On the Use of Nearest Feature Line for Speaker Identification

摘要

As a new pattern classification method. Nearest Feature Line (NFL) provides an effective way to tackle the pattern recognition problems where limited data are available for training. In this paper, we explore the use of NFL for speaker identification in terms of limited data. In order to speed up NFL in decision-making, we propose an alternative method for similarity measure. We have applied the improved NFL to speaker identification in terms of different operating modes. Its performance in the text-dependent case is satisfactory and comparable with the Dynamic Time Warping (DTW) on the Ti46 corpus, while its computational load is much lower than that of DTW. For the text-independent case, we employ the NFL to be a new similarity measure in Vector Quantization (VQ), which causes the VQ to perform better on the KING corpus. Some computational issues on the NFL are also addressed in this paper.