Variable frame rate-based data augmentation to handle speaking-style variability for automatic speaker verification

Abstract

The effects of speaking-style variability on automatic speaker verification were investigated using the UCLA Speaker Variability database which comprises multiple speaking styles per speaker. An x-vector/PLDA (probabilistic linear discriminant analysis) system was trained with the SRE and Switchboard databases with standard augmentation techniques and evaluated with utterances from the UCLA database. The equal error rate (EER) was low when enrollment and test utterances were of the same style (e.g., 0.98% and 0.57% for read and conversational speech, respectively), but it increased substantially when styles were mismatched between enrollment and test utterances. For instance, when enrolled with conversation utterances, the EER increased to 3.03%, 2.96% and 22.12% when tested on read, narrative, and pet-directed speech, respectively. To reduce the effect of style mismatch, we propose an entropy-based variable frame rate technique to artificially generate style-normalized representations for PLDA adaptation. The proposed system significantly improved performance. In the aforementioned conditions, the EERs improved to 2.69% (conversation -read), 2.27% (conversation -narrative), and 18.75% (pet-directedread). Overall, the proposed technique performed comparably to multi-style PLDA adaptation without the need for training data in different speaking styles per speaker.

Effects of language mismatch in automatic forensic voice comparison using deep learning embeddings

“…Few studies have focused on whether these factors actually matter [34][35][36], although if they do, it may bias the evaluation of the evidence. In the present study, we use the three speech styles available in the dataset developed for forensic claims and compare the results depending on sample duration.…”

A Principle Solution for Enroll-Test Mismatch in Speaker Recognition

“…The multi-condition training (MCT) [42] can be regarded as a special normalization approach, belonging to the scoring theme. It pools the data from both enrollment and test conditions and trains a multi-conditional PLDA.…”

“…Our approach (SD/LT as the first and simplest case) also belongs to the scoring theme, but it is fundamentally different from the normalization methods that pursue a conditioninsensitive model as in IDVC [35] or MCT [42]. Instead, it admits the discrepancy between the enrollment and test conditions, and models the statistics of speaker vectors in the two conditions respectively.…”

“…Therefore, any solution that does not address this incoherence is theoretically not optimal. For example, the commonly adopted multi-condition training (MCT) approach that pools data from both enrollment and test condition for PLDA training [42]. By this approach, the resultant PLDA is just an empirical compromise between the enrollment condition and the test condition, but is not optimal for either of them.…”

“…It is often significantly different from the enrollment condition, and varies from one test to another. Significant performance reduction is often observed with this mismatch [39], [40], [41], [42]. Some typical scenarios that involve serious enrollmenttest mismatch are:…”