Pavel Korshunov

TL;DR: A publicly available UHD video dataset designed for evaluation of privacy issues is proposed, which demonstrates that UHD is a significantly more privacy intrusive technology when compared to HD and SD used today, and quantifies the impact of intrusivness.

TL;DR: This paper focuses on the specific task of gender recognition in images when they have been processed by privacy protection filters applied at different strengths, and compares the performance of state of the art deep learning algorithms with a subjective evaluation obtained via crowdsourcing.

TL;DR: This paper describes the speaker diarization systems developed for the Second DIHARD Speech Diarization Challenge (DIHARD II) by the Speed team, and presents several components of the system, including categorization of domains, speech enhancement, speech activity detection, and speaker embeddings.

TL;DR: A long-term study on image aesthetic appeal challenged the crowdsourced assessments with typical lab methodologies in order to identify and analyze the impact of crowdsourcing environment on the reliability of subjective data.

TL;DR: This paper demonstrates that by replacing standard MFCC features with embeddings from a DNN trained for automatic speech recognition, combined with mouth landmarks (visual features), this model can achieve a significant performance improvement on several challenging publicly available databases of speakers, for which it generated sets of tampered data.