Create app.py

app.py

@@ -0,0 +1,107 @@
+import gradio as gr
+import numpy as np
+import librosa
+from PIL import Image
+import tempfile
+import os
+from app.preprocess import preprocess_audio
+from app.model import predict
+from collections import Counter, defaultdict
+
+
+# Process Image Input
+def process_image_input(img):
+    """Classify a spectrogram image directly using model.predict"""
+    label, confidence, probs = predict(img)
+    return label, round(confidence, 3), probs
+
+
+# Process Audio Input
+def process_audio_input(audio_file):
+
+    # Save uploaded audio temporarily
+    with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp:
+        tmp.write(audio_file)
+        tmp_path = tmp.name
+
+    # Preprocess → mel-spectrogram chunks (list of PIL images)
+    imgs = preprocess_audio(tmp_path)
+
+    os.remove(tmp_path)
+
+    # Predict on each chunk
+    all_preds = []
+    all_confs = []
+    all_probs = []
+
+    for img in imgs:
+        label, conf, probs = predict(img)
+        all_preds.append(label)
+        all_confs.append(conf)
+        all_probs.append(probs)
+
+    # Majority Vote
+    counter = Counter(all_preds)
+    max_count = max(counter.values())
+    candidates = [k for k, v in counter.items() if v == max_count]
+
+    if len(candidates) == 1:
+        final_label = candidates[0]
+    else:
+        conf_sums = defaultdict(float)
+        for i, label in enumerate(all_preds):
+            if label in candidates:
+                conf_sums[label] += all_confs[i]
+        final_label = max(conf_sums, key=conf_sums.get)
+
+    final_conf = float(np.mean([all_confs[i] for i, l in enumerate(all_preds) if l == final_label]))
+
+    return final_label, round(final_conf, 3), all_preds, [round(c, 3) for c in all_confs]
+
+
+#  MAIN GRADIO CLASSIFICATION PIPELINE (AUDIO OR IMAGE)
+def classify(audio, image):
+
+    # If image is provided → classify image
+    if image is not None:
+        label, conf, probs = process_image_input(image)
+        return {
+            "Final Label": label,
+            "Confidence": conf,
+            "Details": probs
+        }
+
+    # If audio is provided → preprocess audio → classify
+    if audio is not None:
+        label, conf, all_preds, all_confs = process_audio_input(audio)
+
+        return {
+            "Final Label": label,
+            "Confidence": conf,
+            "All Chunk Labels": all_preds,
+            "All Chunk Confidences": all_confs
+        }
+
+    # Nothing provided
+    return "Please upload an audio file OR a spectrogram image."
+
+
+# GRADIO UI
+interface = gr.Interface(
+    fn=classify,
+    inputs=[
+        gr.Audio(type="bytes", label="Upload Audio (WAV/MP3)"),
+        gr.Image(type="pil", label="Upload Spectrogram Image")
+    ],
+    outputs=gr.JSON(label="Prediction Results"),
+    title="General Audio Classifier (Audio + Spectrogram Support)",
+    description=(
+        "Upload a raw audio file OR a spectrogram image.\n"
+        "The app automatically detects the input type:\n"
+        "• If audio → the model preprocesses it into mel spectrogram chunks.\n"
+        "• If spectrogram → the model classifies it directly.\n"
+        "Built using CNN + Mel-Spectrogram + Gradio."
+    ),
+)
+
+interface.launch()