Update app.py

app.py

@@ -2,15 +2,13 @@ import os
 import requests
 import json
 import re
-import tempfile
 import io
 import numpy as np
 from datetime import datetime, timedelta, timezone
 from bs4 import BeautifulSoup
-from flask import Flask, request, Response, stream_with_context, render_template_string, send_file
+from flask import Flask, request, Response, stream_with_context, render_template_string
 from supertonic import TTS
 from pydub import AudioSegment
-from scipy.io import wavfile
 
 app = Flask(__name__)
 
@@ -19,7 +17,6 @@ app = Flask(__name__)
 # ----------------------------------------------------
 print("Loading Supertonic TTS Model...")
 try:
-    # This runs on CPU by default via ONNX
     tts = TTS(auto_download=True)
     print("✅ TTS Model loaded successfully!")
 except Exception as e:
@@ -314,7 +311,7 @@ STRICT DIRECTIVES:
 
 
 # ----------------------------------------------------
-# 🎙️ TTS DIRECT API ENDPOINT (NEW)
+# 🎙️ TTS DIRECT API ENDPOINT (OPTIMIZED)
 # ----------------------------------------------------
 @app.route('/api/tts', methods=['POST'])
 def generate_tts():
@@ -329,7 +326,7 @@ def generate_tts():
     if not text.strip():
         return Response(json.dumps({"error": "Text is empty."}), status=400, mimetype='application/json')
         
-    # Text cleaning: \u0900-\u097F allows Hindi Devanagari text safely
+    # Text cleaning: allow Hindi Devanagari script safely
     clean_text = re.sub(r'[^\w\s.,!?\'\"\-\<\>\:\u0900-\u097F]', '', text, flags=re.UNICODE)
     
     if not clean_text.strip():
@@ -338,37 +335,38 @@ def generate_tts():
     try:
         lang_code = LANGUAGES.get(language_name, "en")
         
-        # 🚀 OPTIMIZATION: Voice Style Caching
+        # 🚀 Use cached voice style
         if voice not in VOICE_STYLES_CACHE:
             VOICE_STYLES_CACHE[voice] = tts.get_voice_style(voice_name=voice)
         style = VOICE_STYLES_CACHE[voice]
         
-        # Synthesize audio
+        # Synthesize audio (returns numpy array)
         wav, duration = tts.synthesize(clean_text, voice_style=style, lang=lang_code)
         
-        # 🚀 OPTIMIZATION: In-Memory Processing (No Disk I/O)
-        buffer = io.BytesIO()
-        sample_rate = 22050 
-        
+        # Convert to 16-bit PCM if needed
         if wav.dtype != np.int16:
             max_val = np.max(np.abs(wav))
             if max_val > 0:
                 wav_int16 = np.int16(wav / max_val * 32767)
             else:
                 wav_int16 = wav.astype(np.int16)
-            wavfile.write(buffer, sample_rate, wav_int16)
         else:
-            wavfile.write(buffer, sample_rate, wav)
-            
-        buffer.seek(0)
-        
-        # 🚀 OPTIMIZATION: WAV to MP3 Conversion
-        audio_segment = AudioSegment.from_wav(buffer)
+            wav_int16 = wav
+
+        sample_rate = 22050
+        # Create AudioSegment directly from raw bytes (mono, 16‑bit PCM)
+        audio_segment = AudioSegment(
+            wav_int16.tobytes(),
+            frame_rate=sample_rate,
+            sample_width=2,  # 2 bytes = 16 bit
+            channels=1
+        )
+
+        # Export to MP3 in memory
         mp3_buffer = io.BytesIO()
-        audio_segment.export(mp3_buffer, format="mp3", bitrate="128k", parameters=["-ar", "22050"])
+        audio_segment.export(mp3_buffer, format="mp3", bitrate="128k", parameters=["-ar", str(sample_rate)])
         mp3_buffer.seek(0)
-        
-        # Return the actual audio file
+
         return Response(mp3_buffer.read(), mimetype="audio/mpeg")
         
     except Exception as e:
@@ -376,5 +374,4 @@ def generate_tts():
 
 
 if __name__ == '__main__':
-    # Flask App Run
     app.run(host='0.0.0.0', port=7860)