Triplet Based Embedding Distance and Similarity Learning for Text-independent Speaker Verification
This work addresses speaker verification for security or authentication systems, but it is incremental as it builds on existing x-vector systems.
The paper tackled the problem of text-independent speaker verification by proposing two training improvements based on triplet methods, achieving a 9% reduction in equal error rate and detected cost function on the 2016 NIST SRE Test Set.
Speaker embeddings become growing popular in the text-independent speaker verification task. In this paper, we propose two improvements during the training stage. The improvements are both based on triplet cause the training stage and the evaluation stage of the baseline x-vector system focus on different aims. Firstly, we introduce triplet loss for optimizing the Euclidean distances between embeddings while minimizing the multi-class cross entropy loss. Secondly, we design an embedding similarity measurement network for controlling the similarity between the two selected embeddings. We further jointly train the two new methods with the original network and achieve state-of-the-art. The multi-task training synergies are shown with a 9% reduction equal error rate (EER) and detected cost function (DCF) on the 2016 NIST Speaker Recognition Evaluation (SRE) Test Set.