This paper presents results of speaker recognition experiments using short Polish sentences. We developed and analyzed various vector quantization representations in order to first maximize identification effectiveness and second to compare VQ (vecto