Editor’s Note
On June 11, the results of the 2025 Shanghai Jiao Tong University Outstanding Bachelor’s Thesis Selection were announced! A total of 60 theses from 29 schools participated in the evaluation. After departmental recommendations and university-level defense reviews, 45 theses were ultimately honored. Among them, the graduation thesis 《Research on Robust Voice Wake-Up in Real Complex Scenarios》 by Wen Wen, a 2025 undergraduate graduate of SPEIT, was successfully selected.
Since the 2012 graduating class, the university has held the "Shanghai Jiao Tong University Outstanding Bachelor’s Thesis (TOP 1%)" selection. Wen Wen’s recognition fully demonstrates the exceptional research capabilities and potential of SPEIT students.
文雯与论文指导教师俞凯合影
Wen Wen
SPEIT Class of 2025 Undergraduate
High School: Laizhou No.1 Middle School, Shandong Province
Major: Information Engineering
Research Focus: Signal Processing, Voice Wake-Up and Recognition
Next Step: Direct Master’s admission at SPEIT; research at SJTU Intelligent Speech Technology Lab
Supervisor: Prof. Yu Kai, School of Computer Science, SJTU
✦01.Thesis Abstract
✦
✦
This research focuses on voice wake-up technology in complex environments. To address the accuracy degradation of voice assistants caused by noise, reverberation, and interference from other speakers in real-world scenarios, two methods leveraging environmental information are proposed:
"Enhance-then-Wake": Optimizes audio enhancement simulation using environmental information, with limited effectiveness.
"Adaptive Wake-Up": Directly integrates environmental information into model training, significantly improving wake-up rates.
Innovatively, this study introduces no-reference audio evaluation metrics and equal-frequency binning strategies to guide simulated audio closer to real distributions. Findings confirm that environmental information positively enhances voice wake-up performance, showing broad research prospects. Future work should refine no-reference metric systems, improve simulation guidance methods, and expand applications of environmental information in voice interaction.
✦ 02. Q&A Highlights
✦
1
How did you select your thesis topic? Did you encounter difficulties?
During a summer internship in signal processing, I observed that improved audio quality scores did not increase wake-up rates as expected. This sparked my curiosity to investigate audio quality metrics, leading me to explore simulation improvements and noise-resistant models.
Challenges included limited coding skills and unclear research direction. Fortunately, my lab provided tremendous support: my supervisor offered guidance despite his busy schedule, and senior lab members helped debug code and shared experiences. With their help, I overcame these obstacles. I hope to pay this kindness forward to my juniors.
I also thank SPEIT faculty for their guidance in preparing my thesis defense.
2
SPEIT emphasizes foundational math/physics in the first 2.5 years. When did you begin research training?
In my freshman year, I joined the 10th "Jun Zheng" Research Program, studying audio feature detection and recognition. Though my theoretical and practical skills were limited then, this experience ignited my interest in speech signal processing and laid the groundwork for my current focus.
I formally joined my current lab one year ago. Previous non-speech projects were not in vain—they helped me identify my strengths and interests through trial and reflection. Now, I am firmly committed to speech signal processing research.
3
Any advice for SPEIT juniors at graduation?
SPEIT offers great freedom. Its integrated bachelor-master program freed me from postgraduate admission pressures, allowing me to explore my true interests. Here, I could set aside GPA concerns and experiment widely—this environment granted me room for trial and error.
As a critical component of undergraduate education, the thesis/dissertation integrates teaching, research, and practice, significantly enhancing students’ comprehensive abilities. I hope SPEIT students embrace this freedom to explore boldly and strive proactively. Like Wen Wen, accumulate depth and shine brilliantly—fully demonstrating academic excellence and limitless potential.