Use lifespan to load the model just one
Browse files- whisper_fastapi_online_server.py +29 -21
whisper_fastapi_online_server.py
CHANGED
@@ -4,6 +4,7 @@ import asyncio
|
|
4 |
import numpy as np
|
5 |
import ffmpeg
|
6 |
from time import time
|
|
|
7 |
|
8 |
from fastapi import FastAPI, WebSocket, WebSocketDisconnect
|
9 |
from fastapi.responses import HTMLResponse
|
@@ -11,15 +12,8 @@ from fastapi.middleware.cors import CORSMiddleware
|
|
11 |
|
12 |
from src.whisper_streaming.whisper_online import backend_factory, online_factory, add_shared_args
|
13 |
|
14 |
-
app = FastAPI()
|
15 |
-
app.add_middleware(
|
16 |
-
CORSMiddleware,
|
17 |
-
allow_origins=["*"],
|
18 |
-
allow_credentials=True,
|
19 |
-
allow_methods=["*"],
|
20 |
-
allow_headers=["*"],
|
21 |
-
)
|
22 |
|
|
|
23 |
|
24 |
parser = argparse.ArgumentParser(description="Whisper FastAPI Online Server")
|
25 |
parser.add_argument(
|
@@ -49,28 +43,37 @@ parser.add_argument(
|
|
49 |
add_shared_args(parser)
|
50 |
args = parser.parse_args()
|
51 |
|
52 |
-
|
|
|
|
|
|
|
|
|
53 |
|
54 |
if args.diarization:
|
55 |
from src.diarization.diarization_online import DiartDiarization
|
56 |
|
57 |
|
58 |
-
|
59 |
-
with open("src/web/live_transcription.html", "r", encoding="utf-8") as f:
|
60 |
-
html = f.read()
|
61 |
|
|
|
|
|
|
|
|
|
|
|
62 |
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
|
|
|
|
|
|
|
|
|
67 |
|
68 |
-
SAMPLE_RATE = 16000
|
69 |
-
CHANNELS = 1
|
70 |
-
SAMPLES_PER_SEC = SAMPLE_RATE * int(args.min_chunk_size)
|
71 |
-
BYTES_PER_SAMPLE = 2 # s16le = 2 bytes per sample
|
72 |
-
BYTES_PER_SEC = SAMPLES_PER_SEC * BYTES_PER_SAMPLE
|
73 |
|
|
|
|
|
|
|
74 |
|
75 |
async def start_ffmpeg_decoder():
|
76 |
"""
|
@@ -91,6 +94,11 @@ async def start_ffmpeg_decoder():
|
|
91 |
return process
|
92 |
|
93 |
|
|
|
|
|
|
|
|
|
|
|
94 |
|
95 |
@app.websocket("/asr")
|
96 |
async def websocket_endpoint(websocket: WebSocket):
|
|
|
4 |
import numpy as np
|
5 |
import ffmpeg
|
6 |
from time import time
|
7 |
+
from contextlib import asynccontextmanager
|
8 |
|
9 |
from fastapi import FastAPI, WebSocket, WebSocketDisconnect
|
10 |
from fastapi.responses import HTMLResponse
|
|
|
12 |
|
13 |
from src.whisper_streaming.whisper_online import backend_factory, online_factory, add_shared_args
|
14 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
15 |
|
16 |
+
##### LOAD ARGS #####
|
17 |
|
18 |
parser = argparse.ArgumentParser(description="Whisper FastAPI Online Server")
|
19 |
parser.add_argument(
|
|
|
43 |
add_shared_args(parser)
|
44 |
args = parser.parse_args()
|
45 |
|
46 |
+
SAMPLE_RATE = 16000
|
47 |
+
CHANNELS = 1
|
48 |
+
SAMPLES_PER_SEC = SAMPLE_RATE * int(args.min_chunk_size)
|
49 |
+
BYTES_PER_SAMPLE = 2 # s16le = 2 bytes per sample
|
50 |
+
BYTES_PER_SEC = SAMPLES_PER_SEC * BYTES_PER_SAMPLE
|
51 |
|
52 |
if args.diarization:
|
53 |
from src.diarization.diarization_online import DiartDiarization
|
54 |
|
55 |
|
56 |
+
##### LOAD APP #####
|
|
|
|
|
57 |
|
58 |
+
@asynccontextmanager
|
59 |
+
async def lifespan(app: FastAPI):
|
60 |
+
global asr, tokenizer
|
61 |
+
asr, tokenizer = backend_factory(args)
|
62 |
+
yield
|
63 |
|
64 |
+
app = FastAPI(lifespan=lifespan)
|
65 |
+
app.add_middleware(
|
66 |
+
CORSMiddleware,
|
67 |
+
allow_origins=["*"],
|
68 |
+
allow_credentials=True,
|
69 |
+
allow_methods=["*"],
|
70 |
+
allow_headers=["*"],
|
71 |
+
)
|
72 |
|
|
|
|
|
|
|
|
|
|
|
73 |
|
74 |
+
# Load demo HTML for the root endpoint
|
75 |
+
with open("src/web/live_transcription.html", "r", encoding="utf-8") as f:
|
76 |
+
html = f.read()
|
77 |
|
78 |
async def start_ffmpeg_decoder():
|
79 |
"""
|
|
|
94 |
return process
|
95 |
|
96 |
|
97 |
+
##### ENDPOINTS #####
|
98 |
+
|
99 |
+
@app.get("/")
|
100 |
+
async def get():
|
101 |
+
return HTMLResponse(html)
|
102 |
|
103 |
@app.websocket("/asr")
|
104 |
async def websocket_endpoint(websocket: WebSocket):
|