S-Vectors and TESA: Speaker Embeddings and a Speaker Authenticator Based on Transformer Encoder

One of the most popular speaker embeddings is x-vectors, which are obtained from an architecture that gradually builds a larger temporal context with layers. In this paper, we propose to derive Transformer’s encoder trained for classification. Self-attention, on built, attends all features over entire utterance and might be more suitable in capturing characteristics utterance. We refer proposed classification model as s-vectors emphasize they heavily relies self-attention. Through experiments, demonstrate perform better than x-vectors. addition s-vectors, also new based verification replacement conventional probabilistic linear discriminant analysis (PLDA). This inspired by next sentence prediction task bidirectional representations Transformers (BERT), feed two utterances verify whether belong same speaker. name Transformer authenticator (TESA). Our experiments show performance TESA PLDA-based verification.