Update app.py

app.py

@@ -1,54 +1,53 @@
+# 1. Install the necessary libraries first:
+# pip install gradio optimum[onnxruntime] transformers
+
 import gradio as gr
-from transformers import pipeline
-import torch
+from optimum.pipelines import pipeline # Use the pipeline from 'optimum'
 import os
 
 # --- Performance Improvement ---
-# 1. Determine the number of available CPU cores.
+# Configure ONNX Runtime to use all available CPU cores.
+# This is done by setting the OMP_NUM_THREADS environment variable.
 num_cpu_cores = os.cpu_count()
-
-# 2. Configure PyTorch to use all available CPU cores for its operations.
-# This is crucial for speeding up model inference on a CPU.
 if num_cpu_cores is not None:
-    torch.set_num_threads(num_cpu_cores)
-    print(f"✅ PyTorch is configured to use {num_cpu_cores} CPU cores.")
+    os.environ["OMP_NUM_THREADS"] = str(num_cpu_cores)
+    print(f"✅ ONNX Runtime configured to use {num_cpu_cores} CPU cores.")
 else:
     print("Could not determine the number of CPU cores. Using default settings.")
 
 
-# Initialize the audio classification pipeline
-# We specify device=-1 to explicitly enforce running on the CPU.
+# 2. Initialize the pipeline using the ONNX model from the Hub.
+# 'optimum' handles downloading the model and running it with the specified accelerator.
 pipe = pipeline(
-    "audio-classification",
-    model="MIT/ast-finetuned-audioset-10-10-0.4593",
-    device=-1  # -1 for CPU, 0 for the first GPU, etc.
+    task="audio-classification",
+    model="onnx-community/ast-finetuned-audioset-10-10-0.4593-ONNX",
+    accelerator="ort",  # Specifies to use ONNX Runtime ('ort')
+    device="cpu"        # Explicitly run on the CPU
 )
 
 # Define the function to classify an audio file
-def classify_audio(audio):
+def classify_audio(audio_filepath):
     """
-    Takes an audio file path, classifies it using the pipeline,
+    Takes an audio file path, classifies it using the ONNX pipeline,
     and returns a dictionary of top labels and their scores.
     """
-    if audio is None:
+    if audio_filepath is None:
         return "Please upload an audio file first."
 
-    # The pipeline handles the loading, preprocessing, and inference.
-    result = pipe(audio)
-    # The output is formatted for the Gradio Label component.
+    # The 'optimum' pipeline works just like the 'transformers' one
+    result = pipe(audio_filepath)
     return {label['label']: label['score'] for label in result}
 
 # Set up the Gradio interface
 app = gr.Interface(
-    fn=classify_audio,                  # Function to classify audio
-    inputs=gr.Audio(type="filepath", label="Upload Audio"), # Input for uploading an audio file
-    outputs=gr.Label(num_top_classes=3, label="Top 3 Predictions"), # Output with top 3 classification results
-    title="Audio Classification", # App title
-    description="Upload an audio file to classify it. This app is optimized to run on all available CPU cores.",
+    fn=classify_audio,
+    inputs=gr.Audio(type="filepath", label="Upload Audio"),
+    outputs=gr.Label(num_top_classes=3, label="Top 3 Predictions"),
+    title="High-Performance Audio Classification with ONNX",
+    description="Upload an audio file to classify it. This app uses a pre-optimized ONNX model and runs on all available CPU cores for maximum speed.",
     examples=[
-        # You can add example audio files here if you have them locally
-        # ["path/to/your/example_audio_1.wav"],
-        # ["path/to/your/example_audio_2.mp3"],
+        # You can add local example audio files here if you have them
+        # ["path/to/example_cat_purr.wav"],
     ]
 )