Metric Learning Loss Functions to Reduce Domain Mismatch in the x-Vector Space for Language Recognition

Abstract

State-of-the-art language recognition systems are based on discriminative embeddings called x-vectors. Channel and gender distortions produce mismatch in such x-vector space where embeddings corresponding to the same language are not grouped in an unique cluster. To control this mismatch, we propose to train the x-vector DNN with metric learning objective functions. Combining a classification loss with the metric learning n-pair loss allows to improve the language recognition performance. Such a system achieves a robustness comparable to a system trained with a domain adaptation loss function but without using the domain information. We also analyze the mismatch due to channel and gender, in comparison to language proximity, in the x-vector space. This is achieved using the Maximum Mean Discrepancy divergence measure between groups of x-vectors. Our analysis shows that using the metric learning loss function reduces gender and channel mismatch in the x-vector space, even for languages only observed on one channel in the train set.

Investigating model performance in language identification: beyond simple error statistics

“…A total of 7 and 6 systems were submitted to the closed and open tracks of MERLIon CCS challenge respectively. As the LID task (English and Mandarin) is binary and the challenge dataset is highly imbalanced, equal error rate was selected as a primary evaluation metric [23].…”

Cross-corpora spoken language identification with domain diversification and generalization

“…First, this efficiently minimizes the distribution shift between source and target data [32]. Second, the language recognition literature successfully used MMD for reducing domain mismatch [22].…”

Language Recognition on Unknown Conditions: The LORIA-Inria-MULTISPEECH System for AP20-OLR Challenge

“…• additive angular margin softmax (AAM) [27] as an alternative classification loss function , we used a margin parameter m = 0.1 and a radius s = 40 • regularization of the cross-entropy with maximum mean discrepancy (MMD) between mobile channel and unknown channels [10], we compared different weights λ for the regularization loss functions • regularization of the cross-entropy with n-pair loss [28]…”