Update app.py

app.py

@@ -12,11 +12,11 @@ import spaces
 import logging
 from huggingface_hub import login
 import threading
+import time
 
 torch._dynamo.config.disable = True
 torch._dynamo.config.suppress_errors = True
 
-
 logging.basicConfig(level=logging.INFO)
 logger = logging.getLogger(__name__)
 
@@ -25,44 +25,68 @@ if hf_token:
     login(token=hf_token)
 
 
-tts_model = None
-speakers_dict = None
-model_initialized = False
-model_initialized_lock = threading.Lock()
-
-@spaces.GPU()
-def initialize_model():
-    """Initialize the TTS model and speakers - called once with GPU context"""
-    global tts_model, speakers_dict, model_initialized, model_initialized_lock
+class ModelSingleton:
+    _instance = None
+    _lock = threading.Lock()
     
-    # Always acquire lock first for async safety
-    with model_initialized_lock:
-        if not model_initialized:
+    def __new__(cls):
+        if cls._instance is None:
+            with cls._lock:
+                if cls._instance is None:
+                    cls._instance = super(ModelSingleton, cls).__new__(cls)
+                    cls._instance.initialized = False
+                    cls._instance.tts_model = None
+                    cls._instance.speakers_dict = None
+                    cls._instance.init_lock = threading.RLock()  
+        return cls._instance
+    
+    @spaces.GPU()
+    def initialize(self):
+        """Thread-safe initialization with singleton pattern"""
+        if self.initialized:
+            logger.info("Model already initialized, skipping...")
+            return self.tts_model, self.speakers_dict
+            
+        with self.init_lock:
+            # Double-check pattern
+            if self.initialized:
+                logger.info("Model already initialized (double-check), skipping...")
+                return self.tts_model, self.speakers_dict
+                
             logger.info("Initializing Bambara TTS model...")
+            start_time = time.time()
             
             try:
                 from maliba_ai.tts.inference import BambaraTTSInference
                 from maliba_ai.config.speakers import Adame, Moussa, Bourama, Modibo, Seydou
                 
-                # All initialization inside the lock
-                tts_model = BambaraTTSInference()
-                speakers_dict = {
-                    "Adame": Adame,
+                self.tts_model = BambaraTTSInference()
+                self.speakers_dict = {
+                    "Adama": Adame,
                     "Moussa": Moussa, 
                     "Bourama": Bourama,
                     "Modibo": Modibo,
                     "Seydou": Seydou
                 }
                 
-                # Set flag last, after everything is ready
-                model_initialized = True
-                logger.info("Model initialized successfully!")
+                self.initialized = True
+                elapsed = time.time() - start_time
+                logger.info(f"Model initialized successfully in {elapsed:.2f} seconds!")
                 
             except Exception as e:
                 logger.error(f"Failed to initialize model: {e}")
                 raise e
+        
+        return self.tts_model, self.speakers_dict
     
-    return tts_model, speakers_dict
+    def get_model(self):
+        """Get the model, initializing if needed"""
+        if not self.initialized:
+            return self.initialize()
+        return self.tts_model, self.speakers_dict
+
+# Global singleton instance
+model_singleton = ModelSingleton()
 
 def validate_inputs(text, temperature, top_k, top_p, max_tokens):
     if not text or not text.strip():
@@ -81,14 +105,12 @@ def validate_inputs(text, temperature, top_k, top_p, max_tokens):
 
 @spaces.GPU()
 def generate_speech(text, speaker_name, use_advanced, temperature, top_k, top_p, max_tokens):
-
-    
     if not text.strip():
         return None, "Please enter some Bambara text."
     
     try:
-
-        tts, speakers = initialize_model()
+        # Get model through singleton
+        tts, speakers = model_singleton.get_model()
         
         speaker = speakers[speaker_name]
         
@@ -121,6 +143,15 @@ def generate_speech(text, speaker_name, use_advanced, temperature, top_k, top_p,
         logger.error(f"Speech generation failed: {e}")
         return None, f"❌ Error: {str(e)}"
 
+# Preload model on startup (optional - comment out if you prefer lazy loading)
+def preload_model():
+    """Preload the model when the app starts"""
+    try:
+        logger.info("Preloading model...")
+        model_singleton.initialize()
+        logger.info("Model preloaded successfully!")
+    except Exception as e:
+        logger.error(f"Failed to preload model: {e}")
 
 SPEAKER_NAMES = ["Adame", "Moussa", "Bourama", "Modibo", "Seydou"]
 
@@ -129,12 +160,11 @@ examples = [
     ["Mali bɛna diya kɔsɛbɛ,  ka a da a kan baara bɛ ka kɛ.", "Moussa"],
     ["Ne bɛ se ka sɛbɛnni yɛlɛma ka kɛ kuma ye", "Bourama"],
     ["I ka kɛnɛ wa?", "Modibo"],
-    ["Lakɔli karamɔgɔw tun tɛ ka se ka sɛbɛnni kɛ ka ɲɛ walanba kan wa denmisɛnw tun tɛ ka se ka o sɛbɛnni ninnu ye,  kuma tɛ ka u kalan. Denmisɛnw kɛra kunfinw ye.", "Adame"],
+    ["Lakɔli karamɔgɔw tun tɛ ka se ka sɛbɛnni kɛ ka ɲɛ walanba kan wa denmisɛnw tun tɛ ka se ka o sɛbɛnni ninnu ye,  kuma tɛ ka u kalan. Denmisɛnw kɛra kunfinw ye.", "Adama"],
     ["sigikafɔ kɔnɔ jamanaw ni ɲɔgɔn cɛ, olu ye a haminankow ye, wa o ko ninnu ka kan ka kɛ sariya ani tilennenya kɔnɔ.", "Seydou"],
     ["Aw ni ce. Ne tɔgɔ ye Adama. Awɔ,  ne ye maliden de ye. Aw Sanbɛ Sanbɛ. San min tɛ ɲinan ye, an bɛɛ ka jɛ ka o seli ɲɔgɔn fɛ,  hɛɛrɛ  ni lafiya la. Ala ka Mali suma. Ala ka Mali yiriwa. Ala ka Mali taa ɲɛ. Ala ka an ka seliw caya. Ala ka yafa an bɛɛ ma.", "Moussa"],
     ["An dɔlakelen bɛ masike bilenman don ka tɔw gɛn.", "Bourama"],
     ["Aw ni ce. Seidu bɛ aw fo wa aw ka yafa a ma, ka da a kan tuma dɔw la kow ka can.", "Modibo"],
-    
 ]
 
 def build_interface():
@@ -149,11 +179,8 @@ def build_interface():
         Convert Bambara text to speech. This model is currently experimental.
         
         **Bambara** is spoken by millions of people in Mali and West Africa.
-        .
         """)
 
-
-        
         with gr.Row():
             with gr.Column(scale=2):
                 text_input = gr.Textbox(
@@ -237,18 +264,20 @@ def build_interface():
             gr.Markdown("**Click any example below:**")
             
             for i, (text, speaker) in enumerate(examples):
-                btn = gr.Button(f" {text[:30]}{'...' if len(text) > 30 else ''}", size="sm")
+                btn = gr.Button(f"{text[:30]}{'...' if len(text) > 30 else ''}", size="sm")
                 btn.click(
                     fn=lambda t=text, s=speaker: load_example(t, s),
                     outputs=[text_input, speaker_dropdown, use_advanced, temperature, top_k, top_p, max_tokens]
                 )
         
-        with gr.Accordion(" About", open=False):
+        with gr.Accordion("About", open=False):
             gr.Markdown("""
             **⚠️ This is an experimental Bambara TTS model.**
             - **Languages**: Bambara (bm)  
             - **Speakers**: 5 different voice options
             - **Sample Rate**: 16kHz
+            
+            **Status**: Model loads once and reuses for all requests
             """)
         
         def toggle_advanced(use_adv):
@@ -280,6 +309,8 @@ def main():
     """Main function to launch the Gradio interface"""
     logger.info("Starting Bambara TTS Gradio interface.")
     
+    preload_model()
+    
     interface = build_interface()
     interface.launch(
         server_name="0.0.0.0",