app.py

import streamlit as st
import torch
import torchaudio
import torch.nn as nn
import torch.nn.functional as F
import torchvision.models as models
import tempfile
import os
import uuid

if "uploaded_file" not in st.session_state:
    st.session_state.uploaded_file = None
if "uploader_key" not in st.session_state:
    st.session_state.uploader_key = str(uuid.uuid4())

def reset_state():
    st.session_state.uploaded_file = None
    st.session_state.uploader_key = str(uuid.uuid4())
    st.rerun()

st.markdown(
    """
    <style>
    body { background-color: #f5f5f5; }
    .main { background-color: white; padding: 2rem; border-radius: 10px; margin: 2rem auto; max-width: 800px; }
    .stButton>button { background-color: #4CAF50; color: white; border: none; padding: 0.5rem 1rem; border-radius: 5px; }
    </style>
    """,
    unsafe_allow_html=True,
)

def load_and_process_audio(file_path, target_length=3.0, sample_rate=16000):
    waveform, sr = torchaudio.load(file_path)
    if waveform.shape[0] > 1:
        waveform = torch.mean(waveform, dim=0, keepdim=True)
    if sr != sample_rate:
        resampler = torchaudio.transforms.Resample(orig_freq=sr, new_freq=sample_rate)
        waveform = resampler(waveform)
    target_samples = int(target_length * sample_rate)
    current_samples = waveform.shape[1]
    if current_samples > target_samples:
        start = (current_samples - target_samples) // 2
        waveform = waveform[:, start : start + target_samples]
    elif current_samples < target_samples:
        padding = target_samples - current_samples
        pad_left = padding // 2
        pad_right = padding - pad_left
        waveform = F.pad(waveform, (pad_left, pad_right))
    return waveform

def extract_melspectrogram(waveform, sample_rate=16000, n_mels=80, n_fft=1024, hop_length=512):
    mel_spec = torchaudio.transforms.MelSpectrogram(
        sample_rate=sample_rate, n_fft=n_fft, hop_length=hop_length, n_mels=n_mels
    )(waveform)
    mel_spec = torchaudio.transforms.AmplitudeToDB()(mel_spec)
    return mel_spec

class AudioDeepfakeMODEL(nn.Module):
    def __init__(self, num_classes=2):
        super(AudioDeepfakeMODEL, self).__init__()
        self.resnet = models.resnet18(pretrained=True)
        self.resnet.conv1 = nn.Conv2d(
            in_channels=1,
            out_channels=64,
            kernel_size=(7, 7),
            stride=(2, 2),
            padding=(3, 3),
            bias=False,
        )
        self.resnet.fc = nn.Linear(self.resnet.fc.in_features, num_classes)
        for param in self.resnet.parameters():
            param.requires_grad = False
        for param in self.resnet.fc.parameters():
            param.requires_grad = True

    def forward(self, x):
        return self.resnet(x)

def predict_audio_deepfake(wav_path, checkpoint_path, device):
    model = AudioDeepfakeMODEL()
    checkpoint = torch.load(checkpoint_path, map_location=device)
    model.load_state_dict(checkpoint["model_state_dict"])
    model.to(device)
    model.eval()
    waveform = load_and_process_audio(wav_path)
    mel_spec = extract_melspectrogram(waveform)
    mel_spec = mel_spec.unsqueeze(0).to(device)
    with torch.no_grad():
        outputs = model(mel_spec)
        _, predicted = torch.max(outputs, 1)
    return predicted.item()

def main():
    st.markdown('<div class="main">', unsafe_allow_html=True)
    st.title("Audio Deepfake Detector")
    st.write("Upload a **.wav** file to check if it's **Real** or **Fake**.")

    uploaded_file = st.file_uploader(
        "Choose a .wav file",
        type=["wav"],
        key=st.session_state.uploader_key
    )
    if uploaded_file is not None:
        st.session_state.uploaded_file = uploaded_file

    if st.session_state.uploaded_file is not None:
        with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp_file:
            tmp_file.write(st.session_state.uploaded_file.read())
            tmp_path = tmp_file.name

        st.audio(st.session_state.uploaded_file, format="audio/wav")

        device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
        checkpoint_path = "best_model.pth"
        try:
            result = predict_audio_deepfake(tmp_path, checkpoint_path, device)
            label = "Real" if result == 1 else "Fake"
            st.success(f"Prediction: **{label}**")
        except Exception as e:
            st.error(f"Error during prediction: {e}")
        finally:
            os.remove(tmp_path)

        if st.button("Reset"):
            reset_state()

    st.markdown("</div>", unsafe_allow_html=True)

if __name__ == "__main__":
    main()