Update app.py

app.py

@@ -7,19 +7,21 @@ DATASET_CONFIG = "speech_clean"
 DATASET_SPLIT = "train"
 TEXT_COLUMN = "raw_text"
 
-# The CORRECT column names, taken from your data instance example
+# --- THE FINAL, CORRECT COLUMN NAMES ---
+# Based on the official dataset viewer on Hugging Face and the KeyError.
+# This list is now definitive.
 AUDIO_COLUMNS = [
-    "audio.headset_mic",
-    "audio.laryngophone",
-    "audio.soft_in_ear_mic",
-    "audio.rigid_in_ear_mic",
+    "audio.headset_microphone",
+    "audio.throat_microphone",
+    "audio.soft_in_ear_microphone",
+    "audio.rigid_in_ear_microphone",
     "audio.forehead_accelerometer",
     "audio.temple_vibration_pickup"
 ]
 
 # --- Load Dataset ---
 try:
-    # Load the dataset normally, without any 'cast' operation.
+    # Load the dataset normally.
     dataset = load_dataset(DATASET_NAME, DATASET_CONFIG, split=DATASET_SPLIT)
 except Exception as e:
     dataset = None
@@ -32,42 +34,40 @@ def get_audio_row(index: int):
     """
     row_index = int(index)
     sample = dataset[row_index]
-    
+
     sentence = sample[TEXT_COLUMN]
-    
-    # --- THE FIX IS HERE ---
-    # We now extract the raw audio (NumPy array) and sampling rate directly.
-    # We return a list of tuples: (sampling_rate, audio_array).
-    # This is the most robust way and avoids all URL/path errors.
+
+    # This will now work because we are using the correct column names.
+    # We extract the raw audio (NumPy array) and sampling rate directly.
     raw_audio_data = [
         (sample[col]['sampling_rate'], sample[col]['array']) for col in AUDIO_COLUMNS
     ]
-    # --------------------
-    
+
     return [sentence] + raw_audio_data
 
 # --- Build the Gradio Interface ---
 with gr.Blocks(css="footer {display: none !important}") as demo:
     gr.Markdown("# Vibravox Multi-Audio Viewer")
-    
+
     if dataset is None:
         gr.Markdown("## 💥 Application Error")
         gr.Markdown(f"Could not load or process the dataset. Error: `{app_error}`")
     else:
         gr.Markdown("Select a row to listen to all corresponding audio sensor recordings.")
-        
+
         slider = gr.Slider(minimum=0, maximum=len(dataset) - 1, step=1, value=0, label="Select Data Row")
-        
+
         sentence_output = gr.Textbox(label="Raw Text", interactive=False)
-        
+
+        # Labels now match the correct column names
         with gr.Row():
-            audio1 = gr.Audio(label="Headset Mic")
-            audio2 = gr.Audio(label="Laryngophone")
-            audio3 = gr.Audio(label="Soft In-Ear Mic")
+            audio1 = gr.Audio(label="Headset Microphone")
+            audio2 = gr.Audio(label="Laryngophone (Throat Mic)")
+            audio3 = gr.Audio(label="Soft In-Ear Microphone")
         with gr.Row():
-            audio4 = gr.Audio(label="Rigid In-Ear Mic")
+            audio4 = gr.Audio(label="Rigid In-Ear Microphone")
             audio5 = gr.Audio(label="Forehead Accelerometer")
-            audio6 = gr.Audio(label="Temple Pickup")
+            audio6 = gr.Audio(label="Temple Vibration Pickup")
 
         outputs = [sentence_output, audio1, audio2, audio3, audio4, audio5, audio6]