Update app.py

app.py

@@ -167,19 +167,34 @@ def tts_interface(text_input, speaker_audio):
         print("Sentiment analyzer not loaded.")
 
     try:
-        print("Attempting to generate audio...")
-        model.tts_to_file(
+        print("Attempting to generate audio using model.inference...")
+        # Extract speaker embedding
+        try:
+            (
+                gpt_cond_latent,
+                speaker_embedding,
+            ) = model.get_conditioning_latents(audio_path=speaker_audio, gpt_cond_len=30, gpt_cond_chunk_len=4, max_ref_length=60)
+            print("Speaker embedding extracted successfully.")
+        except Exception as e:
+            print("Speaker encoding error:", str(e))
+            return f"Error during speaker encoding: {e}"
+
+        # Perform inference
+        out = model.inference(
             text=text_input,
-            file_path=audio_output_path,
-            emotion=emotion,
-            speaker_wav=speaker_audio,
-            language=language
+            language=language,
+            gpt_cond_latent=gpt_cond_latent,
+            speaker_embedding=speaker_embedding,
+            emotion=emotion # You might need to adjust how emotion is used, Xtts might not directly take an 'emotion' parameter like this.
         )
+
+        # Save the generated audio
+        torchaudio.save(audio_output_path, torch.tensor(out["wav"]).unsqueeze(0).cpu(), 24000)
         print(f"Audio generated and saved to: {audio_output_path}")
         return audio_output_path
     except Exception as e:
-        print(f"Error during TTS: {e}")
-        return f"Error during TTS: {e}"
+        print(f"Error during TTS inference: {e}")
+        return f"Error during TTS inference: {e}"
 
 iface = gr.Interface(
     fn=tts_interface,