The research by Le Duy Khanh, "Gen Z" engineer from Zalo AI, which improves the accuracy of real-time speech recognition models (Streaming Automatic Speech Recognition), will be presented for the first time at an international scientific conference in Greece in September 2024.



With the topic "Improving Streaming Speech Recognition With Time-Shifted Contextual Attention And Dynamic Right Context Masking," Khanh's research, focused on enhancing real-time speech recognition using time-shifted attention mechanisms and dynamic future context masking, scored nearly perfect—11 out of 12 points. It passed the rigorous selection process from over 2,000 submissions and will be presented at the Interspeech Conference in an oral session.

“I’m very proud that my first scientific paper has been recognized by such a prestigious conference, and that I have the opportunity to showcase the research achievements of Vietnam to big-tech, experts, and the international community,” Le Duy Khanh shared.

Under the guidance of Dr. Chau Thanh Duc, Head of Research and Development at Zalo AI and a lecturer at the University of Science (Vietnam National University, Ho Chi Minh City), this research is expected to make a significant contribution to improving speech recognition models. It aims to enhance the accuracy of features like voice dictation and voice-to-text conversion on the Zalo app.

"The process of compiling Zalo AI’s practical research into scientific papers and presenting them at international conferences is highly valuable. It not only demonstrates the capability of Vietnamese engineers but also reflects our desire to share knowledge and contribute to the development of the global AI community,” said Dr. Chau Thanh Duc

Earlier, Zalo integrated this research into its messaging app in late 2023, significantly improving the accuracy of the "voice dictation" feature. This feature allows users to dictate messages by voice instead of typing, saving time and offering greater convenience in various use cases. The accuracy of this feature has reached 95% in real-world use, and the need to correct text after voice dictation has decreased from 6.4% to just 4.8%.



According to Zalo's statistics, although the feature is still in the testing phase, it generates nearly 4.5 million messages daily and attracts about 3.2 million monthly users (as of June 2024).

Since embarking on its pioneering journey in AI research in 2017, Zalo has consistently believed in "empowering" the younger generation. Currently, 31% of Zalo’s workforce belongs to Gen Z. In 2021, two other research projects by Zalo AI engineers on speech processing technology were recognized at the Asia-Pacific International Conference on Artificial Intelligence (PRICAI 2021). Notably, the authors of both projects were young researchers under the age of 30.

Interspeech is one of the oldest, most comprehensive, and prestigious international conferences on speech processing, organized by the International Speech Communication Association. This year’s conference, themed "Speech and Beyond," will take place from September 1-5, 2024, on the island of Kos, Greece.