Spaces:
Runtime error
Runtime error
File size: 2,978 Bytes
41d4efd 817f64d 41d4efd 817f64d 41d4efd a6fde96 41d4efd 817f64d 41d4efd 817f64d 5de0202 16d09d3 817f64d 41d4efd 817f64d acba8bc dd50305 817f64d |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 |
import gradio as gr
import numpy as np
import librosa
from tensorflow.keras.models import load_model
import matplotlib.pyplot as plt
# Constants
MAX_TIME_STEPS = 109
SAMPLE_RATE = 16000
DURATION = 5
N_MELS = 128
MODEL_PATH = "audio_classifier.h5" # Replace with the actual path to your saved model
# Load the pre-trained model
model = load_model(MODEL_PATH, compile=False)
model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])
def visualize(mel_spectrogram):
median_decibels = np.median(mel_spectrogram)
median_human_voice_range = -65
diff_decibels = abs(median_decibels - median_human_voice_range)
plt.figure(figsize=(12, 6))
plt.subplot(1, 1, 1)
librosa.display.specshow(mel_spectrogram, sr=SAMPLE_RATE, x_axis='time', y_axis='mel')
plt.colorbar(format='%+2.0f dB')
plt.title(f'Difference from Median Human Voice Range: {diff_decibels:.2f} dB')
plt.savefig("mel_spectrogram.png") # Save the image
plt.close()
def classify_audio(audio):
# Convert the audio data to NumPy array
rate, ar = audio
arone = ar.astype(np.float32)
if arone.ndim > 1:
arone = arone.flatten()
mel_spectrogram = librosa.feature.melspectrogram(y=arone, sr=SAMPLE_RATE, n_mels=N_MELS)
mel_spectrogram = librosa.power_to_db(mel_spectrogram, ref=np.max)
# Ensure all spectrograms have the same width (time steps)
if mel_spectrogram.shape[1] < MAX_TIME_STEPS:
mel_spectrogram = np.pad(mel_spectrogram, ((0, 0), (0, MAX_TIME_STEPS - mel_spectrogram.shape[1])), mode='constant')
else:
mel_spectrogram = mel_spectrogram[:, :MAX_TIME_STEPS]
# Reshape for the model
X_test = np.expand_dims(mel_spectrogram, axis=-1)
X_test = np.expand_dims(X_test, axis=0)
# Predict using the loaded model
y_pred = model.predict(X_test)
# Convert probabilities to predicted classes
y_pred_classes = np.argmax(y_pred, axis=1)
if y_pred_classes[0] == 1:
prediction = "Not Spoof : High chances of original voice"
else:
prediction = "Spoof : Possible voice cloning"
median_decibels = np.median(mel_spectrogram)
median_human_voice_range = -65
diff_decibels = abs(median_decibels - median_human_voice_range)
if diff_decibels<3 and y_pred_classes[0] == 0 :
prediction = "Not Spoof : High chances of original voice"
visualize(mel_spectrogram)
return prediction,"mel_spectrogram.png"
title=" Group-2 Audio Spoof detection using CNN"
description="The model was trained on the ASVspoof 2019 dataset with an aim to detect spoof audios through deep learning.To use it please upload an audio file of suitable length. The Mel spectrogram used for inferencing is also available for the user to understand the classification and compare it with the median Human decibel range."
iface = gr.Interface(classify_audio, inputs=["audio"], outputs=["text","image"],title=title,description=description)
iface.launch() |