HMM-Based Phrase-Independent i-Vector Extractor for Text-Dependent Speaker Verification

• http://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=7902120
• http://www.fit.vutbr.cz/research/groups/speech/publi/2017/zeinali_ieee_acm%20transactions2017_07902120.pdf PDF

Bottleneck features, DNN, hidden Markov model (HMM), i-vector, text-dependent speaker verification.

This article is describes a new HMM structure for text-dependent speaker verification, which enabled the authors to use the potential of the HMM to model time sequences along with the established i-vector technique.

Abstract-The low-dimensional i-vector representation of speech segments is used in the state-of-the-art text-independent speaker verification systems. However, i-vectors were deemed unsuitable for the text-dependent task, where simpler and older speaker recognition approaches were found more effective. In this work,we propose a straightforward hiddenMarkovmodel (HMM) based extension of the i-vector approach, which allows i-vectors to be successfully applied to text-dependent speaker verification. In our approach, the Universal Background Model (UBM) for training phrase-independent i-vector extractor is based on a set of monophone HMMs instead of the standard Gaussian Mixture Model (GMM). To compensate for the channel variability, we propose to precondition i-vectors using a regularized variant of within-class covariance normalization, which can be robustly estimated in a phrase-dependent fashion on the small datasets available for the text-dependent task. The verification scores are cosine similarities between the i-vectors normalized using phrase-dependent s-norm. The experimental results on RSR2015 and RedDots databases confirm the effectiveness of the proposed approach, especially in rejecting test utterances with a wrong phrase. A simpleMFCC based i-vector/HMM system performs competitively when compared to very computationally expensive DNN-based approaches or the conventional relevance MAP GMM-UBM, which does not allow for compact speaker representations. To our knowledge, this paper presents the best published results obtained with a single system on both RSR2015 and RedDots dataset.