play_lesson Forensic Masterclasses & Step-by-Step Guides

Practical Forensic Tutorials

Learn how to operate professional forensic auditing tools, analyze spectrograms, humanize essays for academic defense, and configure enterprise workflows.

Tutorial 01 Audio Forensics • 4 Steps • 5 mins

How to Detect Voice Deepfakes in 4 Steps Using Spectral & Jitter Analysis

When suspicious audio is sent via WhatsApp, voicemail, or podcast, follow this exact laboratory protocol in SmrAI Voice Studio:

Step 1: Upload / Ingest

Drag the raw audio into Voice Studio. The audio is decoded into uncompressed 48 kHz PCM without lossy compression.

Step 2: Inspect 16 kHz

Toggle Layer 1 (Mel-Spectrogram). Look for an abrupt horizontal black void above 16 kHz—a telltale sign of neural vocoders.

Step 3: Measure F0 Jitter

Switch to Layer 2 (Pitch Jitter). Real human voices display 0.65% to 1.25% variance. Jitter below 0.05% indicates robotic flatness.

Step 4: Issue Audit Cert

Click Download Audit Cert to export a cryptographically signed SHA-256 PDF for legal or journalistic verification.

Try in Voice Studio arrow_forward

Tutorial 02 Academic Integrity • 3 Steps • 4 mins

How to Humanize Academic Essays While Protecting Citations & Technical Vocabulary

When an original student paper is falsely flagged for high AI probability, standard paraphrasers ruin citations. Here is how to restore human cadence safely:

Step 1: Lock Citations

In AI Humanizer Pro, verify that "Preserve In-Text Citations" is active. All brackets (e.g. [Vance et al., 2026]) are strictly locked.

Step 2: Inject Natural Burstiness

Select Academic Defense Mode. The engine restructures sentence lengths, breaking uniform mechanical cadence into organic rhythms.

Step 3: Verify with Multi-Detector

Click Cross-Verify to test the output simultaneously against Turnitin v3, GPTZero, and Winston AI with zero storage.

Try in Humanizer Studio arrow_forward

Tutorial 03 Image Forensics • 3 Steps • 3 mins

How to Spot Midjourney v6 & Flux Diffusion Artifacts via Error Level Analysis (ELA)

Modern generative images look flawless to the naked eye. Learn to use mathematical error residual layers to identify artificial generation:

Step 1: Upload to Vision Studio

Ingest the image in Image Detector. The pipeline parses embedded EXIF and C2PA Content Credentials.

Step 2: Activate ELA Lens

Error Level Analysis highlights compression differences. AI generated regions produce uniform glow whereas real camera sensors exhibit noisy photonic grain.

Step 3: DCT High-Frequency Pass

Check the frequency spectrum. Generative upsamplers leave grid-like checkerboard harmonics in high-frequency cosine coefficients.

Try in Image Studio arrow_forward