Speaker recognition over HF radio after automatic speaker segmentation

The author presents a new method to segment speakers in HF radio conversations. An experiment is conducted to evaluate how this segmentation method affects automatic speaker identification. It is found that, for the database, the segmentation method does not degrade identification performance relative to manual speaker segmentation and also that the automatic speaker identification is promising when compared to human performance. The author discusses problems associated with speaker recognition in general (and specifically HF radio) and reviews several techniques from the literature that may be applied to address these problems.