Update app.py

app.py

@@ -6,6 +6,7 @@ import numpy as np
 import re
 from scipy import signal
 import warnings
+import os
 warnings.filterwarnings('ignore')
 
 print("🚀 Starting Enhanced Hindi Speech Sentiment Analysis App...")
@@ -34,17 +35,24 @@ except Exception as e:
 # Load IndicWhisper for Hindi ASR (Best for Indian languages)
 print("🎤 Loading IndicWhisper Hindi ASR model...")
 try:
+    from transformers import AutoProcessor, AutoModelForSpeechSeq2Seq
+    
+    asr_processor = AutoProcessor.from_pretrained("vasista22/whisper-hindi-medium")
+    asr_model = AutoModelForSpeechSeq2Seq.from_pretrained("vasista22/whisper-hindi-medium")
+    
+    # Create pipeline with the loaded model
     asr_pipeline = pipeline(
         "automatic-speech-recognition",
-        model="vasista22/whisper-hindi-medium",  # IndicWhisper variant
+        model=asr_model,
+        tokenizer=asr_processor.tokenizer,
+        feature_extractor=asr_processor.feature_extractor,
         device="cpu",
         chunk_length_s=30
     )
     print("✅ IndicWhisper Hindi ASR model loaded successfully")
 except Exception as e:
-    print(f"⚠️ Error loading IndicWhisper, trying fallback: {e}")
+    print(f"❌ Error loading IndicWhisper, trying fallback: {e}")
     try:
-        # Fallback to OpenAI Whisper with Hindi optimization
         asr_pipeline = pipeline(
             "automatic-speech-recognition",
             model="openai/whisper-small",
@@ -195,15 +203,6 @@ def validate_hindi_text(text):
     
     return True, "Valid Hindi/Hinglish", hindi_ratio
 
-def transliterate_to_hindi(text):
-    """
-    If text is in Roman script, attempt to keep Hindi words
-    This is a placeholder - in production, use proper transliteration library
-    """
-    # For now, just return original text
-    # In production, use: indic-transliteration or aksharamukha library
-    return text
-
 # ============================================
 # 5. ENHANCED SENTIMENT ANALYSIS
 # ============================================
@@ -375,18 +374,11 @@ def predict(audio_filepath):
         # ============================================
         # STEP 2: Speech-to-Text (ASR)
         # ============================================
-        print("🔄 Transcribing audio with IndicWhisper...")
+        print("🔄 Transcribing audio with Whisper...")
         try:
-            # Save preprocessed audio temporarily
-            import tempfile
-            with tempfile.NamedTemporaryFile(suffix='.wav', delete=False) as temp_audio:
-                import soundfile as sf
-                sf.write(temp_audio.name, audio_processed, sr)
-                temp_audio_path = temp_audio.name
-            
             # Transcribe with Hindi language setting
             result = asr_pipeline(
-                temp_audio_path,
+                audio_filepath,
                 generate_kwargs={
                     "language": "hindi",
                     "task": "transcribe"
@@ -396,10 +388,6 @@ def predict(audio_filepath):
             transcription = result["text"].strip()
             print(f"📝 Raw transcription: '{transcription}'")
             
-            # Clean up temp file
-            import os
-            os.unlink(temp_audio_path)
-            
         except Exception as asr_error:
             print(f"❌ ASR Error: {asr_error}")
             return {
@@ -491,7 +479,7 @@ def predict(audio_filepath):
 # ============================================
 
 demo = gr.Interface(
-    fn=predict,  # Removed async - not needed for this implementation
+    fn=predict,
     inputs=gr.Audio(
         type="filepath",
         label="🎤 Record or Upload Hindi Audio",
@@ -506,7 +494,7 @@ demo = gr.Interface(
     ## 🇮🇳 Professional-grade Hindi/Hinglish Speech Emotion Analysis
     
     ### ✨ Advanced Features:
-    - **🎙️ IndicWhisper ASR** - Best-in-class Hindi transcription
+    - **🎙️ IndicWhisper ASR** - Specialized Hindi transcription model
     - **🧠 XLM-RoBERTa** - Multilingual sentiment analysis
     - **🎵 Prosodic Analysis** - Voice tone, pitch, energy detection
     - **🔄 Mixed Emotion Detection** - Handles complex feelings
@@ -554,9 +542,5 @@ demo = gr.Interface(
 
 if __name__ == "__main__":
     print("🌐 Starting server...")
-    demo.launch(
-        server_name="0.0.0.0",
-        server_port=7860,
-        show_error=True
-    )
+    demo.launch()
     print("🎉 Enhanced Hindi Sentiment Analysis App is ready!")