DG^VoiC Voice Clustering Achieves 96% AMI for Fraud Detection
June 29, 2026
DG^VoiC is a voice clustering framework designed for real call-centre conditions, utilizing sliding-window speaker embeddings and cosine similarity. The system achieves 96% Adjusted Mutual Information (AMI) and 95% Adjusted Rand Index (ARI) for identifying repeated speakers across anonymized audio.
HOW THIS AFFECTS YOU
●
builderYou can implement this clustering approach to link identities across telephony data.
●
founderThis provides a scalable method for reducing insurance fraud via speaker identity signals.