Subphase
Listening beneath the surface. An adversarial audio detection library combining signal processing and deep learning to identify manipulated audio.
An adversarial audio detection library, preparing for release on PyPI.
- Dual-layer digital signal processing (DSP) and convolutional neural network (CNN) architecture.
- Validated a mel-spectrogram and raw-waveform ensemble at 0.932 balanced accuracy.
- Won 1st place at the Devin Hackathon: Agent of Chaos in Vancouver, July 2026.
- Built with Python, FastAPI, NumPy, SciPy, and Librosa.





