Update app.py

app.py

@@ -1,12 +1,12 @@
 import os
 
-# Disable problematic optimizations for ZeroGPU compatibility
+
 os.environ["TORCHDYNAMO_DISABLE"] = "1"
 os.environ["TORCH_COMPILE_DISABLE"] = "1"
 os.environ["PYTORCH_DISABLE_CUDNN_BENCHMARK"] = "1"
 os.environ["TOKENIZERS_PARALLELISM"] = "false"
 
-# Check if we're in ZeroGPU or similar restricted environment
+
 def is_restricted_environment():
     return (
         os.getenv("ZERO_GPU") or 
@@ -15,7 +15,7 @@ def is_restricted_environment():
         "spaces" in str(os.getenv("HOSTNAME", "")).lower()
     )
 
-# Disable Unsloth optimizations in restricted environments
+
 if is_restricted_environment():
     os.environ["UNSLOTH_DISABLE"] = "1"
     os.environ["DISABLE_UNSLOTH"] = "1" 
@@ -44,7 +44,7 @@ hf_token = os.getenv("HF_TOKEN")
 if hf_token:
     login(token=hf_token)
 
-# Global variables for model caching
+
 _tts_model = None
 _speakers_dict = None
 _model_initialized = False
@@ -53,10 +53,7 @@ _initialization_in_progress = False
 def get_speakers_dict():
     """Get speakers dictionary using the correct SDK structure"""
     try:
-        # Import the Speakers class (not individual speakers)
         from maliba_ai.config.settings import Speakers
-        
-        # Access all 10 speakers through the Speakers class
         speakers_dict = {
             "Adama": Speakers.Adama,
             "Moussa": Speakers.Moussa,
@@ -190,11 +187,9 @@ def generate_speech(text, speaker_name, use_advanced, temperature, top_k, top_p,
         if waveform is None or waveform.size == 0:
             return None, "Failed to generate audio. Please try again."
         
-        # Convert to numpy if it's a tensor
         if isinstance(waveform, torch.Tensor):
             waveform = waveform.cpu().numpy()
         
-        # Ensure proper audio format (convert float32 to int16 range but keep as float for Gradio)
         if waveform.dtype == np.float32:
             # Normalize to [-1, 1] range if needed
             if np.max(np.abs(waveform)) > 1.0:
@@ -209,22 +204,20 @@ 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)}"
 
-# Use available speakers (detect what's actually available, prioritize Bourama)
+
 def get_speaker_names():
     speakers = get_speakers_dict()
     if speakers:
         speaker_list = list(speakers.keys())
-        # Reorder to match preferred order (Bourama first)
+
         preferred_order = ["Bourama", "Adama", "Moussa", "Modibo", "Seydou", 
                           "Amadou", "Bakary", "Ngolo", "Ibrahima", "Amara"]
         
-        # Sort available speakers according to preferred order
         ordered_speakers = []
         for speaker in preferred_order:
             if speaker in speaker_list:
                 ordered_speakers.append(speaker)
         
-        # Add any remaining speakers not in preferred list
         for speaker in speaker_list:
             if speaker not in ordered_speakers:
                 ordered_speakers.append(speaker)
@@ -232,51 +225,48 @@ def get_speaker_names():
         logger.info(f"Available speakers: {ordered_speakers}")
         return ordered_speakers
     else:
-        # Final fallback with Bourama first
+
         logger.warning("No speakers loaded, using fallback list")
         return ["Bourama", "Adama", "Moussa", "Modibo", "Seydou"]
 
 SPEAKER_NAMES = get_speaker_names()
 
-# Examples representing ALL 10 speakers - with fallbacks for missing speakers
+
 examples = [
-    ["Aw ni ce", "Adama"],  # Natural conversational greeting
-    ["Mali bɛna diya kɔsɛbɛ, ka a da a kan baara bɛ ka kɛ.", "Moussa"],  # Clear pronunciation for informative content
-    ["Ne bɛ se ka sɛbɛnni yɛlɛma ka kɛ kuma ye", "Bourama"],  # Most stable for educational content
-    ["I ka kɛnɛ wa?", "Modibo"],  # Expressive delivery for questions
-    ["Lakɔli karamɔgɔw tun tɛ ka se ka sɛbɛnni kɛ ka ɲɛ walanda kan wa denmisɛnw tun tɛ ka se ka o sɛbɛnni ninnu ye, kuma tɛ ka u kalan. Denmisɛnw kɛra kunfinw ye.", "Adama"],  # Natural conversational tone for longer explanation
-    ["sigikafɔ kɔnɔ jamanaw ni ɲɔgɔn cɛ, olu ye a haminankow ye, wa o ko ninnu ka kan ka kɛ sariya ani tilennenya kɔnɔ.", "Seydou"],  # Balanced characteristics for formal content
-    ["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"],  # Clear pronunciation for heartfelt long message
-    ["An dɔlakelen bɛ masike bilenman don ka tɔw gɛn.", "Bourama"],  # Most stable for complex statement
-    ["Aw ni ce. Seidu bɛ aw fo wa aw ka yafa a ma, ka da a kan tuma dɔw la kow ka can.", "Modibo"],  # Expressive delivery for personal greeting
-    ["To tɔ nantan ni lafiya, o ka fisa ni so fa dumuniba kɛlɛma ye.", "Amadou"],  # Warm and friendly voice for wisdom saying
-    ["Mali ye jamana ɲuman ye!", "Bakary"],  # Deep, authoritative tone for patriotic statement
-    ["An ka ɲɔgɔn dɛmɛ ka baara kɛ ɲɔgɔn fɛ", "Ngolo"],  # Youthful and energetic for collaboration
-    ["Hakili to yɔrɔ min na, sabali bɛ yen", "Ibrahima"],  # Calm and measured for philosophical thought
-    ["Dɔnko ɲuman ye, a bɛ dɔn mɔgɔ kɔnɔ", "Amara"],  # Melodic and smooth for poetic expression
+    ["Aw ni ce", "Adama"],  
+    ["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 walanda 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"],  
+    ["To tɔ nantan ni lafiya, o ka fisa ni so fa dumuniba kɛlɛma ye.", "Amadou"],  
+    ["Mali ye jamana ɲuman ye!", "Bakary"],  
+    ["An ka ɲɔgɔn dɛmɛ ka baara kɛ ɲɔgɔn fɛ", "Ngolo"],  
+    ["Hakili to yɔrɔ min na, sabali bɛ yen", "Ibrahima"],  
+    ["Dɔnko ɲuman ye, a bɛ dɔn mɔgɔ kɔnɔ", "Amara"],  
 ]
 
 def get_safe_examples():
     """Get examples with speaker fallbacks for missing speakers"""
     safe_examples = []
     
-    # Fallback mapping for missing speakers
     fallback_speakers = {
-        "Amadou": "Adama",    # Warm -> Natural conversational
-        "Bakary": "Modibo",   # Authoritative -> Expressive
-        "Ngolo": "Adama",     # Youthful -> Natural conversational
-        "Ibrahima": "Seydou", # Calm -> Balanced
-        "Amara": "Moussa"     # Melodic -> Clear pronunciation
+        "Amadou": "Adama",    
+        "Bakary": "Modibo",   
+        "Ngolo": "Adama",    
+        "Ibrahima": "Seydou", 
+        "Amara": "Moussa"     
     }
     
     for text, speaker in examples:
-        # Use original speaker if available, otherwise use fallback
         if speaker in SPEAKER_NAMES:
             safe_examples.append([text, speaker])
         elif speaker in fallback_speakers and fallback_speakers[speaker] in SPEAKER_NAMES:
             safe_examples.append([text, fallback_speakers[speaker]])
         else:
-            # Final fallback to first available speaker
             safe_examples.append([text, SPEAKER_NAMES[0]])
     
     return safe_examples