Audio Analysis Tool

Audio Fake
Detector PRO

Detects lossy-to-lossless transcodes and artificially upscaled audio files mislabeled as high-quality sources — using spectral analysis and half-or-strict-majority wall voting across multiple segments.

v6.1.1  ·  Windows 10/11 (x86 / x64)  ·  PowerShell 5+  ·  © 2026 Alessandro Comito

↓ Download ZIP archive via Google Drive  ·  Standard + Portable editions

Overview

Audio Fake Detector PRO analyses whether an audio file genuinely contains the frequency content its container bitrate implies. It targets the common case of lossy material (MP3, AAC) re-encoded into lossless containers (FLAC, WAV) or inflated to a higher lossy bitrate to appear higher quality than it actually is.

The engine splits each file into multiple non-final segments plus one end segment. Each non-final segment is analysed independently via spectrogram bitmap inspection. A half-or-strict-majority wall vote aggregates the per-segment findings into a single file-level verdict. Lossless files are additionally validated through auCDtect statistical PCM analysis.

[OK] Full-band content consistent with the declared bitrate. No spectral wall detected.
[SUSPECT] Best segment falls in the 16.5–18.5 kHz grey zone. Possibly high-bitrate lossy; manual review advised.
[FAKE] Half or strict majority of non-final segments show a spectral wall ≤ 16.5 kHz. Declared bitrate is not representative.

Supported Formats

Lossy

MP3 AAC M4A (AAC) OGG Vorbis OPUS WMA

Lossless

FLAC WAV AIFF AIF APE WavPack (WV)

Detection Engine

Each file is split into non-final segments used for voting, plus one end segment used only as a secondary indicator. Every non-final segment is scanned via a GDI+ LockBits spectrogram for high-frequency cutoff behaviour, roll-off ratio, peak energy, and anomaly flags.

HALF-OR-STRICT-MAJORITY WALL VOTE

Rule  (N = non-final segments):
  N even  → FAKE if walled ≥ N/2   (tie counts as FAKE)
  N odd   → FAKE if walled > N/2   (strict majority required)

Spectral wall threshold:  ≤ 16 500 Hz
Grey zone (excl. from vote):  16 500 – 18 500 Hz → SUSPECT
Example N = 3:  2/3 walled → FAKE  (66.7 %)
Example N = 4:  2/4 walled → FAKE  (50 %, tie-break)

Analysis Methods

Processing Pipeline

Format detection & decoding ffprobe / ffmpeg — 3-step metadata probe with size-based fallback for lossless
Segment selection Non-final slots at fixed intervals + one random segment + one end segment
Spectrogram extraction ffmpeg → bitmap · GDI+ LockBits pixel scan for cutoff, ratio, peak
Per-segment analysis Wall detection · abrupt-end check · joint-stereo anomaly · NR flag
Vote aggregation Half-or-strict-majority wall vote across non-final segments
Final classification [OK] · [SUSPECT] · [FAKE] — with cutoff Hz and bitrate context in log

Output

Technical Notes

Platform
Windows 10 / 11 · 32-bit & 64-bit
Runtime
PowerShell 5 or higher
Dependencies
ffmpeg · ffprobe · auCDtect — auto-downloaded on first run
Build
PS2EXE — high-performance PowerShell executable
Temp (Standard)
%LOCALAPPDATA%\AudioFakeDetector\
Temp (Portable)
.\Data\App\Temp\ — deleted after each run
Character support
Full Unicode — Japanese, symbols, hidden files
Portable edition
Zero footprint — no admin rights, no system-wide changes

Limitations

The tool is less reliable on the following material:

License

This software is freeware. It is provided "AS IS", without warranty of any kind. The source code is provided for reference purposes only.

You are allowed to view and study the source code; however, you are not permitted to modify, redistribute, reuse, or create derivative works based on it, in whole or in part, without explicit written permission from the author. This notice applies to all versions of the software, including any previous releases.