Update app.py

app.py

@@ -4,12 +4,12 @@ import requests
 import threading
 import torch
 import librosa
-import psutil
+#import psutil  # Not needed for concurrency gating, only for CPU usage checks
 
 from flask import Flask, request, jsonify
 from transformers import WhisperProcessor, WhisperForConditionalGeneration
 
-# GLOBAL concurrency lock or counter
+# GLOBAL concurrency counter & lock
 concurrent_requests = 0
 concurrent_requests_lock = threading.Lock()
 
@@ -29,24 +29,24 @@ WEBHOOK_URL = "https://hook.eu1.make.com/86zogci73u394k2uqpulp5yjjwgm8b9x"
 def transcribe_in_background(audio_url, file_id, company, user, file_name):
     global concurrent_requests
     try:
-        # 1) Download audio
+        # Download audio
         r = requests.get(audio_url)
         audio_path = "/tmp/temp_audio.wav"
         with open(audio_path, "wb") as f:
             f.write(r.content)
 
-        # 2) Load audio
+        # Load & 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)
 
-        # 3) Transcribe in chunks
+        # Split into 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)
@@ -57,7 +57,7 @@ def transcribe_in_background(audio_url, file_id, company, user, file_name):
             transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)[0]
             partial_text += transcription + "\n"
 
-        # 4) Post final transcription
+        # Post final transcription
         payload = {
             "Transcription": partial_text.strip(),
             "callDuration": call_duration,
@@ -69,7 +69,6 @@ def transcribe_in_background(audio_url, file_id, company, user, file_name):
         requests.post(WEBHOOK_URL, json=payload)
 
     except Exception as e:
-        # 5) Handle errors
         error_payload = {
             "error": str(e),
             "fileId": file_id,
@@ -80,47 +79,44 @@ def transcribe_in_background(audio_url, file_id, company, user, file_name):
         requests.post(WEBHOOK_URL, json=error_payload)
 
     finally:
-        # Always decrement concurrency, even on error
+        # Decrement concurrency counter
         with concurrent_requests_lock:
-            global concurrent_requests
             concurrent_requests -= 1
 
 @app.route("/transcribe", methods=["POST"])
 def transcribe_endpoint():
     global concurrent_requests
 
-    # 1) Check concurrency
+    # Concurrency check: only 1 job at a time
     with concurrent_requests_lock:
         if concurrent_requests >= 1:
-            # We only allow ONE job at a time
-            return jsonify({"error": "Server is busy with another transcription"}), 503
-        
-        # If it's free, claim the slot
+            # Return 429 if we already have a job in progress
+            return jsonify({"error": "Too many requests, server is already processing another job."}), 429
         concurrent_requests += 1
 
-    # 2) Parse request
+    # Parse JSON
     data = request.get_json()
     audio_url = data.get("audio_url")
     if not audio_url:
-        # Since we've already claimed concurrency=1, we should free it
+        # Free the concurrency slot since we're not using it
         with concurrent_requests_lock:
             concurrent_requests -= 1
         return jsonify({"error": "Missing 'audio_url' in request"}), 400
 
-    # 3) Read custom headers
+    # Read custom headers
     file_id = request.headers.get("fileId", "")
     company = request.headers.get("company", "")
     user = request.headers.get("user", "")
     file_name = request.headers.get("fileName", "")
 
-    # 4) Spawn a background thread
+    # Spawn a background thread
     thread = threading.Thread(
         target=transcribe_in_background,
         args=(audio_url, file_id, company, user, file_name)
     )
     thread.start()
 
-    # 5) Return an immediate response
+    # Return immediately
     return jsonify({
         "status": "Received. Transcription in progress.",
         "note": "Results will be sent via webhook once done."