Module 02

Spectral Analysis Basics

Spectral analysis is the foundation of audio forensics. This module introduces frequency-domain techniques for identifying synthetic speech.

Understanding the Frequency Domain

All audio can be represented as a combination of frequencies. A spectrogram visualizes these frequencies over time, revealing patterns invisible to the ear. Synthetic speech leaves distinct frequency-domain artifacts.

Reading Spectrograms

Key features to examine include formant patterns (resonant frequencies of vowels), harmonic spacing, noise floor characteristics, and transition smoothness between phonemes. AI-generated speech often shows unnaturally uniform formant patterns.

// Frequency analysis parameters Sample Rate: 44.1 kHz FFT Size: 2048 Window: Hann Overlap: 75% Frequency Range: 20 Hz - 20 kHz

Detecting Voice Cloning

Voice cloning technology has advanced significantly, but current models still produce detectable artifacts in the frequency domain. Learn to identify these markers and build confidence in your spectral analysis skills.