app.py

import os
import json
import requests
import threading
import torch
import librosa

from flask import Flask, request, jsonify
from transformers import WhisperProcessor, WhisperForConditionalGeneration

# GLOBAL concurrency counter & lock
concurrent_requests = 0
concurrent_requests_lock = threading.Lock()

app = Flask(__name__)

model_id = "ivrit-ai/whisper-large-v3-turbo"
processor = WhisperProcessor.from_pretrained(model_id)
model = WhisperForConditionalGeneration.from_pretrained(model_id)

device = "cuda" if torch.cuda.is_available() else "cpu"
model.to(device)

forced_decoder_ids = processor.get_decoder_prompt_ids(language="he", task="transcribe")

WEBHOOK_URL = "https://hook.eu1.make.com/86zogci73u394k2uqpulp5yjjwgm8b9x"


def transcribe_in_background(audio_url, file_id, company, user, file_name):
    global concurrent_requests
    try:
        # Download audio
        r = requests.get(audio_url)
        audio_path = "/tmp/temp_audio.wav"
        with open(audio_path, "wb") as f:
            f.write(r.content)

        # Load audio & limit to 1 hour
        waveform, sr = librosa.load(audio_path, sr=16000)
        max_sec = 3600
        waveform = waveform[: sr * max_sec]

        call_duration = int(len(waveform) / sr)

        # Transcribe in 25-second chunks
        chunk_sec = 25
        chunk_size = sr * chunk_sec
        chunks = [waveform[i : i + chunk_size] for i in range(0, len(waveform), chunk_size)]

        partial_text = ""
        for chunk in chunks:
            inputs = processor(chunk, sampling_rate=sr, return_tensors="pt", padding=True)
            input_features = inputs.input_features.to(device)

            with torch.no_grad():
                predicted_ids = model.generate(
                    input_features,
                    forced_decoder_ids=forced_decoder_ids
                )
            transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)[0]
            partial_text += transcription + "\n"

        # Send result to webhook
        payload = {
            "Transcription": partial_text.strip(),
            "callDuration": call_duration,
            "fileId": file_id,
            "company": company,
            "user": user,
            "fileName": file_name
        }
        requests.post(WEBHOOK_URL, json=payload)

    except Exception as e:
        error_payload = {
            "error": str(e),
            "fileId": file_id,
            "company": company,
            "user": user,
            "fileName": file_name
        }
        requests.post(WEBHOOK_URL, json=error_payload)

    finally:
        # Decrement concurrency count
        with concurrent_requests_lock:
            concurrent_requests -= 1


@app.route("/transcribe", methods=["POST"])
def transcribe_endpoint():
    global concurrent_requests

    # We only allow ONE job at a time:
    with concurrent_requests_lock:
        if concurrent_requests >= 1:
            # Return a 200 (OK) and a JSON message 
            return jsonify({
                "message": "Server is already processing another job, please try again later."
            }), 200
        
        # If it's free, occupy the slot
        concurrent_requests += 1

    data = request.get_json()
    audio_url = data.get("audio_url")
    if not audio_url:
        # If missing the audio_url, free the slot we claimed
        with concurrent_requests_lock:
            concurrent_requests -= 1
        return jsonify({"error": "Missing 'audio_url' in request"}), 400

    # Read headers
    file_id = request.headers.get("fileId", "")
    company = request.headers.get("company", "")
    user = request.headers.get("user", "")
    file_name = request.headers.get("fileName", "")

    # Spawn a background thread
    thread = threading.Thread(
        target=transcribe_in_background,
        args=(audio_url, file_id, company, user, file_name)
    )
    thread.start()

    return jsonify({
        "status": "Received. Transcription in progress.",
        "note": "Results will be sent via webhook once done."
    }), 202


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