from transformers import ASTFeatureExtractor, ASTForAudioClassification
import torch
import librosa
import numpy as np
import pandas as pd

device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

model = ASTForAudioClassification.from_pretrained("MIT/ast-finetuned-audioset-10-10-0.4593").to(device)
feature_extractor = ASTFeatureExtractor.from_pretrained("MIT/ast-finetuned-audioset-10-10-0.4593")

def detect_siren(audio_path):
    audio, sr = librosa.load(audio_path, sr=16000)
    if len(audio.shape) == 1:
        audio = np.expand_dims(audio, axis=0)
    inputs = feature_extractor(audio, sampling_rate=16000, return_tensors="pt")
    inputs = {k: v.to(device) for k, v in inputs.items()}
    with torch.no_grad():
        outputs = model(**inputs)
        logits = outputs.logits
        predicted_class_idx = logits.argmax(-1).item()
    return model.config.id2label[predicted_class_idx]

# Example: Analyze multiple files
audio_files = ["/kaggle/input/emergency-vehicle-siren-sounds/sounds/ambulance/sound_1.wav", "/kaggle/input/emergency-vehicle-siren-sounds/sounds/ambulance/sound_10.wav", "/kaggle/input/emergency-vehicle-siren-sounds/sounds/ambulance/sound_100.wav"]
results = []

for file in audio_files:
    label = detect_siren(file)
    results.append({"file": file, "label": label})

# Save results
df = pd.DataFrame(results)
df.to_csv("detection_results.csv", index=False)

# app.py
import streamlit as st

st.title("🚨 Emergency Vehicle Siren Detection")

df = pd.read_csv("detection_results.csv")

st.dataframe(df)

for index, row in df.iterrows():
    if "siren" in row['label'].lower():
        st.error(f"🚨 {row['file']} => Siren Detected!")
    else:
        st.success(f"✅ {row['file']} => No Siren.")