Update app.py

app.py

@@ -1,29 +1,26 @@
 import gradio as gr
-from datasets import load_dataset, Audio
+from datasets import load_dataset
 
 # --- Configuration ---
 DATASET_NAME = "Cnam-LMSSC/vibravox-test"
 DATASET_CONFIG = "speech_clean"
 DATASET_SPLIT = "train"
 TEXT_COLUMN = "raw_text"
+
+# The CORRECT column names, taken from your data instance example
 AUDIO_COLUMNS = [
-    "audio.headset_microphone",
-    "audio.throat_microphone",
-    "audio.soft_in_ear_microphone",
-    "audio.rigid_in_ear_microphone",
+    "audio.headset_mic",
+    "audio.laryngophone",
+    "audio.soft_in_ear_mic",
+    "audio.rigid_in_ear_mic",
     "audio.forehead_accelerometer",
     "audio.temple_vibration_pickup"
 ]
 
 # --- Load Dataset ---
 try:
+    # Load the dataset normally, without any 'cast' operation.
     dataset = load_dataset(DATASET_NAME, DATASET_CONFIG, split=DATASET_SPLIT)
-    
-    new_features = dataset.features.copy()
-    for col in AUDIO_COLUMNS:
-        new_features[col] = Audio(decode=False)
-    dataset = dataset.cast(new_features)
-
 except Exception as e:
     dataset = None
     app_error = e
@@ -31,27 +28,23 @@ except Exception as e:
 # --- App Logic ---
 def get_audio_row(index: int):
     """
-    Retrieves a row and returns the text and the full URLs to the audio files.
+    Retrieves a row and returns the text and the RAW audio data.
     """
     row_index = int(index)
     sample = dataset[row_index]
     
     sentence = sample[TEXT_COLUMN]
     
-    # This is the URL structure that *should* be correct.
-    base_url = f"https://huggingface.co/datasets/{DATASET_NAME}/resolve/main/{DATASET_CONFIG}"
+    # --- 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.
+    raw_audio_data = [
+        (sample[col]['sampling_rate'], sample[col]['array']) for col in AUDIO_COLUMNS
+    ]
+    # --------------------
     
-    # --- !! THIS IS THE DEBUGGING STEP !! ---
-    # We will print the first URL to the logs to see exactly what is being generated.
-    # This will help us find any hidden typos or path errors.
-    first_file_path = sample[AUDIO_COLUMNS[0]]['path']
-    first_full_url = f"{base_url}/{first_file_path}"
-    print(f"DEBUGGING URL: '{first_full_url}'")
-    # ----------------------------------------
-
-    audio_urls = [f"{base_url}/{sample[col]['path']}" for col in AUDIO_COLUMNS]
-    
-    return [sentence] + audio_urls
+    return [sentence] + raw_audio_data
 
 # --- Build the Gradio Interface ---
 with gr.Blocks(css="footer {display: none !important}") as demo:
@@ -61,15 +54,23 @@ with gr.Blocks(css="footer {display: none !important}") as demo:
         gr.Markdown("## 💥 Application Error")
         gr.Markdown(f"Could not load or process the dataset. Error: `{app_error}`")
     else:
-        # The UI part remains the same
         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)
+        
         with gr.Row():
-            audio1, audio2, audio3 = gr.Audio(label="Headset Microphone"), gr.Audio(label="Laryngophone"), gr.Audio(label="Soft In-Ear Microphone")
+            audio1 = gr.Audio(label="Headset Mic")
+            audio2 = gr.Audio(label="Laryngophone")
+            audio3 = gr.Audio(label="Soft In-Ear Mic")
         with gr.Row():
-            audio4, audio5, audio6 = gr.Audio(label="Rigid In-Ear Microphone"), gr.Audio(label="Forehead Accelerometer"), gr.Audio(label="Temple Vibration Pickup")
+            audio4 = gr.Audio(label="Rigid In-Ear Mic")
+            audio5 = gr.Audio(label="Forehead Accelerometer")
+            audio6 = gr.Audio(label="Temple Pickup")
+
         outputs = [sentence_output, audio1, audio2, audio3, audio4, audio5, audio6]
+
         demo.load(fn=get_audio_row, inputs=gr.State(0), outputs=outputs)
         slider.change(fn=get_audio_row, inputs=slider, outputs=outputs)