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
 
@@ -45,10 +46,10 @@ def get_ai_answer(question):
     q_lower = question.lower()
     if "salesforce" in q_lower:
         return "Salesforce is a cloud-based customer relationship management platform that helps businesses manage sales, marketing, and customer service tracking in one place."
-    return f"I processed your query about '{question}'. The neural voice matrix is active and generating the voice output track."
+    return f"I processed your query about '{question}'. The voice cloning engine is computing the response track now."
 
 # ==========================================
-# 3. Dynamic Voice Synthesis (Fixed Disk Write)
+# 3. Dynamic Voice Synthesis (Adaptive Unpacking)
 # ==========================================
 def clone_and_speak(text, dynamic_ref_audio):
     try:
@@ -56,7 +57,7 @@ def clone_and_speak(text, dynamic_ref_audio):
             print("⚠️ Error: No reference enrollment voice track provided.")
             return None
 
-        # Automatically extract text from reference file
+        # Automatically extract text from the uploaded reference file
         print(f"🔊 Analyzing voice file characteristics: {dynamic_ref_audio}")
         ref_transcription = whisper_model.transcribe(dynamic_ref_audio)
         extracted_ref_text = ref_transcription["text"].strip()
@@ -65,11 +66,11 @@ def clone_and_speak(text, dynamic_ref_audio):
             extracted_ref_text = "Hello"
         print(f"🎯 Successfully extracted reference transcript: '{extracted_ref_text}'")
 
-        # Create unique name to bypass browser caching issues
+        # Create unique filename to clear browser playback cache issues
         unique_filename = f"voice_clone_{uuid.uuid4().hex[:8]}.wav"
         
-        # FIX: Extract the raw audio arrays returned by the F5TTS function
-        wav_data, sr, spect = f5tts.infer(
+        # Execute F5-TTS simulation matrix inference
+        outputs = f5tts.infer(
             ref_file=dynamic_ref_audio,
             ref_text=extracted_ref_text,
             gen_text=text,
@@ -77,7 +78,18 @@ def clone_and_speak(text, dynamic_ref_audio):
             speed=1.0
         )
         
-        # FIX: Explicitly write the generated array onto your server disk space
+        # FIX: Adaptively unpack output regardless of package version layout
+        if isinstance(outputs, tuple):
+            wav_data = outputs[0]
+            sr = outputs[1]
+        else:
+            wav_data, sr = outputs
+
+        # Convert torch tensor back to NumPy structure if needed
+        if hasattr(wav_data, "cpu"):
+            wav_data = wav_data.cpu().numpy()
+
+        # FIX: Save raw memory structures explicitly onto your host disk space
         sf.write(unique_filename, wav_data, sr, format='WAV', subtype='PCM_16')
         
         if os.path.exists(unique_filename) and os.path.getsize(unique_filename) > 0: