With the increasing popularity of automatic speaker verification (ASV), the reliability of ASV systems has also gained importance. ASV is vulnerable to various spoofing attacks, especially replay attacks. Thus, recent public competitions and studies based on spoofing attack detection for ASV have mainly focused on the detection of replay attacks. Generally, replayed speech includes the attributes of one playback and two recording devices: the playback device, the recording device used by the attacker, and the recording device embedded in any system to verify input utterances. Therefore, the main attributes differentiating a replayed speech from the genuine speech are the attributes of the playback and the recording devices used by the attacker. In this paper, we propose a novel replay attack and its defense through observation of the general speech-spoofing process. The proposed attack includes only the attribute of one recording device embedded in an ASV system; genuine speech passes through the recording device only once, and the replayed speech produced for the proposed attack passes through the same recording device twice. Because the proposed attack is feasible, it can be considered a new task for replay countermeasures in the training process in order to develop a robust ASV protection system. The experimental results show that this novel replay attack cannot be detected by several of the existing state-of-the-art replay attack detection systems. Furthermore, the new attack can be detected by the same systems successfully if they are retrained with an appropriate dataset designed for the new task.

Recurrent neural networks (RNNs) can model the time-dependency of time-series data. It has also been widely used in text-dependent speaker verification to extract speaker-and-phrase-discriminant embeddings. As with other neural networks, RNNs are trained in mini-batch units. In order to feed input sequences into an RNN in mini-batch units, all the sequences in each mini-batch must have the same length. However, the sequences have variable lengths and we have no knowledge of these lengths in advance. Truncation/padding are most commonly used to make all sequences the same length. However, the truncation/padding causes information distortion because some information is lost and/or unnecessary information is added, which can degrade the performance of text-dependent speaker verification. In this paper, we propose a method to handle variable length sequences for RNNs without adding information distortion by truncating the output sequence so that it has the same length as corresponding original input sequence. The experimental results for the text-dependent speaker verification task in part 2 of RSR 2015 show that our method reduces the relative equal error rate by approximately 1.3% to 27.1%, depending on the task, compared to the baselines but with an associated, small overhead in execution time.

We propose a method, called bi-point input, for convolutional neural networks (CNNs) that handle variable-length input features (e.g., speech utterances). Feeding input features into a CNN in a mini-batch unit requires that all features in each mini-batch have the same shape. A set of variable-length features cannot be directly fed into a CNN because they commonly have different lengths. Feature segmentation is a dominant method for CNNs to handle variable-length features, where each feature is decomposed into fixed-length segments. A CNN receives one segment as an input at one time. However, a CNN can consider only the information of one segment at one time, not the entire feature. This drawback limits the amount of information available at one time and consequently results in suboptimal solutions. Our proposed method alleviates this problem by increasing the amount of information available at one time. With the proposed method, a CNN receives a pair of two segments obtained from a feature as an input at one time. Each of the two segments generally covers different time ranges and therefore has different information. We also propose various combination methods and provide a rough guidance to set a proper segment length without evaluation. We evaluate the proposed method on the spoofing detection tasks using the ASVspoof 2019 database under various conditions. The experimental results reveal that the proposed method reduces the relative equal error rate (EER) by approximately 17.2% and 43.8% on average for the logical access (LA) and physical access (PA) tasks, respectively.

In the field of speaker verification, probabilistic linear discriminant analysis (PLDA) is the dominant method for back-end scoring. To estimate the PLDA model, the between-class covariance and within-class precision matrices must be estimated from samples. However, the empirical covariance/precision estimated from samples has estimation errors due to the limited number of samples available. In this paper, we propose a method to improve the conventional PLDA by estimating the PLDA model using the regularized within-class precision matrix. We use graphical least absolute shrinking and selection operator (GLASSO) for the regularization. The GLASSO regularization decreases the estimation errors in the empirical precision matrix by making the precision matrix sparse, which corresponds to the reflection of the conditional independence structure. The experimental results on text-dependent speaker verification reveal that the proposed method reduce the relative equal error rate by up to 23% compared with the conventional PLDA.