Applied ML with an evaluation protocol that refuses to fool itself: a CNN plus 11 spectral rules catch lossy transcodes hiding in lossless audio, and a conviction needs two independent evidence families to agree. 11K files validated, zero false positives — failed approaches published alongside. CLI + Python API + Docker.
python docker machine-learning cnn pytorch audio-analysis flac vorbis opus spectral-analysis model-evaluation cli-tool pypi-package applied-ml lossless-audio fake-flac audio-repair mp3-detection transcode-detection
-
Updated
Sep 4, 2026 - Python