- Visiting period:
December 2019 to March 2020 - Visiting university:
Johns Hopkins University - Visiting country and region:
Maryland, U.S.A
Content of the study program
I worked on speech recognition research under Dr. Shinji Watanabe at the Center for Language and Speech Processing (CLSP), Johns Hopkins University (JHU). Speech recognition realizes speech-to-text conversion by a single deep neural network (DNN) model. Although recent research has drastically improved the precision of speech recognition, challenges related to extremely slow inference-speed due to the network characteristics remain. This program aims to build an end-to-end speech recognition model with a high inference-speed while maintaining recognition precision.
Study results
I worked on research for high inference-speed end-to-end speech recognition. I read many papers, including ones outside the field of speech. I spent time considering different ideas through discussions with professors and senior students. Most ideas did not produce the expected outcomes, and I was frustrated with the time limitation. In the end, I was able to set the direction for my research. I also learned the importance of perseverance and continuing to contemplate ideas through the knowledge gained from failures. Aside from the end-to-end speech recognition research, I incorporated my previous research. I deeply understood how new research could result from a gathering of students with similar research themes to share the knowledge and tools with one another. Such activities allow researchers to utilize diverse experiences as tools to enhance the enjoyment of research.
My experience abroad
Baltimore, Maryland is notorious for its rough neighborhoods in the U.S. Since this was my destination, I was filled with anxiety before leaving Japan. However, the area surrounding the university is relatively safe, and I did not encounter trouble. I enjoyed typical American events such as attending a Christmas party at the professor’s home and watching American football games. At the university, I participated in the research-paper read-aloud and progress meetings to present and to discuss in English. This was the first time that I used English for work, and it was great practice to explain the research content and answer/ask questions. I attended lectures on the topic of speech recognition, which are not offered in Waseda University, and participated in seminars that invited outside researchers. These experiences deepened my understanding of the state-of-the-art research content and my foundational knowledge of specialization.
