Fix performance issues: updated requirements with exact versions and added model caching

app.py

@@ -12,8 +12,9 @@ from typing import Dict, Any, Optional
 from dotenv import load_dotenv
 import numpy as np
 
-# Import audio processors (only the 3 ML models + external API)
+# Import audio processors (only essential ones for deployment)
 from audio_processors.external_api import ExternalAPIProcessor
+from audio_processors.whisper_digit_processor import WhisperDigitProcessor
 from audio_processors.ml_mfcc_processor import MLMFCCProcessor
 from audio_processors.ml_mel_cnn_processor import MLMelCNNProcessor
 from audio_processors.ml_raw_cnn_processor import MLRawCNNProcessor
@@ -44,8 +45,18 @@ def allowed_file(filename: str) -> bool:
     """Check if file extension is allowed."""
     return '.' in filename and filename.rsplit('.', 1)[1].lower() in ALLOWED_EXTENSIONS
 
+# Global processor cache for model persistence
+_processor_cache = {}
+
 def initialize_processors():
-    """Initialize audio processors optimized for HF Spaces deployment."""
+    """Initialize audio processors optimized for HF Spaces deployment with caching."""
+    global _processor_cache
+    
+    # Return cached processors if already initialized
+    if _processor_cache:
+        app.logger.info(f"Using cached processors: {len(_processor_cache)} available")
+        return _processor_cache
+    
     procs = {}
     
     # ML-trained processors (high priority - use best models only)
@@ -58,11 +69,13 @@ def initialize_processors():
     ml_working_count = 0
     for proc_key, proc_class, proc_name in ml_processors:
         try:
+            # Initialize once and cache
+            app.logger.info(f"Loading {proc_name}...")
             processor = proc_class()
             if processor.is_configured():
                 procs[proc_key] = processor
                 ml_working_count += 1
-                app.logger.info(f"[OK] {proc_name} loaded successfully")
+                app.logger.info(f"[OK] {proc_name} loaded successfully (cached)")
             else:
                 app.logger.warning(f"[WARN] {proc_name} not configured (model files missing)")
         except Exception as e:
@@ -73,20 +86,31 @@ def initialize_processors():
         external_processor = ExternalAPIProcessor()
         if external_processor.is_configured():
             procs['external_api'] = external_processor
-            app.logger.info("[OK] External API processor initialized")
+            app.logger.info("[OK] External API processor initialized (cached)")
         else:
             app.logger.warning("[WARN] External API not configured")
     except Exception as e:
         app.logger.error(f"[FAIL] Failed to initialize External API: {str(e)}")
     
-    # Removed whisper processors to reduce dependencies and build size
+    # Whisper digit processor as another fallback
+    try:
+        whisper_processor = WhisperDigitProcessor()
+        if whisper_processor.is_configured():
+            procs['whisper_digit'] = whisper_processor
+            app.logger.info("[OK] Whisper digit processor initialized (cached)")
+    except Exception as e:
+        app.logger.error(f"[FAIL] Failed to initialize Whisper: {str(e)}")
+    
+    # Cache the processors globally
+    _processor_cache = procs
     
     app.logger.info(f"Processor initialization complete:")
     app.logger.info(f"  ML Models loaded: {ml_working_count}/3")
-    app.logger.info(f"  Total processors: {len(procs)}")
+    app.logger.info(f"  Total processors cached: {len(procs)}")
     
     return procs
 
+# Initialize processors on startup (cached globally)
 processors = initialize_processors()
 
 @app.route('/')
@@ -95,7 +119,7 @@ def index():
     return jsonify({
         'message': 'Streaming Digit Classifier API',
         'status': 'running',
-        'version': '1.2.0',
+        'version': '1.0.0',
         'available_processors': list(processors.keys()),
         'documentation': 'Frontend at Vercel, Backend API at HF Spaces'
     })
@@ -173,7 +197,7 @@ def process_audio():
         result.update({
             'audio_duration': round(duration, 3),
             'file_size': len(audio_data),
-            'api_version': '1.2.0'
+            'api_version': '1.0.0'
         })
         
         app.logger.info(f"Processed audio with {method}: '{result['predicted_digit']}' in {result['inference_time']}s")
@@ -231,7 +255,7 @@ def process_audio_chunk():
             'segment_index': 0,
             'segment_size': len(standardized_audio),
             'is_streaming': True,
-            'api_version': '1.2.0'
+            'api_version': '1.0.0'
         })
         
         app.logger.info(f"Streaming prediction: '{result['predicted_digit']}' "
@@ -294,7 +318,7 @@ def health_check():
             'status': 'healthy',
             'timestamp': time.time(),
             'processors': processor_health,
-            'version': '1.2.0',
+            'version': '1.0.0',
             'deployment': 'huggingface-spaces'
         })
     

requirements_hf.txt

@@ -1,14 +1,14 @@
-# HF Spaces Requirements - Python 3.10 compatible versions (streamlined)
+# HF Spaces Requirements - Exact working versions from local system
 # Core Flask API
 Flask==2.3.3
 Flask-CORS==4.0.0
 requests==2.31.0
 python-dotenv==1.0.0
 
-# Audio Processing Core
-numpy==1.25.0
+# Audio Processing Core - Fixed versions (avoid yanked scipy 1.11.0)
+numpy==1.21.6
 librosa==0.10.1
-scipy==1.11.0
+scipy==1.9.3
 soundfile==0.12.1
 
 # Critical Audio Processing Libraries (Fix audio corruption)
@@ -17,12 +17,18 @@ ffmpeg-python==0.2.0
 audioread==3.0.0
 resampy==0.4.2
 
-# ML Models - PyTorch CPU 
-torch==2.0.1
-torchaudio==2.0.2
+# ML Models - PyTorch CPU with exact versions
+torch==1.12.1
+torchaudio==0.12.1
+torchvision==0.13.1
 
-# Essential ML utilities (no pandas/datasets needed for inference)
-scikit-learn==1.3.0
+# Essential ML utilities
+scikit-learn==1.1.3
+transformers==4.21.3
+collections-extended==2.0.2
+
+# VAD processing
+webrtcvad==2.0.10
 
 # Logging and utilities
 tqdm==4.65.0