app.py

from flask import Flask, request, jsonify
from faster_whisper import WhisperModel
import torch
import io
import time
from threading import Lock
from queue import Queue
import datetime

app = Flask(__name__)

# Device check for faster-whisper
device = "cuda" if torch.cuda.is_available() else "cpu"
compute_type = "float16" if device == "cuda" else "int8"
print(f"Using device: {device} with compute_type: {compute_type}")

# Faster Whisper setup
beamsize = 2
wmodel = WhisperModel("guillaumekln/faster-whisper-small", device=device, compute_type=compute_type)

# Server status tracking
active_requests = 0
request_queue = Queue()
status_lock = Lock()
MAX_CONCURRENT_REQUESTS = 2 # Adjust based on your server capacity

@app.route("/health", methods=["GET"])
def health_check():
    """Endpoint to check if API is running"""
    return jsonify({
        'status': 'API is running',
        'timestamp': datetime.datetime.now().isoformat(),
        'device': device,
        'compute_type': compute_type
    })

@app.route("/status/busy", methods=["GET"])
def server_busy():
    """Endpoint to check if server is busy"""
    with status_lock:
        is_busy = active_requests >= MAX_CONCURRENT_REQUESTS
        return jsonify({
            'is_busy': is_busy,
            'active_requests': active_requests,
            'max_capacity': MAX_CONCURRENT_REQUESTS,
            'queue_size': request_queue.qsize()
        })

@app.route("/status/queue", methods=["GET"])
def queue_status():
    """Endpoint to get current queue size"""
    return jsonify({
        'queue_size': request_queue.qsize(),
        'active_requests': active_requests
    })

@app.route("/whisper_transcribe", methods=["POST"])
def whisper_transcribe():
    global active_requests
    
    # Check if server is at capacity
    with status_lock:
        if active_requests >= MAX_CONCURRENT_REQUESTS:
            request_queue.put(datetime.datetime.now())
            return jsonify({
                'status': 'Server busy',
                'message': f'Currently processing {active_requests} requests',
                'queue_position': request_queue.qsize()
            }), 503
        
        active_requests += 1
    
    try:
        if 'audio' not in request.files:
            return jsonify({'error': 'No file provided'}), 400

        audio_file = request.files['audio']
        allowed_extensions = {'mp3', 'wav', 'ogg', 'm4a'}
        if not (audio_file and audio_file.filename.lower().split('.')[-1] in allowed_extensions):
            return jsonify({'error': 'Invalid file format'}), 400

        print(f"Transcribing audio on {device} (Active requests: {active_requests})")
        audio_bytes = audio_file.read()
        audio_file = io.BytesIO(audio_bytes)

        try:
            segments, info = wmodel.transcribe(audio_file, beam_size=beamsize)
            text = ''
            starttime = time.time()
            for segment in segments:
                text += segment.text
            print(f"Time to transcribe: {time.time() - starttime} seconds")
            return jsonify({'transcription': text})
        except Exception as e:
            print(f"Transcription error: {str(e)}")
            return jsonify({'error': 'Transcription failed'}), 500
    finally:
        with status_lock:
            active_requests -= 1
            # Remove oldest queued request if any
            if not request_queue.empty():
                try:
                    request_queue.get_nowait()
                except:
                    pass

if __name__ == "__main__":
    app.run(host="0.0.0.0", debug=True, port=7860, threaded=True)