.sleep(len(chunk) / 32000)
45
            await socket.send(json.dumps({"type": "input_audio_buffer.commit"}))
46
47
        async def read_results():
48
            while True:
49
                event = await receive(socket)
50
                if event["type"] == "transcript.partial":
51
                    print("Partial:", event["transcript"])
52
                elif event["type"] == "transcript.completed":
53
                    print("Final:", event["transcript"])
54
                    return
55
56
        await asyncio.gather(send_audio(), read_results())
57
58
59
asyncio.run(main())