Weakly Supervised PLDA Training
This work addresses the problem of expensive data labeling for speaker verification systems, offering an incremental improvement in efficiency for domain-specific applications.
The paper tackles the high cost of labeled data for PLDA training in speaker verification by proposing a weakly supervised approach using cheap, imperfect labels, achieving good performance on real-life telephony data when labeled data is limited.
PLDA is a popular normalization approach for the i-vector model, and it has delivered state-of-the-art performance in speaker verification. However, PLDA training requires a large amount of labelled development data, which is highly expensive in most cases. We present a cheap PLDA training approach, which assumes that speakers in the same session can be easily separated, and speakers in different sessions are simply different. This results in `weak labels' which are not fully accurate but cheap, leading to a weak PLDA training. Our experimental results on real-life large-scale telephony customer service achieves demonstrated that the weak training can offer good performance when human-labelled data are limited. More interestingly, the weak training can be employed as a discriminative adaptation approach, which is more efficient than the prevailing unsupervised method when human-labelled data are insufficient.