Front-End Factor Analysis for Speaker Verification

Abstract

Abstract-This paper presents an extension of our previous work which proposes a new speaker representation for speaker verification. In this modeling, a new low-dimensional speaker-and channel-dependent space is defined using a simple factor analysis. This space is named the total variability space because it models both speaker and channel variabilities. Two speaker verification systems are proposed which use this new representation. The first system is a support vector machine-based system that uses the cosine kernel to estimate the similarity between the input data. The second system directly uses the cosine similarity as the final decision score. We tested three channel compensation techniques in the total variability space, which are within-class covariance normalization (WCCN), linear discriminate analysis (LDA), and nuisance attribute projection (NAP). We found that the best results are obtained when LDA is followed by WCCN. We achieved an equal error rate (EER) of 1.12% and MinDCF of 0.0094 using the cosine distance scoring on the male English trials of the core condition of the NIST 2008 Speaker Recognition Evaluation dataset. We also obtained 4% absolute EER improvement for both-gender trials on the 10 s-10 s condition compared to the classical joint factor analysis scoring.Index Terms-Cosine distance scoring, joint factor analysis (JFA), support vector machines (SVMs), total variability space.

Multitaper MFCC and PLP features for speaker verification using i-vectors

“…An i-vector extractor represents entire speech segments as low-dimensional feature vectors called i-vectors [4,5,14]. The i-vector extractors studied in [4,5,14] areaccording to long traditions in speaker verification research following NIST SRE evaluation protocol -gender-dependent and they are followed by gender-dependent generative modeling stages. In this paper, however, we use a gender-independent i-vector extractor, as shown in Fig.…”

“…Therefore, in this work, we provide detailed comparison between different taper weight selections in the popular Thomson multi-taper method. The recent i-vector model [4,5,6] includes elegant intersession variability compensation, with demonstrated significant improvements on the recent NIST speaker recognition evaluation corpora. Since i-vectors already do a good job in compensating for variabilities in the speaker model space, one may argue that improvements in the front-end may not translate to the full recognition system.…”

Investigation of Segmentation in i-Vector Based Speaker Diarization of Telephone Speech

“…Our goal is to determine whether the SCD approach offers any improvement under such conditions.For this purpose, we implement an i-vector based speaker diarization system. The use of i-vectors in speaker diarization has become increasingly popular in recent years [2,5], following their success in speaker recognition tasks [6,7]. This paper is organized as follows: The i-vector based speaker diarization system is described in Section 2.…”

Frame-by-frame language identification in short utterances using deep neural networks

“…Currently, most acoustic approaches to perform LID rely on i-vector technology [22]. All such approaches, while sharing i-vectors as a feature representation, differ in the type of classifier used to perform the final language identification [23].…”

“…From those features, we built a Universal Background Model of 1024 components. The Total Variability matrix was trained by using PCA and a posterior refinement of 10 EM iterations [22], keeping just the top 400 eigenvectors. We then derived the i-vectors using the standard methodology presented in Section 2.1.…”