Update app.py

app.py

@@ -4,7 +4,6 @@ import requests
 import threading
 import torch
 import librosa
-#import psutil  # Not needed for concurrency gating, only for CPU usage checks
 
 from flask import Flask, request, jsonify
 from transformers import WhisperProcessor, WhisperForConditionalGeneration
@@ -26,6 +25,7 @@ forced_decoder_ids = processor.get_decoder_prompt_ids(language="he", task="trans
 
 WEBHOOK_URL = "https://hook.eu1.make.com/86zogci73u394k2uqpulp5yjjwgm8b9x"
 
+
 def transcribe_in_background(audio_url, file_id, company, user, file_name):
     global concurrent_requests
     try:
@@ -35,14 +35,14 @@ def transcribe_in_background(audio_url, file_id, company, user, file_name):
         with open(audio_path, "wb") as f:
             f.write(r.content)
 
-        # Load & limit to 1 hour
+        # 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)
 
-        # Split into 25-second chunks
+        # 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)]
@@ -53,11 +53,14 @@ def transcribe_in_background(audio_url, file_id, company, user, file_name):
             input_features = inputs.input_features.to(device)
 
             with torch.no_grad():
-                predicted_ids = model.generate(input_features, forced_decoder_ids=forced_decoder_ids)
+                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"
 
-        # Post final transcription
+        # Send result to webhook
         payload = {
             "Transcription": partial_text.strip(),
             "callDuration": call_duration,
@@ -79,31 +82,35 @@ def transcribe_in_background(audio_url, file_id, company, user, file_name):
         requests.post(WEBHOOK_URL, json=error_payload)
 
     finally:
-        # Decrement concurrency counter
+        # Decrement concurrency count
         with concurrent_requests_lock:
             concurrent_requests -= 1
 
+
 @app.route("/transcribe", methods=["POST"])
 def transcribe_endpoint():
     global concurrent_requests
 
-    # Concurrency check: only 1 job at a time
+    # We only allow ONE job at a time:
     with concurrent_requests_lock:
         if concurrent_requests >= 1:
-            # Return 429 if we already have a job in progress
-            return jsonify({"error": "Too many requests, server is already processing another job."}), 429
+            # 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
 
-    # Parse JSON
     data = request.get_json()
     audio_url = data.get("audio_url")
     if not audio_url:
-        # Free the concurrency slot since we're not using it
+        # 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 custom headers
+    # Read headers
     file_id = request.headers.get("fileId", "")
     company = request.headers.get("company", "")
     user = request.headers.get("user", "")
@@ -116,11 +123,11 @@ def transcribe_endpoint():
     )
     thread.start()
 
-    # Return immediately
     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)