app.py

import gradio as gr
import librosa
import numpy as np
import torch
from torch import Tensor
import torch.nn as nn
from model import Model

model_path = 'final_model.pth'
def load_data(path):
    X, fs = librosa.load(path)
    X_pad = pad(X,64600)
    x_inp = Tensor(X_pad).unsqueeze(0)
    return x_inp,fs

def pad(x, max_len=64600):
    x_len = x.shape[0]
    if x_len >= max_len:
        return x[:max_len]
    # need to pad
    num_repeats = int(max_len / x_len)+1
    padded_x = np.tile(x, (1, num_repeats))[:, :max_len][0]
    return padded_x	

device = 'cuda' if torch.cuda.is_available() else 'cpu'
model = Model(None, device)
nb_params = sum([param.view(-1).size()[0] for param in model.parameters()])
model =nn.DataParallel(model).to(device)

model.load_state_dict(torch.load(model_path, map_location=device))
print("Model loaded : {}".format(model_path))

model.eval()
prediction_dict = {0: 'Fake', 1: 'Real'}
def Detection(audio_1):
    
    x_inp,fs = load_data(audio_1)
    print(x_inp.shape)
    validity_probs = model(x_inp)
    validity_probs = torch.nn.functional.softmax(validity_probs, dim=1)
    
    emotion = torch.argmax(validity_probs).item()
    print(emotion)
    validity = prediction_dict[emotion]
    # validity as a dictionary of class probabilities
    # validity = {prediction_dict[i]: float(validity_probs[0][i]) for i in range(2)}

    return validity

audio_1 = gr.Audio(type="filepath", label="Audio 1")
# text_output = gr.Textbox(label="Prediction")
text_output = gr.Textbox(label="Similarity Score")
gr.Interface(
    fn=Detection,
    inputs=audio_1,
    outputs=text_output,
    title="Audio Deepfake Detection",
    description="Audio Deepfake Detection using finetuned model on for-2seconds dataset.",
).launch()