README.md

metadata

title: Audio Forensic Analyzer
emoji: 🎵
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 6.1.0
app_file: app.py
pinned: false

🎧 Audio Forensic Analyzer

A complete forensic-grade audio analysis tool running entirely in HuggingFace Spaces.
Uploads any audio file and generates a detailed PNG report + Markdown summary.

---

🚀 Features

🔍 Forensic Analysis

• Full spectrogram visualization
• Peak, RMS, Crest Factor
• Noise floor & SNR
• Zero-Crossing Rate
• Loudness (LUFS, ITU BS.1770)

🎚 Spectral Analysis

• 4096-point STFT
• Energy distribution across speech bands
• 85% and 95% rolloff
• Highest usable frequency (–60 dB)
• Spectral centroid, bandwidth, flatness

🧪 Artifact Detection

Detects:

• Low-pass filters (LPF)
• High-pass filters (HPF)
• Brick-wall filtering
• Compression / Over-compression
• Noise reduction artifacts
• Spectral notches
• Clipping
• De-esser behavior
• Multiband compression

🤖 Synthetic Voice Probability (Informational Only)

A lightweight heuristic AI-vs-Human estimator using:

• MFCC stability
• Pitch jitter
• HF spectral smoothness

Does NOT affect scoring.
Displayed in the PNG report as a separate block.

---

📊 Output

The tool generates:

1. PNG Forensic Report

Includes:

• Spectrogram
• File metadata
• Time-domain metrics
• Spectral details
• Energy distribution chart
• Issue list
• Quality score, grade, and recommendation
• Synthetic voice probability block

2. Markdown Summary

Clean, readable, fully formatted analysis results.

---

📁 Usage

1. Upload any audio file (WAV, MP3, FLAC, AAC, OGG, M4A).
2. Click Analyze Audio.
3. View the PNG report and the summary.

Works fully offline inside the Space — no external API calls.

---