AI Voice Turing Test Model

Built for the Social Bias Hackathon hosted by Ethical Spectacle Research and Arizona State University. The system combines an audio feature pipeline (MFCCs, pitch, cadence, spectral entropy extracted with Librosa) with NLP sentiment analysis on call transcripts to classify whether a caller is human or AI-generated. Beyond classification accuracy, the project included a demographic bias audit: evaluating whether the model's error rates varied across population groups and documenting those disparities. Earned 3rd place recognition from ASU and Ethical Spectacle Research.