Cluster-Guided Unsupervised Domain Adaptation for Deep Speaker Embedding
This work addresses the problem of speaker verification across domains without target labels, offering a significant performance improvement for applications in speech technology.
The paper tackles unsupervised domain adaptation for speaker verification by proposing a cluster-guided framework that uses clustering to pseudo-label target domain data and trains a speaker embedding network with contrastive center loss, achieving an equal error rate of 8.10% on CN-Celeb1, which outperforms the supervised baseline by 39.6% and sets a state-of-the-art result.
Recent studies have shown that pseudo labels can contribute to unsupervised domain adaptation (UDA) for speaker verification. Inspired by the self-training strategies that use an existing classifier to label the unlabeled data for retraining, we propose a cluster-guided UDA framework that labels the target domain data by clustering and combines the labeled source domain data and pseudo-labeled target domain data to train a speaker embedding network. To improve the cluster quality, we train a speaker embedding network dedicated for clustering by minimizing the contrastive center loss. The goal is to reduce the distance between an embedding and its assigned cluster center while enlarging the distance between the embedding and the other cluster centers. Using VoxCeleb2 as the source domain and CN-Celeb1 as the target domain, we demonstrate that the proposed method can achieve an equal error rate (EER) of 8.10% on the CN-Celeb1 evaluation set without using any labels from the target domain. This result outperforms the supervised baseline by 39.6% and is the state-of-the-art UDA performance on this corpus.