Upload 2 files
Browse files- app.py +123 -0
- requirements.txt +3 -0
app.py
ADDED
@@ -0,0 +1,123 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from flask import Flask, request, jsonify, Response # Add Response here
|
2 |
+
from faster_whisper import WhisperModel
|
3 |
+
import torch
|
4 |
+
import io
|
5 |
+
import time
|
6 |
+
import datetime
|
7 |
+
from threading import Semaphore
|
8 |
+
import os
|
9 |
+
from werkzeug.utils import secure_filename
|
10 |
+
import tempfile
|
11 |
+
|
12 |
+
app = Flask(__name__)
|
13 |
+
|
14 |
+
# Configuration
|
15 |
+
MAX_CONCURRENT_REQUESTS = 2 # Adjust based on your server capacity
|
16 |
+
MAX_AUDIO_DURATION = 60 * 30 # 30 minutes maximum audio duration (adjust as needed)
|
17 |
+
TEMPORARY_FOLDER = tempfile.gettempdir()
|
18 |
+
ALLOWED_EXTENSIONS = {'mp3', 'wav', 'ogg', 'm4a', 'flac'}
|
19 |
+
|
20 |
+
# Device check for faster-whisper
|
21 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
22 |
+
compute_type = "float16" if device == "cuda" else "int8"
|
23 |
+
print(f"Using device: {device} with compute_type: {compute_type}")
|
24 |
+
|
25 |
+
# Faster Whisper setup with optimized parameters for long audio
|
26 |
+
beamsize = 2 # Slightly larger beam size can help with long-form accuracy
|
27 |
+
wmodel = WhisperModel(
|
28 |
+
"guillaumekln/faster-whisper-small",
|
29 |
+
device=device,
|
30 |
+
compute_type=compute_type,
|
31 |
+
download_root="./model_cache" # Cache model to avoid re-downloading
|
32 |
+
)
|
33 |
+
|
34 |
+
# Concurrency control
|
35 |
+
request_semaphore = Semaphore(MAX_CONCURRENT_REQUESTS)
|
36 |
+
active_requests = 0
|
37 |
+
|
38 |
+
def allowed_file(filename):
|
39 |
+
return '.' in filename and \
|
40 |
+
filename.rsplit('.', 1)[1].lower() in ALLOWED_EXTENSIONS
|
41 |
+
|
42 |
+
def cleanup_temp_files(file_path):
|
43 |
+
"""Ensure temporary files are deleted after processing"""
|
44 |
+
try:
|
45 |
+
if os.path.exists(file_path):
|
46 |
+
os.remove(file_path)
|
47 |
+
except Exception as e:
|
48 |
+
print(f"Error cleaning up temp file {file_path}: {str(e)}")
|
49 |
+
|
50 |
+
@app.route("/health", methods=["GET"])
|
51 |
+
def health_check():
|
52 |
+
"""Endpoint to check if API is running"""
|
53 |
+
return jsonify({
|
54 |
+
'status': 'API is running',
|
55 |
+
'timestamp': datetime.datetime.now().isoformat(),
|
56 |
+
'device': device,
|
57 |
+
'compute_type': compute_type,
|
58 |
+
'active_requests': active_requests,
|
59 |
+
'max_duration_supported': MAX_AUDIO_DURATION
|
60 |
+
})
|
61 |
+
|
62 |
+
@app.route("/status/busy", methods=["GET"])
|
63 |
+
def server_busy():
|
64 |
+
"""Endpoint to check if server is busy"""
|
65 |
+
is_busy = active_requests >= MAX_CONCURRENT_REQUESTS
|
66 |
+
return jsonify({
|
67 |
+
'is_busy': is_busy,
|
68 |
+
'active_requests': active_requests,
|
69 |
+
'max_capacity': MAX_CONCURRENT_REQUESTS
|
70 |
+
})
|
71 |
+
|
72 |
+
@app.route("/whisper_transcribe", methods=["POST"])
|
73 |
+
def whisper_transcribe():
|
74 |
+
global active_requests
|
75 |
+
|
76 |
+
if not request_semaphore.acquire(blocking=False):
|
77 |
+
return jsonify({'error': 'Server busy'}), 503
|
78 |
+
|
79 |
+
active_requests += 1
|
80 |
+
start_time = time.time()
|
81 |
+
temp_file_path = None
|
82 |
+
|
83 |
+
try:
|
84 |
+
if 'audio' not in request.files:
|
85 |
+
return jsonify({'error': 'No file provided'}), 400
|
86 |
+
|
87 |
+
audio_file = request.files['audio']
|
88 |
+
if not (audio_file and allowed_file(audio_file.filename)):
|
89 |
+
return jsonify({'error': 'Invalid file format'}), 400
|
90 |
+
|
91 |
+
temp_file_path = os.path.join(TEMPORARY_FOLDER, secure_filename(audio_file.filename))
|
92 |
+
audio_file.save(temp_file_path)
|
93 |
+
|
94 |
+
segments, _ = wmodel.transcribe(
|
95 |
+
temp_file_path,
|
96 |
+
beam_size=beamsize,
|
97 |
+
vad_filter=True,
|
98 |
+
without_timestamps=True, # Ensure timestamps are not included
|
99 |
+
compression_ratio_threshold=2.4,
|
100 |
+
word_timestamps=False
|
101 |
+
)
|
102 |
+
|
103 |
+
full_text = " ".join(segment.text for segment in segments)
|
104 |
+
return jsonify({'transcription': full_text}), 200
|
105 |
+
|
106 |
+
|
107 |
+
|
108 |
+
except Exception as e:
|
109 |
+
return jsonify({'error': str(e)}), 500
|
110 |
+
|
111 |
+
finally:
|
112 |
+
if temp_file_path:
|
113 |
+
cleanup_temp_files(temp_file_path)
|
114 |
+
active_requests -= 1
|
115 |
+
request_semaphore.release()
|
116 |
+
print(f"Processed in {time.time()-start_time:.2f}s (Active: {active_requests})")
|
117 |
+
|
118 |
+
if __name__ == "__main__":
|
119 |
+
# Create temporary folder if it doesn't exist
|
120 |
+
if not os.path.exists(TEMPORARY_FOLDER):
|
121 |
+
os.makedirs(TEMPORARY_FOLDER)
|
122 |
+
|
123 |
+
app.run(host="0.0.0.0", port=7860, threaded=True)
|
requirements.txt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
flask
|
2 |
+
faster-whisper
|
3 |
+
torch
|