Anti-Spoofing for Text-Independent Speaker Verification: An Initial Database, Comparison of Countermeasures, and Human Performance

Abstract

In this paper, we present a systematic study of the vulnerability of automatic speaker verification to a diverse range of spoofing attacks. We start with a thorough analysis of the spoofing effects of five speech synthesis and eight voice conversion systems, and the vulnerability of three speaker verification systems under those attacks. We then introduce a number of countermeasures to prevent spoofing attacks from both known and unknown attackers. Known attackers are spoofing systems whose output was used to train the countermeasures, whilst an unknown attacker is a spoofing system whose output was not available to the countermeasures during training. Finally, we benchmark automatic systems against human performance on both speaker verification and spoofing detection tasks.

The Voice Conversion Challenge 2016

“…The baseline system is based on the voice conversion toolkit within the open-source Festvox system 2 , as in our previous work [40], we found the toolkit can achieve similar performance to other state-of-the-art voice conversion or speech synthesis adaptation techniques. The toolkit is based on the joint density Gaussian mixture model with maximum likelihood parameter trajectory generation considering global variance as proposed in [27].…”

ASVspoof: The Automatic Speaker Verification Spoofing and Countermeasures Challenge

“…software available to the public, including fraudsters. More details including protocols used to generate spoofed speech can be found in [6], [7]. Figure 2 provides an intuitive visualisation of the acoustic similarities between each spoofing-attack algorithm and genuine speech.…”

“…The S1 and S2 spoofing-attack algorithms are two of the most easily implemented voiceconversion algorithms. The S3, S4, and S5 spoofing-attack algorithms were all implemented using either the HTS 6 or Festvox 7 , both of which are publicly available and open source. The remaining unknown attacks (S6-S10) are generally more sophisticated.…”

“…They include the series of LivDet evaluations for fingerprint recognition [8] and a similar initiative for face recognition [9]. As of 2014, [6], [7], the first ASVspoof challenge was co-organized in 2015. The initiative lead to the submission of 16 countermeasure systems, all benchmarked and ranked using a common database, protocol, and metric.…”

“…Stemming from the special session at INTERSPEECH, the creation of a large dataset of genuine and spoofed speech [6], [7] started later in 2013. Following refinements and improvements to the dataset and protocols, the first ASVspoof challenge [11] 1 was organised as a special session at IN-TERSPEECH 2015 held in Dresden, Germany.…”