Matthieu Hebert

Parameterization of the score threshold for a text-dependent adaptive speaker verification system

We present a computationally efficient strategy for setting a priori thresholds in an adaptive speaker verification system. We have two motivations: to eliminate the externally preset overall system thresholds and replace them with automatically-set internal thresholds conditioned by a target FA rate and calculated at runtime; to counter the verification score shifts resulting from online adaptation. Our approach entails calculating the trajectory of the score threshold as a function of 1) length of the password, 2) target FA, 3) the number of training frames in the speaker model. The solution is successful at both achieving the target FA rates and keeping the FA rate constant during online adaptation. Furthermore, it is algorithmically simple and requires negligible computational resources. The threshold function is calibrated on a Japanese database and experimental results are presented on 12 databases in four different languages.

T-Norm for text-dependent commercial speaker verification applications: effect of lexical mismatch

We describe a test-time score normalization technique (T-Norm) for text-dependent speaker verification that is robust to lexical mismatch. The main challenge to the deployment of T-Norm in a text-dependent task is the mismatch between the lexicon of the target speaker model in the application and that of the cohort speaker models. We show the negative effect of that mismatch in controlled experiments and propose a hybrid scoring scheme (T-Norm and background model) to remedy it. In a lexically mismatched scenario, which is inherent to the deployment of T-Norm in a text-dependent system, we show a 31% relative error rate reduction using the hybrid scoring over T-Norm alone. A 22% relative error rate reduction is measured over the baseline (no T-Norm) system.

Improved normalization without recourse to an impostor database for speaker verification

Score normalization has become an important facet of modern speaker verification systems. That is, the score of the verification attempt with the claimant voiceprint is usually normalized by the score from a background model or a cohort model. Typically, the determination of these normalizing models requires a priori a database of impostors and is clearly incompatible with real world applications. We propose a scheme to generate a normalizing model by biasing a speaker independent model with the customers enrollment tokens. For each password of each customer, there exists a privileged direction in model parameter space defined by the speaker independent and dependent models. By changing the location of the speaker independent model along this direction, a family of modified normalizing models can be generated without any knowledge about the impostors. In the difficult task of speaker verification with same-gender impostors, who know the password, 10%-15% error rate reduction is achieved using various biasing mechanisms.

Desperately seeking impostors: data-mining for competitive impostor testing in a text-dependent speaker verification system

Precise determination of the operating point of a real-world verification application is of great importance. For a text-dependent password-based security system, this can be a challenging task, as lexically matched impostor test data may be nonexistent. We present a data mining approach for extracting suitable impostor data. The approach may be applied to either the target database (the application data itself) or the stock databases (data from other applications). The method entails: 1) determining Levenstein distances of impostor text utterances with respect to the claimant password; 2) selecting subsets of impostor data at various levels of lexical distance; 3) calculating the score threshold using such subsets; 4) extrapolating the score threshold (and hence the operating point) for lexically perfectly-matched data. Experiments on four databases in two languages are presented. This approach, as applied to the target database, provides an accurate and inexpensive solution to a formidable real-world problem.