Design Choices for X-Vector Based Speaker Anonymization

The recently proposed x-vector based anonymization scheme converts any input into that of a random pseudo-speaker. In this paper, we present a pseudo-speaker selection technique as a baseline for the first Challenge. We explore several design choices for the distance between speakers, the region of x-vector space where the pseudo-speaker picked, and gender selection. To assess the strength of anonymization, we consider attackers using an x-vector based speaker verification who may use original or anonymized speech for enrollment, depending on knowledge of the anonymization scheme. The Equal Error Rate (EER) by the attackers and the decoding Word Error Rate (WER) over data are reported as the measures of privacy and utility. are performed using datasets derived from LibriSpeech to find the combination of design choices in terms of privacy and utility.