Update app.py

app.py

@@ -33,25 +33,31 @@ _initialization_in_progress = False
 def get_speakers_dict():
     """Get speakers dictionary using the new SDK structure"""
     try:
-        # Use new structure from maliba_ai.config.settings (as shown in README)
+        # Try the new structure first - check what's actually available
         from maliba_ai.config.settings import Speakers
-        return {
-            "Adama": Speakers.Adama,
-            "Moussa": Speakers.Moussa, 
-            "Bourama": Speakers.Bourama,
-            "Modibo": Speakers.Modibo,
-            "Seydou": Speakers.Seydou,
-            "Amadou": Speakers.Amadou,
-            "Bakary": Speakers.Bakary,
-            "Ngolo": Speakers.Ngolo,
-            "Ibrahima": Speakers.Ibrahima,
-            "Amara": Speakers.Amara
-        }
+        
+        # Get all available speaker attributes dynamically
+        available_speakers = {}
+        # Updated speaker list with all 10 speakers in preferred order
+        speaker_names = ["Bourama", "Adama", "Moussa", "Modibo", "Seydou", 
+                        "Amadou", "Bakary", "Ngolo", "Ibrahima", "Amara"]
+        
+        for name in speaker_names:
+            if hasattr(Speakers, name):
+                available_speakers[name] = getattr(Speakers, name)
+                
+        if available_speakers:
+            logger.info(f"Loaded {len(available_speakers)} speakers from new structure: {list(available_speakers.keys())}")
+            return available_speakers
+        else:
+            raise AttributeError("No speakers found in new structure")
+            
     except Exception as e:
         logger.error(f"Failed to import from new settings structure: {e}")
         # Fallback to old structure if new one fails
         try:
             from maliba_ai.config.speakers import Adame, Moussa, Bourama, Modibo, Seydou
+            logger.info("Using fallback old speaker structure")
             return {
                 "Adama": Adame,
                 "Moussa": Moussa, 
@@ -167,6 +173,18 @@ 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:
+                waveform = waveform / np.max(np.abs(waveform))
+            # Keep as float32 but ensure proper range for Gradio
+            waveform = np.clip(waveform, -1.0, 1.0)
+        
         sample_rate = 16000
         return (sample_rate, waveform), f"✅ Audio generated successfully for speaker {speaker_name}"
         
@@ -174,32 +192,51 @@ 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 (prioritize new SDK structure with 10 speakers)
+# Use available speakers (detect what's actually available, prioritize Bourama)
 def get_speaker_names():
     speakers = get_speakers_dict()
-    if speakers and len(speakers) >= 10:
-        # New SDK with all 10 speakers
-        return ["Bourama", "Adama", "Moussa", "Modibo", "Seydou", "Amadou", "Bakary", "Ngolo", "Ibrahima", "Amara"]
-    elif speakers:
-        # Return whatever speakers are available
-        return list(speakers.keys())
+    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)
+                
+        logger.info(f"Available speakers: {ordered_speakers}")
+        return ordered_speakers
     else:
-        # Fallback to old 5 speakers
-        return ["Adama", "Moussa", "Bourama", "Modibo", "Seydou"]
+        # 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 with variety of lengths and speakers matched to content
+# Examples with variety of lengths and speakers matched to their characteristics
 examples = [
     ["Aw ni ce", "Adama"],  # Natural conversational greeting
-    ["Mali bɛna diya kɔsɛbɛ, ka a da a kan baara bɛ ka kɛ.", "Bakary" if "Bakary" in SPEAKER_NAMES else "Moussa"],  
-    ["Ne bɛ se ka sɛbɛnni yɛlɛma ka kɛ kuma ye", "Moussa"],  
-    ["I ka kɛnɛ wa?", "Ngolo" if "Ngolo" in SPEAKER_NAMES else "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.", "Bourama"],  
-    ["sigikafɔ kɔnɔ jamanaw ni ɲɔgɔn cɛ, olu ye a haminankow ye, wa o ko ninnu ka kan ka kɛ sariya ani tilennenya kɔnɔ.", "Ibrahima" if "Ibrahima" in SPEAKER_NAMES else "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.", "Amara" if "Amara" in SPEAKER_NAMES else "Moussa"],  
-    ["An dɔlakelen bɛ masike bilenman don ka tɔw gɛn.", "Modibo"],  
-    ["Aw ni ce. Seidu bɛ aw fo wa aw ka yafa a ma, ka da a kan tuma dɔw la kow ka can.", "Amadou" if "Amadou" in SPEAKER_NAMES else "Modibo"],
+    ["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
 ]
 
 def build_interface():
@@ -283,7 +320,8 @@ def build_interface():
         audio_output = gr.Audio(
             label="Generated Speech",
             type="numpy",
-            interactive=False
+            interactive=False,
+            format="wav"  # Specify WAV format to help with conversion
         )
             
         status_output = gr.Textbox(