Multi-Label Training for Text-Independent Speaker Identification
Yuqi Xue
TL;DR
This work tackles robust text-independent speaker identification by introducing Multi-Label Training (MLT), which partitions each speaker's data into $N$ subgroups with distinct label sets and trains over an expanded output space of $N\times C$, enabling ensemble-like benefits without substantial overhead. The method is implemented on a Shon et al.-based baseline, expanding the final layer and balancing hidden-layer ratios, and augmented with a speech enhancement module that feeds into a 1D-convolutional speaker-ID backbone. The authors demonstrate that MLT improves identification accuracy in both clean and enhanced noisy conditions on VoxCeleb1, with the best gains typically at $N=2$ for their setup, and show that speech enhancement is particularly beneficial for robustness. The proposed approach is easily applicable to other text-independent speaker-ID models, offering a practical route to improved performance without large computational costs.
Abstract
In this paper, we propose a novel strategy for text-independent speaker identification system: Multi-Label Training (MLT). Instead of the commonly used one-to-one correspondence between the speech and the speaker label, we divide all the speeches of each speaker into several subgroups, with each subgroup assigned a different set of labels. During the identification process, a specific speaker is identified as long as the predicted label is the same as one of his/her corresponding labels. We found that this method can force the model to distinguish the data more accurately, and somehow takes advantages of ensemble learning, while avoiding the significant increase of computation and storage burden. In the experiments, we found that not only in clean conditions, but also in noisy conditions with speech enhancement, Multi-Label Training can still achieve better identification performance than commom methods. It should be noted that the proposed strategy can be easily applied to almost all current text-independent speaker identification models to achieve further improvements.