Skip to content

Audio Detection Guide

Comprehensive guide to detecting synthetic voice and deepfake audio using Veridex.

Overview

Veridex offers multiple audio detection signals, from lightweight spectral analysis to foundation model-based detection.

Available Detectors

1. SpectralSignal (Fastest)

from veridex.audio import SpectralSignal

detector = SpectralSignal()
result = detector.run("audio.wav")

print(f"AI Probability: {result.score:.2%}")

Best for: Quick screening, real-time applications

2. AASISTSignal (Anti-Spoofing)

from veridex.audio import AASISTSignal

detector = AASISTSignal()
result = detector.run("audio.wav")

print(f"AI Probability: {result.score:.2%}")

Best for: Voice spoofing detection

3. Wav2VecSignal (Highest Accuracy)

from veridex.audio import Wav2VecSignal

detector = Wav2VecSignal()
result = detector.run("audio.wav")

print(f"AI Probability: {result.score:.2%}")

Best for: Production-grade detection

4. SilenceSignal (Pattern Analysis)

from veridex.audio import SilenceSignal

detector = SilenceSignal()
result = detector.run("audio.wav")

print(f"AI Probability: {result.score:.2%}")

Best for: Detecting synthetic speech patterns

Best Practices

  1. Ensure proper audio format (WAV recommended)
  2. Check sample rate (16kHz optimal)
  3. Combine multiple signals
  4. Consider computational resources

Next Steps