Denis Bibik
About Denis Bibik
Denis Bibik is a Middle Machine Learning Engineer who teaches a course on speech synthesis at Moscow State University. He has a Bachelor's degree, a Master's degree, and is currently a Ph.D. candidate in Computer Science, with significant experience in improving speech recognition and text-to-speech technologies.
Work at Mail.ru Group
Denis Bibik has been employed at Mail.ru Group since 2019, currently holding the position of Middle Machine Learning Engineer. His work focuses on enhancing speech recognition technologies. He previously served as a Junior Machine Learning Engineer at the same company from 2018 to 2019. During his tenure, he made significant contributions, including improving offline speech recognition through the implementation of MWER loss and domain adaptation, which led to a fourfold acceleration of the speech recognition pipeline and a reduction in word error rate from 20% to 7%. Additionally, he reduced the latency of online speech recognition responses by 35%.
Education and Expertise
Denis Bibik has a strong educational background in Computer Science from Moscow State University. He earned his Bachelor's degree from 2014 to 2018, followed by a Master's degree from 2020 to 2021. Currently, he is a Ph.D. candidate at the same institution, pursuing his studies from 2021 to 2025. In addition to his academic pursuits, he teaches a course on speech synthesis for graduate students at Moscow State University, sharing his expertise in this specialized field.
Previous Experience at PicsArt Inc.
Denis Bibik worked at PicsArt Inc. as a Machine Learning Engineer for a duration of four months in 2019. During this time, he developed a model for text normalization that enabled the implementation of a text-to-speech module capable of processing free formatted text. This development was crucial for applications involving facts and calendar functionalities.
Achievements in Speech Recognition Technology
Throughout his career, Denis Bibik has achieved notable advancements in speech recognition technology. He enhanced a text-to-speech model by incorporating prosody control, pauses control, and intonation cloning. This improvement resulted in a 7% increase in performance in side-by-side comparisons with the previous production model. His work has significantly contributed to the field of machine learning, particularly in the areas of speech synthesis and recognition.