Home

Introduction

I recently finished my PhD at the Centre for Speech Technology Research (CSTR), School of Informatics, the University of Edinburgh, supervised by Prof Peter Bell. My research interest lies in Speech Technology, including Automatic Speech Recognition (ASR), especially End-to-End ASR, and Text-to-Speech (TTS). For more information, please refer to my CV (updated 2 August 2025). I am also a contributor of SpeechBrain. Currently, I am a Research Officer, a postdoc, at Swansea University working on Generative and Interactive AI.

• Github
• Google Scholar
• LinkedIn

Recent Work and News

• (December 2024) One paper accepted in IEEE ICASSP 2025. See you in Hyderabad, India!
• (August 2024) One paper accepted in IEEE SLT Workshop 2024. See you in Macau!
• (August 2024) I submitted my PhD thesis in August and will start to work at Swansea University as a Research Officer in September.
• (March 2024) I am now working on a new differentiable WFST-based ASR toolkit, named BenNevis, where the topology can be freely defined.

Education

• PhD
  • September 2020 - July 2025
  • Supervised by Prof Peter Bell
  • Centre for Speech Technology Research (CSTR)
  • University of Edinburgh
• Master
  • September 2017 - August 2020
  • Supervised by Dr Wei-Qiang Zhang
  • Speech and Audio Technology Lab
  • Tsinghua Univeristy
• Bachelor
  • September 2013 - June 2017
  • School of Information and Electronics
  • Beijing Institute of Technology

Selected Publications

Here are some publications of mine. For the whole list, please refer to publications and my Google Scholar

• Zeyu Zhao and Peter Bell, Regarding the Existence of the Internal Language Model in CTC-Based E2E ASR, ICASSP 2025. link pdf code

• Zeyu Zhao and Peter Bell, Advancing CTC Models For Better Speech Alignment: A Topological Approach, SLT 2024. pdf link code

• Zeyu Zhao, Pinzhen Chen and Peter Bell, Regarding Topology and Adaptability in Differentiable WFST-Based E2E ASR, ICASSP 2024 XAI-SA Workshop. pdf link code video

• Zeyu Zhao, Peter Bell and Ondrej Klejch, Exploring Dominant Paths in CTC-Like ASR Models: Unraveling the Effectiveness of Viterbi Decoding, ICASSP 2024 XAI-SA Workshop. pdf link code video

• Zeyu Zhao and Peter Bell, Regarding Topology and Variant Frame Rates for Differentiable WFST-based End-to-End ASR, Interspeech 2023. pdf link code

• Zeyu Zhao and Peter Bell, Investigating Sequence-Level Normalisation For CTC-Like End-to-End ASR, ICASSP 2022. pdf link

Personal Hobbies

• Mechanical keyboard
• Guitar
• Video game

About my English name

Jarvis has a similar pronunciation to my Chinese name (Zeyu) and is short for Just A Rather Very Intelligent System, which is the speech assistant of Iron Man suit. It is a coincidence that I also major in Speech Recognition.