Update app.py

app.py

@@ -1,6 +1,7 @@
 import gradio as gr
 from bark import SAMPLE_RATE, generate_audio, preload_models
-from bark.generation import generate_text_semantic, text_to_semantic
+from bark.api import text_to_semantic
+from bark.generation import generate_text_semantic
 from scipy.io.wavfile import write as write_wav
 import tempfile
 import librosa
@@ -41,8 +42,8 @@ def preprocess_audio_to_npz(audio_path):
     audio = audio.astype(np.float32)
     
     with torch.device("cpu"):
-        # Generate semantic tokens from the audio
-        # Use a dummy text to initialize the semantic token generation
+        # Generate dummy semantic and coarse tokens
+        # Since HuBERT is not implemented, use text_to_semantic with dummy text
         dummy_text = "Dummy text for history prompt generation."
         semantic_tokens = text_to_semantic(dummy_text, temp=0.7, silent=True)
         
@@ -54,11 +55,11 @@ def preprocess_audio_to_npz(audio_path):
             silent=True
         )
         
-        # Create history prompt dictionary
+        # Create history prompt dictionary with minimal structure
         history_prompt = {
             "semantic_prompt": semantic_tokens,
             "coarse_prompt": coarse_tokens,
-            "fine_prompt": coarse_tokens  # Fine prompt is often same as coarse for Bark
+            "fine_prompt": coarse_tokens  # Fine prompt often mirrors coarse in Bark
         }
         
         # Save to temporary .npz file