[P] Deezer showed CNN detection fails on compressed audio, here's a dual-engine approach that survives MP3
About this article
I've been working on detecting AI-generated music and ran into the same wall that Deezer's team documented in their paper, CNN-based detection on mel-spectrograms breaks when audio is compressed to MP3. The problem: A ResNet18 trained on mel-spectrograms works well on WAV files, but real-world music is distributed as MP3/AAC. Compression destroys the subtle spectral artifacts the CNN relies on. What actually worked: Instead of trying to make the CNN more robust, I added a second engine based ...
You've been blocked by network security.To continue, log in to your Reddit account or use your developer tokenIf you think you've been blocked by mistake, file a ticket below and we'll look into it.Log in File a ticket