Mikhail Kudinov

Vadim Popov, Ivan Vovk, Vladimir Gogoryan et al.

Voice conversion is a common speech synthesis task which can be solved in different ways depending on a particular real-world scenario. The most challenging one often referred to as one-shot many-to-many voice conversion consists in copying the target voice from only one reference utterance in the most general case when both source and target speakers do not belong to the training dataset. We present a scalable high-quality solution based on diffusion probabilistic modeling and demonstrate its superior quality compared to state-of-the-art one-shot voice conversion approaches. Moreover, focusing on real-time applications, we investigate general principles which can make diffusion models faster while keeping synthesis quality at a high level. As a result, we develop a novel Stochastic Differential Equations solver suitable for various diffusion model types and generative tasks as shown through empirical studies and justify it by theoretical analysis.

Vadim Popov, Ivan Vovk, Vladimir Gogoryan et al.

Recently, denoising diffusion probabilistic models and generative score matching have shown high potential in modelling complex data distributions while stochastic calculus has provided a unified point of view on these techniques allowing for flexible inference schemes. In this paper we introduce Grad-TTS, a novel text-to-speech model with score-based decoder producing mel-spectrograms by gradually transforming noise predicted by encoder and aligned with text input by means of Monotonic Alignment Search. The framework of stochastic differential equations helps us to generalize conventional diffusion probabilistic models to the case of reconstructing data from noise with different parameters and allows to make this reconstruction flexible by explicitly controlling trade-off between sound quality and inference speed. Subjective human evaluation shows that Grad-TTS is competitive with state-of-the-art text-to-speech approaches in terms of Mean Opinion Score. We will make the code publicly available shortly.

S. E. Yunakovsky, M. Kot, N. O. Pozhar et al.

Quantum computing technologies pose a significant threat to the currently employed public-key cryptography protocols. In this paper, we discuss the impact of the quantum threat on public key infrastructures (PKIs), which are used as a part of security systems for protecting production environments. We analyze security issues of existing models with a focus on requirements for a fast transition to post-quantum solutions. Although our primary focus is on the attacks with quantum computing, we also discuss some security issues that are not directly related to the used cryptographic algorithms but are essential for the overall security of the PKI. We attempt to provide a set of security recommendations regarding the PKI from the viewpoints of attacks with quantum computers.

M. A. Kudinov, E. O. Kiktenko, A. K. Fedorov

In this work, we discuss in detail a flaw in the original security proof of the W-OTS${^+}$ variant of the Winternitz one-time signature scheme, which is an important component for various stateless and stateful many-time hash-based digital signature schemes. We update the security proof for the W-OTS${^+}$ scheme and derive the corresponding security level. Our result is of importance for the security analysis of hash-based digital signature schemes.

E. O. Kiktenko, M. A. Kudinov, A. A. Bulychev et al.

In the present work, a peculiar property of hash-based signatures allowing detection of their forgery event is explored. This property relies on the fact that a successful forgery of a hash-based signature most likely results in a collision with respect to the employed hash function, while the demonstration of this collision could serve as convincing evidence of the forgery. Here we prove that with properly adjusted parameters Lamport and Winternitz one-time signatures schemes could exhibit a forgery detection availability property. This property is of significant importance in the framework of crypto-agility paradigm since the considered forgery detection serves as an alarm that the employed cryptographic hash function becomes insecure to use and the corresponding scheme has to be replaced.

Vadim Popov, Mikhail Kudinov

Cross-entropy loss is a common choice when it comes to multiclass classification tasks and language modeling in particular. Minimizing this loss results in language models of very good quality. We show that it is possible to fine-tune these models and make them perform even better if they are fine-tuned with sum of cross-entropy loss and reverse Kullback-Leibler divergence. The latter is estimated using discriminator network that we train in advance. During fine-tuning probabilities of rare words that are usually underestimated by language models become bigger. The novel approach that we propose allows us to reach state-of-the-art quality on Penn Treebank: perplexity decreases from 52.4 to 52.1. Our fine-tuning algorithm is rather fast, scales well to different architectures and datasets and requires almost no hyperparameter tuning: the only hyperparameter that needs to be tuned is learning rate.

Vadim Popov, Mikhail Kudinov, Irina Piontkovskaya et al.

One of the big challenges in machine learning applications is that training data can be different from the real-world data faced by the algorithm. In language modeling, users' language (e.g. in private messaging) could change in a year and be completely different from what we observe in publicly available data. At the same time, public data can be used for obtaining general knowledge (i.e. general model of English). We study approaches to distributed fine-tuning of a general model on user private data with the additional requirements of maintaining the quality on the general data and minimization of communication costs. We propose a novel technique that significantly improves prediction quality on users' language compared to a general model and outperforms gradient compression methods in terms of communication efficiency. The proposed procedure is fast and leads to an almost 70% perplexity reduction and 8.7 percentage point improvement in keystroke saving rate on informal English texts. We also show that the range of tasks our approach is applicable to is not limited by language modeling only. Finally, we propose an experimental framework for evaluating differential privacy of distributed training of language models and show that our approach has good privacy guarantees.