About Me

Hi, this is Linkai Peng. I am a first-year Ph.D. student in UConn’s Psychology program, and I am fortunate to be advised by Dr. Jim Magnuson. I am curious about the process of human/machine speech perception. My research interests intersect in the areas of speech processing, phonetics, second language acquisition, neuroscience, and cognitive science.

📌 Research Themes Overview

Image from: Li, Y., Anumanchipalli, G., Mohamed, A., Chen, P., Carney, L. H., Lu, J., Wu, J., Chang, E.F. (2023) Dissecting neural computations of the human auditory pathway using deep neural networks for speech. Nature Neuroscience, 26, 1-30.

How do humans perceive and process speech?
We investigated the contribution of several acoustic parameters (phonological and fluency (Rian et al.'2022), lexical tone (Rian et al.'2022), and contextual information (Rian et al.'2022)) to the comprehensibility of second language speech through behavioral experiments and Functional Load Principle.
Do deep learning to have similar speech processing mechanisms as the human brain? (or is it necessary)
Advanced deep learning models have brought new research paradigms to the exploration of how the human brain processes speech. Many studies have found a strong alignment between the hierarchical representations of self-supervised speech models and the neural activity in human auditory pathway (left picture). (Project #1'2023 / 中文 )
How to recover speech/semantic information from brain recordings?
Intracranial recordings have made significant advancements in this area while the challenge of decoding speech from non-invasive brain recordings remains. (Project #2'2023).

✨ Past Research Themes

Non-native Speech Processing: Mispronunciation detection and diagnosis with self supervised model. (Linkai et al'2021;2022)
Mandarin Tone Recognition: modeling tone with multi-scale temporal-frequency features. (Linkai et al'2020;)

Updates

[06/2024] A new work with Zuoyebang AI Research: https://arxiv.org/pdf/2407.09209
[11/2023] The setup of two research project pages has been completed [Project #1'2023 / 中文 | Project #2'2023]. If you have similar research interests and relevant background (computer science, cognitive science, and linguistics), please feel free to take a look at them and any suggestions are greatly appreciated.
[08/2023] New preprint [ArXiv | Website], Spoken Language Intelligence of Large Language Models for Language Learning, We conducted preliminary explorations on large language models and confirmed through various prompts that text-based large language models (LLM, e.g. ChatGPT, GPT4, LLaMa2) have a good understanding of concepts in phonetics, phonology, and second language acquisition.

Research Experience

University of Connecticut, CT, US, August. 2024 - Now
Computational Cognitive Neuroscience of Language Lab
- Research Assistant
NetEase, Beijing, China, July 2022 - May 2024
- Machine Learning Engineer
ByteDance AI-Lab, Beijing, China, November. 2020 - February. 2021
- Research Intern
Beijing Language and Culture University, Beijing, China, September. 2019 - June. 2022
Speech Acquisition and Intelligent Technology Lab
- Research Assistant

Last update: 08/2024. Modified from Website of Prof. Hanjie Chen