TLDR AI · RESEARCH & INNOVATION
SPEAKER IDENTIFICATION WITH WHISPER
Summary from the TLDR AI newsletter
WSI repurposes the Whisper ASR encoder for multilingual speaker identification using joint loss optimization. It outperforms Pyannote, ECAPA TDNN, and Xvector in recognizing speakers across diverse languages and environments.