Voice conversion versus speaker verification: an overview

Abstract

A speaker verification system automatically accepts or rejects a claimed identity of a speaker based on a speech sample. Recently, a major progress was made in speaker verification which leads to mass market adoption, such as in smartphone and in online commerce for user authentication. A major concern when deploying speaker verification technology is whether a system is robust against spoofing attacks. Speaker verification studies provided us a good insight into speaker characterization, which has contributed to the progress of voice conversion technology. Unfortunately, voice conversion has become one of the most easily accessible techniques to carry out spoofing attacks; therefore, presents a threat to speaker verification systems. In this paper, we will briefly introduce the fundamentals of voice conversion and speaker verification technologies. We then give an overview of recent spoofing attack studies under different conditions with a focus on voice conversion spoofing attack. We will also discuss anti-spoofing attack measures for speaker verification.

Development of a computationally efficient voice conversion system on mobile phones

“…(a) Split a sentence for parallel conversion (b) Logistic function for segment merging. subsequent steps such as the estimation of GMM parameters in equation (8). For example, in the work of Wojciech Kwedlo [47], a shared memory parallelization of the standard EM algorithm based on data decomposition is proposed to learn the GMM parameters on a multi-core system for higher performance.…”

SAS: A speaker verification spoofing database containing diverse attacks

“…The eigenvoice GMM (EV-GMM) was constructed from the training data from one pivot speaker in the ATR Japanese speech database [35], and 273 speakers (137 male, 136 female) from the JNAS database. 4 Settings were the same as in [36]. The 272-dimensional weight vectors were estimated by using 4 http://www.milab.is.tsukuba.ac.jp/jnas/instruct.html the Part-A (small) training data.…”

“…4 Settings were the same as in [36]. The 272-dimensional weight vectors were estimated by using 4 http://www.milab.is.tsukuba.ac.jp/jnas/instruct.html the Part-A (small) training data. Covariance matrices in EV-GMM were not updated, i.e.…”

“…A major concern when deploying an ASV system is whether the system is secure against spoofing attacks. Recently, an increasing number of studies have assessed the vulnerability of ASV systems to various forms of spoofing attacks [2,3,4], including impersonation [5,6], replay [7], speech synthesis [8,9] and voice conversion [10,11,12,13,14]. Efforts have also been made to develop individual countermeasures to protect ASV systems against specific spoofing attacks.…”

On the study of replay and voice conversion attacks to text-dependent speaker verification

“…As discussed in [50], feature extraction in speaker verification is one of the weak links. In a replay attack, an attacker plays a pre-recorded speech from the exact target speaker to spoof a text-dependent speaker verification system.…”