File size: 8,457 Bytes
d8240b4 |
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 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 |
import os
import torch
import gradio as gr
import numpy as np
import librosa
import soundfile as sf
from transformers import Wav2Vec2FeatureExtractor, Wav2Vec2ForSequenceClassification
from pydub import AudioSegment
import tempfile
import matplotlib
matplotlib.use('Agg')
# Constants
MODEL_ID = "MelodyMachine/Deepfake-audio-detection-V2"
SAMPLE_RATE = 16000
MAX_DURATION = 30 # maximum audio duration in seconds
class DeepfakeDetector:
def __init__(self, model_id=MODEL_ID):
self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(f"Using device: {self.device}")
print(f"Loading model from {model_id}...")
self.feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained(model_id)
self.model = Wav2Vec2ForSequenceClassification.from_pretrained(model_id).to(self.device)
print("Model loaded successfully!")
# Labels for classification
self.id2label = {0: "Real", 1: "Deepfake"}
def preprocess_audio(self, audio_path):
"""Process audio file to match model requirements."""
try:
# Load audio file
y, sr = librosa.load(audio_path, sr=SAMPLE_RATE, mono=True)
# Trim silence from the beginning and end
y, _ = librosa.effects.trim(y, top_db=20)
# If audio is longer than MAX_DURATION seconds, take the first MAX_DURATION seconds
if len(y) > MAX_DURATION * SAMPLE_RATE:
y = y[:MAX_DURATION * SAMPLE_RATE]
return y
except Exception as e:
raise ValueError(f"Error preprocessing audio: {str(e)}")
def detect(self, audio_path):
"""Detect if audio is real or deepfake."""
try:
# Preprocess audio
audio_array = self.preprocess_audio(audio_path)
# Extract features
inputs = self.feature_extractor(
audio_array,
sampling_rate=SAMPLE_RATE,
return_tensors="pt",
padding=True
).to(self.device)
# Get prediction
with torch.no_grad():
outputs = self.model(**inputs)
logits = outputs.logits
predictions = torch.softmax(logits, dim=1)
# Get results
predicted_class = torch.argmax(predictions, dim=1).item()
confidence = predictions[0][predicted_class].item()
result = {
"prediction": self.id2label[predicted_class],
"confidence": float(confidence),
"probabilities": {
"Real": float(predictions[0][0].item()),
"Deepfake": float(predictions[0][1].item())
}
}
return result
except Exception as e:
raise ValueError(f"Error during detection: {str(e)}")
def convert_audio(input_file):
"""Convert the audio file to the required format."""
# Create temp file with .wav extension
temp_dir = tempfile.gettempdir()
temp_path = os.path.join(temp_dir, "temp_audio_file.wav")
# Handle various input formats
if input_file.endswith('.mp3'):
audio = AudioSegment.from_mp3(input_file)
audio = audio.set_channels(1) # Convert to mono
audio = audio.set_frame_rate(SAMPLE_RATE) # Set sample rate
audio.export(temp_path, format="wav")
elif input_file.endswith('.wav'):
audio = AudioSegment.from_wav(input_file)
audio = audio.set_channels(1) # Convert to mono
audio = audio.set_frame_rate(SAMPLE_RATE) # Set sample rate
audio.export(temp_path, format="wav")
elif input_file.endswith('.ogg'):
audio = AudioSegment.from_ogg(input_file)
audio = audio.set_channels(1) # Convert to mono
audio = audio.set_frame_rate(SAMPLE_RATE) # Set sample rate
audio.export(temp_path, format="wav")
elif input_file.endswith('.flac'):
audio = AudioSegment.from_file(input_file, format="flac")
audio = audio.set_channels(1) # Convert to mono
audio = audio.set_frame_rate(SAMPLE_RATE) # Set sample rate
audio.export(temp_path, format="wav")
else:
# Try to convert using pydub's generic from_file
try:
audio = AudioSegment.from_file(input_file)
audio = audio.set_channels(1) # Convert to mono
audio = audio.set_frame_rate(SAMPLE_RATE) # Set sample rate
audio.export(temp_path, format="wav")
except:
raise ValueError(f"Unsupported audio format for file: {input_file}")
return temp_path
def detect_deepfake(audio_file, detector):
"""Process audio and detect if it's a deepfake."""
if audio_file is None:
return {
"error": "Please upload an audio file."
}
try:
# Convert audio to required format
processed_audio = convert_audio(audio_file)
# Detect deepfake
result = detector.detect(processed_audio)
# Create a visually appealing output
prediction = result["prediction"]
confidence = result["confidence"] * 100
# Prepare visualization data
labels = list(result["probabilities"].keys())
values = list(result["probabilities"].values())
output = {
"prediction": prediction,
"confidence": f"{confidence:.2f}%",
"chart_labels": labels,
"chart_values": values
}
# Create result text with confidence
result_text = f"Prediction: {prediction} (Confidence: {confidence:.2f}%)"
return result_text, output
except Exception as e:
return f"Error: {str(e)}", None
def create_interface():
"""Create Gradio interface for the application."""
# Initialize the deepfake detector
detector = DeepfakeDetector()
with gr.Blocks(title="Deepfake Voice Detector") as interface:
gr.Markdown("""
# Deepfake Voice Detector
Upload an audio file to check if it's a real human voice or an AI-generated deepfake.
**Model:** MelodyMachine/Deepfake-audio-detection-V2 (Accuracy: 99.73%)
""")
with gr.Row():
with gr.Column(scale=1):
audio_input = gr.Audio(
type="filepath",
label="Upload Audio File",
sources=["upload", "microphone"]
)
submit_btn = gr.Button("Analyze Audio", variant="primary")
with gr.Column(scale=1):
result_text = gr.Textbox(label="Result")
# Visualization component
with gr.Accordion("Detailed Analysis", open=False):
gr.Markdown("### Confidence Scores")
confidence_plot = gr.Plot(label="Confidence Scores")
# Process function for the submit button
def process_and_visualize(audio_file):
result_text, output = detect_deepfake(audio_file, detector)
if output:
# Create bar chart visualization
import matplotlib.pyplot as plt
fig, ax = plt.subplots(figsize=(6, 4))
bars = ax.bar(output["chart_labels"], output["chart_values"], color=['green', 'red'])
# Add percentage labels on top of each bar
for bar in bars:
height = bar.get_height()
ax.text(bar.get_x() + bar.get_width()/2., height + 0.02,
f'{height*100:.1f}%', ha='center', va='bottom')
ax.set_ylim(0, 1.1)
ax.set_title('Confidence Scores')
ax.set_ylabel('Probability')
return result_text, fig
else:
return result_text, None
submit_btn.click(
process_and_visualize,
inputs=[audio_input],
outputs=[result_text, confidence_plot]
)
return interface
if __name__ == "__main__":
interface = create_interface()
interface.launch() |