Stochastic Feature Transformation with Divergence-Based Out-of-Handset Rejection for Robust Speaker Verification

Abstract

The performance of telephone-based speaker verification systems can be severely degraded by linear and nonlinear acoustic distortion caused by telephone handsets. This paper proposes to combine a handset selector with stochastic feature transformation to reduce the distortion. Specifically, a Gaussian mixture model (GMM)-based handset selector is trained to identify the most likely handset used by the claimants, and then handset-specific stochastic feature transformations are applied to the distorted feature vectors. This paper also proposes a divergence-based handset selector with out-of-handset (OOH) rejection capability to identify the “unseen” handsets. This is achieved by measuring the Jensen difference between the selector's output and a constant vector with identical elements. The resulting handset selector is combined with the proposed feature transformation technique for telephone-based speaker verification. Experimental results based on 150 speakers of the HTIMIT corpus show that the handset selector, either with or without OOH rejection capability, is able to identify the “seen” handsets accurately (98.3% in both cases). Results also demonstrate that feature transformation performs significantly better than the classical cepstral mean normalization approach. Finally, by using the transformation parameters of the seen handsets to transform the utterances with correctly identified handsets and processing those utterances with unseen handsets by cepstral mean subtraction (CMS), verification error rates are reduced significantly (from 12.41% to 6.59% on average)

An overview of text-independent speaker recognition: From features to supervectors

“…The graph structure was motivated by invariance against the affine feature distortion model for cepstral features (e.g. [151,155]). The method requires further development to validate the assumptions of the feature distortion model and to improve computational efficiency.…”

Extraction of Speaker Features from Different Stages of DSR Front-Ends for Distributed Speaker Verification

“…In this work, the feature transformation was combined with a handset selector (Tsang et al, 2002;Mak et al, 2004) for robust speaker verification. Specifically, before verification takes place, we compute one set of transformation parameters for each type of handsets that claimants are likely to use.…”

“…Similar to our previous work Mak et al, 2004;Yiu et al, 2003;Tsang et al, 2002), we trained a personalized 32-center GMM to model the characteristics of each client speaker in the system. 1 The feature vectors derived from the SA and SX sentence sets of the corresponding speaker were used for training, i.e., 7 sentences per GMM.…”

“…Because the LMS algorithm used in the ETSI standard is a kind of linear equalization algorithm, it may not perform satisfactorily on telephone handsets with nonlinear characteristics. To overcome the limitation of the LMS algorithm, this paper incorporates a feature transformation algorithm Mak et al, 2004;Tsang et al, 2002Tsang et al, , 2003 into the back-end recognizer to enhance the robustness of the speaker verification system against handset variations. Experiments with and without using the LMS-based blind equalization were also performed for comparison.…”

Blind Stochastic Feature Transformation for Channel Robust Speaker Verification

“…However, this approach may not be practical because users may use a new handset, which is not well represented in the training set, during verification. While this problem can be partially resolved by using a handset classifier with out-of-handset rejection capability [8,9], it is difficult to find a threshold for detecting unseen handsets. On the other hand, unsupervised (blind) compensation does not assume any knowledge of the channel characteristics.…”