Speaker- and gender-dependent maskers for efficient speech privacy protection

We investigated on an efficient masker for speech privacy protection. We first proposed a speaker-dependent (SD) masker created from speech spoken by the speaker to be masked (maskee). This masker decreased the speech intelligibility to a much lower level than conventional babble maskers, by up to 40%. However, this masker needs to be prepared for each potential speaker, and may be expensive to create and maintain. We also investigated a same gender single-speaker (SSGD) masker, where the masker is created from a different speaker of the same gender. We found that this masker works just as well as the SD maskers for some speakers, while it was not as efficient for some other speakers. Accordingly, we also created gender-dependent maskers from multiple same-gender speakers (MSGD). This masker was proven to mask as well as the SD maskers for all the same gender speakers tested. We now only need to maintain two maskers, one masker per gender, which makes the maintenance and administration of these maskers much more efficient. In order to mask a speaker, we merely need to determine the gender of the speaker, and play out the gender-dependent masker of the estimated gender.