import streamlit as st
from keras.models import load_model
import librosa
import numpy as np
import pickle

# Model ve eğitim tarihçesini yükle
model = load_model('my_model.h5')
with open('pkl.pkl', 'rb') as file_pi:
    history = pickle.load(file_pi)

def detect_fake(sound_file):
    sound_signal, sample_rate = librosa.load(sound_file, res_type="kaiser_fast")
    mfcc_features = librosa.feature.mfcc(y=sound_signal, sr=sample_rate, n_mfcc=40)
    mfccs_features_scaled = np.mean(mfcc_features.T, axis=0)
    mfccs_features_scaled = mfccs_features_scaled.reshape(1, -1)
    result_array = model.predict(mfccs_features_scaled)
    result_classes = ["FAKE", "REAL"]
    result = np.argmax(result_array[0])
    return result_classes[result]

# Streamlit arayüzü
st.title('Ses Doğrulama Sistemi')

uploaded_file = st.file_uploader("Ses dosyası yükle", type=["wav", "mp3", "ogg"])
if uploaded_file is not None:
    # Dosyayı kaydet
    with open(uploaded_file.name, "wb") as f:
        f.write(uploaded_file.getbuffer())
    result = detect_fake(uploaded_file.name)
    st.write(f"Tahmin: {result}")