SdSV Challenge 2020: Large-Scale Evaluation of Short-duration Speaker Verification

Modern approaches to speaker verification represent speech utterancesas fixed-length embeddings. With these approaches,we implicitly assume that speaker characteristics are independentof the spoken content. Such an assumption generally holdswhen sufficiently long utterances are given. In this context,speaker embeddings, like i-vector and x-vector, have shown tobe extremely effective. For speech utterances of short duration(in the order of a few seconds), speaker embeddings have shownsignificant dependency on the phonetic content. In this regard,the SdSV Challenge 2020 was organized with a broad focus onsystematic benchmark and analysis on varying degrees of phoneticvariability on short-duration speaker verification (SdSV).In addition to text-dependent and text-independent tasks, thechallenge features an unusual and difficult task of cross-lingualspeaker verification (English vs. Persian). This paper describesthe dataset and tasks, the evaluation rules and protocols, the performancemetric, baseline systems, and challenge results. Wealso present insights gained from the evaluation and future researchdirections.