In this paper, we discuss the quality assessment of a new method for thegeneration of a masker for speech privacy protection. This masker includes speech characteristicsthat prevent eavesdroppers from overhearing conversations in public spaces.Previous research shows that maskers generated from the target speech perform betterin interfering with the listening, than the other maskers. Therefore, we propose a bootstrap(BS) masker method that efficiently generates a masker from a small sample of therecorded speech. We evaluate the subjective speech intelligibility and establish that the BSmasker can achieve the same level of intelligibility as that of the conventional additionalmasker at an approximately 4 dB lower target-to-masker ratio.
雑誌名
Journal of Information Hiding and Multimedia Signal Processing
A Case Study of Bootstrap Masker Quality Assessment for Speech-Privacy Protection
JIHMSP_8_6_1460_1469
(6.75MB)
