app.py

from flask import Flask, request, jsonify
from faster_whisper import WhisperModel
import torch
import io
import time
import datetime
from threading import Semaphore

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)

# Concurrency control
MAX_CONCURRENT_REQUESTS = 2  # Adjust based on your server capacity
request_semaphore = Semaphore(MAX_CONCURRENT_REQUESTS)
active_requests = 0

# Warm up the model (important for CUDA)
print("Warming up the model...")
try:
    dummy_audio = io.BytesIO(b'')  # Empty audio for warmup
    segments, info = wmodel.transcribe(dummy_audio, beam_size=beamsize)
    _ = [segment.text for segment in segments]  # Force execution
    print("Model warmup complete")
except Exception as e:
    print(f"Model warmup failed: {str(e)}")

@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,
        'active_requests': active_requests
    })

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

@app.route("/whisper_transcribe", methods=["POST"])
def whisper_transcribe():
    global active_requests
    
    # Check if server is at capacity
    if not request_semaphore.acquire(blocking=False):
        return jsonify({
            'status': 'Server busy',
            'message': f'Currently processing {active_requests} requests',
            'suggestion': 'Please try again shortly'
        }), 503
    
    active_requests += 1
    print(f"Starting transcription (Active requests: {active_requests})")
    
    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

        audio_bytes = audio_file.read()
        audio_file = io.BytesIO(audio_bytes)

        try:
            # Timeout handling (60 seconds max processing time)
            start_time = time.time()
            segments, info = wmodel.transcribe(audio_file, beam_size=beamsize)
            
            text = ''
            for segment in segments:
                if time.time() - start_time > 60:  # Timeout after 60 seconds
                    raise TimeoutError("Transcription took too long")
                text += segment.text
            
            processing_time = time.time() - start_time
            print(f"Transcription completed in {processing_time:.2f} seconds")
            
            return jsonify({
                'transcription': text,
                'processing_time': processing_time,
                'language': info.language,
                'language_probability': info.language_probability
            })
            
        except TimeoutError:
            print("Transcription timeout")
            return jsonify({'error': 'Transcription timeout'}), 504
        except Exception as e:
            print(f"Transcription error: {str(e)}")
            return jsonify({'error': 'Transcription failed'}), 500
            
    finally:
        active_requests -= 1
        request_semaphore.release()
        print(f"Request completed (Active requests: {active_requests})")

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