Update app.py

app.py

@@ -3,6 +3,7 @@ import gradio as gr
 import requests
 import uuid
 import os
+import torch
 import soundfile as sf
 from f5_tts.api import F5TTS
 
@@ -16,10 +17,10 @@ whisper_model = whisper.load_model("base")
 
 print("Loading F5-TTS Engine...")
 f5tts = F5TTS()
-print("System Operational!")
+print("Voice Simulation Matrix Active!")
 
 # ==========================================
-# 2. Dynamic Text Response Generation
+# 2. Pure Dynamic Text Response Generation
 # ==========================================
 def get_ai_answer(question):
     response = requests.post(
@@ -35,61 +36,76 @@ def get_ai_answer(question):
         },
         timeout=10
     )
-    # Extracts text purely from the live API response array
     data = response.json()
-    return data["choices"][0]["message"]["content"].strip()
+    return data["choices"]["message"]["content"].strip()
 
 # ==========================================
-# 3. Dynamic Voice Synthesis Matrix
+# 3. Resilient Voice Synthesis Engine
 # ==========================================
 def clone_and_speak(text, dynamic_ref_audio):
     try:
         if not dynamic_ref_audio or not os.path.exists(dynamic_ref_audio):
+            print("❌ Reference enrollment track missing.")
             return None
 
         # Automatically extract text from reference clip
-        print(f"🔊 Transcribing profile: {dynamic_ref_audio}")
+        print(f"🔊 Transcribing profile properties: {dynamic_ref_audio}")
         ref_transcription = whisper_model.transcribe(dynamic_ref_audio)
         extracted_ref_text = ref_transcription["text"].strip()
         
         if not extracted_ref_text:
             extracted_ref_text = "Hello"
 
-        # Unique filename generation
+        # Unique filename generation to bypass browser cache freezes
         unique_filename = f"voice_clone_{uuid.uuid4().hex[:8]}.wav"
         
-        # Execute voice clone generation
-        outputs = f5tts.infer(
-            ref_file=dynamic_ref_audio,
-            ref_text=extracted_ref_text,
-            gen_text=text,
-            nfe_step=32,
-            speed=1.0
-        )
-        
-        # Universal unpacking handler for different package versions
-        if isinstance(outputs, tuple):
-            if len(outputs) == 3:
-                wav_data, sr, _ = outputs
+        # EXPLICIT BRIDGE: Try both common signatures of the F5TTS inference layer
+        print("🔊 Processing voice clone generation...")
+        try:
+            # Format A: Directly write to file using file_wave parameter
+            f5tts.infer(
+                ref_file=dynamic_ref_audio,
+                ref_text=extracted_ref_text,
+                gen_text=text,
+                file_wave=unique_filename,
+                nfe_step=32,
+                speed=1.0
+            )
+        except TypeError:
+            # Format B: Object returns raw arrays instead of writing files directly
+            outputs = f5tts.infer(
+                ref_file=dynamic_ref_audio,
+                ref_text=extracted_ref_text,
+                gen_text=text,
+                nfe_step=32,
+                speed=1.0
+            )
+            # Safe tuple array parsing extraction boundary
+            if isinstance(outputs, tuple):
+                if len(outputs) == 3:
+                    wav_data, sr, _ = outputs
+                else:
+                    wav_data, sr = outputs
             else:
-                wav_data, sr = outputs
-        else:
-            wav_data, sr = outputs, 24000
+                wav_data, sr = outputs, 24000
 
-        # Cast torch tensor to NumPy structure if present
-        if hasattr(wav_data, "cpu"):
-            wav_data = wav_data.cpu().numpy()
+            # Convert underlying torch tensor blocks to NumPy structure if present
+            if hasattr(wav_data, "cpu"):
+                wav_data = wav_data.cpu().numpy()
 
-        # Write data to physical disk space for browser playback mapping
-        sf.write(unique_filename, wav_data, sr, format='WAV', subtype='PCM_16')
-        
+            # Save arrays onto your server disk workspace
+            sf.write(unique_filename, wav_data, sr, format='WAV', subtype='PCM_16')
+
+        # Double check data properties on disk space
         if os.path.exists(unique_filename) and os.path.getsize(unique_filename) > 0:
-            print(f"🎉 Generated track successfully: {unique_filename}")
+            print(f"🎉 Generated track successfully saved to: {unique_filename}")
             return unique_filename
+        
+        print("❌ Generated audio file was empty or missing from disk.")
         return None
             
     except Exception as e:
-        print(f"❌ Core Exception: {e}")
+        print(f"❌ Core TTS System Exception triggered: {e}")
         return None
 
 # ==========================================
@@ -107,14 +123,14 @@ def chatbot_engine(text_input, audio_input, target_voice_profile):
     if target_voice_profile is None:
         return None
 
-    # Processes text and directly converts it into output audio path
+    # Pipeline generation flow execution
     ai_text_response = get_ai_answer(question)
     generated_audio_track = clone_and_speak(ai_text_response, target_voice_profile)
     
     return generated_audio_track
 
 # ==========================================
-# 5. Gradio Audio-Only UI Canvas
+# 5. Gradio Audio-Only UI Canvas (Single Page)
 # ==========================================
 with gr.Blocks(title="Dynamic Voice Simulation Studio") as interface:
     gr.Markdown("# 🎙️ Single-Page Voice Cloning Studio")