Verbal Information Verification

Cantonese verbal information verification system using GMM-based anti-model

“…However, verbal information is obviously a useful knowledge for authentication. Verbal information verification was proposed as a complementary technique for speaker authentication [1]. It requires the user to speak out his/her personal information such as name, birth date and residence address in order to verify the identity.…”

“…UV is preferred to ASR since the ASR approach doesn't effectively utilize the registered information in the user's profile [1]. In a UV system, there are three key modules that determine the system's performance [1].…”

“…This is done with speaker-independent (SI) acoustic models. Afterwards, for each subword, the likelihood scores from the corresponding target model and the anti-model are computed for hypothesis test [1] [2]. Decisions will be made on both the subword and the utterance levels.…”

“…In this study, the confidence measure for subword unit n in an observed speech segment O n is defined as [1], where n λ and n λ denote respectively the corresponding target model and anti-model for the subword unit n. For utterance-level decision, the results of subword tests need to be combined. In this work, we adopt the normalized confidence measure as defined in [1]. If the utterance contains N subword units, the normalized confidence measure is given by,…”

“…

ABSTRACTVerbal information verification (VIV) is one of the approaches for speaker authentication [1]. It is a process in which the spoken utterance of a claimed speaker is verified against the key information in speaker's registered profile.

…”